Enterprise Service Oriented Architectures

ENTERPRISE SERVICE ORIENTED ARCHITECTURES
Enterprise Service Oriented

Architectures
Concepts, Challenges, Recommendations

by
JAMES MCGOVERN
OLIVER SIMS
ASHISH JAIN
MARK LITTLE
A C.I.P. Catalogue record for this book is available from the Library of Congress.
ISBN-10
ISBN-13
ISBN-10
ISBN-13
1-4020-3704-X (HB)
978-1-4020-3704-7 (HB)
1-4020-3705-8 (e-book)
978-1-4020-3705-4 (e-book)
Published by Springer,
P.O. Box 17, 3300 AA Dordrecht, The Netherlands.
www.springer.com
Printed on acid-free paper
All Rights Reserved

2006 Springer
No part of this work may be reproduced, stored in a retrieval system, or transmitted
in any form or by any means, electronic, mechanical, photocopying, microfilming, recording
or otherwise, without written permission from the Publisher, with the exception
of any material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work.
Printed in the Netherlands.
The condition of an enlightened mind is a surrendered heart.

Alan Redpath
Author Team
To those who are savage in the pursuit of excellence
James
To my wife Sherry and sons James and Sylvester who provide and replenish the
energy necessary for me to complete the exciting work as well as the mundane. To
Mom and Dad, thanks for the encouragement and persistence.
Oliver
To my wife, Heather, for just about everything, and to my children Christopher,
Richard, and David, of whom I am inordinately proud, and who have kept me
rmly rooted in reality.
Ashish
To my wife Nishma and children Eshan and Ronit for their love, patience and
support. To my parents, for their encouragement throughout the years.
Mark
Id like to send my love to all my family, particularly my wife Paula and our children
Daniel and Adam, who have acted as an anchor for me through the years, keeping
me sane throughout the storms.
TABLE OF CONTENTS
ENDORSEMENTS
xi
ABOUT THE SERIES
xiii
FOREWORD
xvii
PREFACE
xxi
ABOUT THIS BOOK
xxv
ACKNOWLEDGEMENTS
xxxi
ABOUT THE REVIEWERS
xxxiii
CHAPTER 1: UNDERSTANDING SERVICE-ORIENTED

ARCHITECTURE
Introduction
1
3
1. Introducing Service-Oriented Architectures
2. Service-Based Collaboration through Federation
19
3. The Platform
32
4. Transitioning to ESOA
45
5. Summary
48
CHAPTER 2: COMPONENT-BASED SERVICES
49
1. Component-Based Software Engineering (CBSE)
51
2. A Component Denition
55
3. Component Granularity
64
vii
viii
Table of Contents
4. From Requirements to Design
81
5. Summary
94
CHAPTER 3: ORCHESTRATION
1. Workow and Business Process Management
95
97
2. The Business Process Execution Language (BPEL)
101
3. A Worked Example of Web Services Orchestration
128
4. Design-Time Demonstration
129
5. Run-Time Demonstration
145
6. Summary
148
CHAPTER 4: WORKING WITH REGISTRY AND UDDI
151
1. Introducing the Registry
152
2. Universal Description, Discovery and Integration (UDDI)
154
3. Programming UDDI
169
4. Internationalization
179
5. Summary
187
CHAPTER 5: UNDERSTANDING ENTERPRISE SECURITY
189
1. Need for a Message Level Security Solution
191
2. Security Concepts
193
3. Security Technologies
196
4. Web Services Security (WSS)
225
5. WS-Policy
230
6. WS-Trust
231
7. WS-Privacy
232
8. WS-SecureConversation
232
9. WS-Federation
233
Table of Contents
ix
10. WS-Authorization
233
11. Summary
233
CHAPTER 6: SOA MANAGEMENT
235
1. Problem Space
236
2. Systems Management
244
3. Alerting
250
4. Provisioning
252
5. Leasing
253
6. Billing
254
7. Pricing/Chargeback Models
255
8. Lifecycle Management
257
9. Management Architecture
271
10. Policy Architecture
277
11. Framework Vendors
279
12. Summary
280
CHAPTER 7: TRANSACTIONS
281
1. What Are ACID Transactions?
281
2. Why ACID Is Too Strong for Web Services
288
3. A Brief History of Web Services Transactions
290
4. The Coordination Frameworks
291
5. Web Services Transactions
299
6. Security Implications
312
7. Interoperability Considerations
314
8. Summary
315
Table of Contents
CHAPTER 8: EVENT-DRIVEN ARCHITECTURE
317
1. Overview
319
2. Events
320
3. Agents
323
4. Threads
329
5. Alternative Pattern-Based Approaches
332
6. Language Specic Constructs
338
7. Finite State Machines
344
8. Event Notication
347
9. Practical Considerations
352
10. Summary
355
OUTTRO
357
APPENDIX A: UNDERSTANDING DISTRIBUTED COMPUTING
359
1. Distributed Computing
360
385
3. Summary
385
APPENDIX B: QUALITY ATTRIBUTES
387
1. System Qualities
387
2. Design vs Run-Time
391
APPENDIX C: REFERENCES
395
APPENDIX D: ADDITIONAL READING
403
APPENDIX E: UPCOMING BOOKS
405
ENDORSEMENTS
You cant live a perfect day without doing something
for someone who will never be able to repay you.
John Wooden
Enterprise SOA is well written and insightful. This book covers enormous ground and
readers will nd it the best, single source on SOA. Highly recommended!
Ron Widitz
Enterprise Architect
Discover Financial
This book was truly a guide for all levels of individuals developing business applications
in todays global, open market. It clearly summarizes key concepts for executive
management, provides the framework of architectural guidelines and standards, as well
as provided detailed coding examples for entry level developers. This book should be
a must read for all interested in leading their organizations business model into the future.
Damon Rothstein
Enterprise Network Architect
Piper Jaffray and Companies
Concise, readable and useful as a tool in the running of a business. You truly pull
concepts together with real world examples.
W.M. Douglas Crawford
VP NASDAQ Technology & Operations
Advest
Enterprise SOA provides architects and developers with an excellent source of much
needed information on understanding how to utilize enterprise technologies such as
xi
xii
Endorsements
SOA, orchestration, components, and registries to solve problems that currently face the
enterprise. Understanding these technologies will help architects develop systems that can
solve current problems as well as lay down an architecture that will adapt to on going
changes in the business environment.
Suneet Shah
CTO and Chief Architect
Diamelle
Enterprise Service-Oriented Architectures provides a unique and worthwhile lifecycleperspective to realizing a SOA. A number of concepts such as components, registries,
web-service security, management, business processes, etc. are addressed in the context of
different stages during the realization of a SOA, including : translating SOA requirements
to design, design to implementation, and implementation to deployment.
Sekhar Sarukkai
Technical Evangelist
Oblix
This book is an outstanding and insightful work on the perspectives and potential of
service-oriented architecture. A must read for every Enterprise Architect who needs to
know how to succeed in the face of architectural challenges presented as part of his/her
daily chores.
Nitin Narayan
CEO
Mavenz, India
This book is the product of some of the leading thinkers in Information Technology
today. The concepts included in this book are being debated and analyzed by most of the
Information Ofcers in the world right now. This book provides a history of how we got
to SOAs, what they mean today, and where they will lead tomorrow. The implications of
SOAs, Web Services, Federation, BPEL, and Grid computing, will revolutionize the IT
industry. We are living in truly interesting times. Those of us in the IT community have
our work cut out for us to lead our companies and customers into the next generation of
computing. Thank you for this great book to help spearhead the charge!
Joe Gibson
Senior Practice Director, Integration
East Area Technology Services
Oracle
ABOUT THE SERIES

Before you can inspire with emotion, you must be swamped with it yourself. Before you
can move their tears, your own must ow. To convince them, you must yourself believe.
Winston Churchill
The new enterprise requires a new type of leadership based on the logical continuation
of historical effort, while not doing what has been done just because it has been
done that way in the past. Agility and leadership when combined is a cohesive
presentation of common sense mined from the few truly successful projects as
opposed to the aggregation of every (predominately failed) effort declared complete
not only successful projects but projects that really add value to a business imperative.
We are living in a new era where one seeks uncommon approaches while maintaining
common virtues.
The art of leadership is about knowing and inuencing people so that they can come
to share common values resulting in more efciency in achieving the strategic vision
for the enterprise. Leadership must also embrace diversity which will lead to a much
more lively dialectic.
The Enterprise Series has earned its place as a valuable resource to those who want
to ramp up quickly and stay ahead of the curve. The authors of books within this
series are not writers hired to cover the hot topic of the minute. Instead they are
thought leaders and expert practitioners who bring insight to the community at large
based on real-world experience. More importantly, they are not theorists but actually
practice what they preach.
This series is founded on the conviction that enterprises should differentiate
themselves, their architecture and their people based on the way they think as
much as on the products or services they sell. Thought leadership is all about
time-honored expertise and comprehensive capabilities. Its inection point however
is new thinking and original perspectives.
xiii
xiv
About the Series
We hope you nd this series and this book to be a practical guide and trusted advisor
leading you successfully on your journey.
James McGovern
Rajanish Dass
Anthony Finkelstein
John Gtze
Series Editors
James McGovern
James is an industry thought leader and the co-author of several recent books on
service-oriented architectures, enterprise architectures and technology leadership. He
is employed as an Enterprise Architect for The Hartford Financial Services Group,
Inc. He holds industry certications from Microsoft, Cisco and Sun. James is a
popular speaker at elite technology conferences around the globe. He is member of
the Java Community Process and the Worldwide Institute of Software Architects.
Rajanish Dass
Rajanish is a Fellow at IIM Calcutta and his primary research and teaching interests
are in the area of Information Systems Management and its applications to various
arenas of doing business. Particular interests lie in the area of real-time Data
Mining and Business Intelligence techniques, developing algorithms and heuristics
for performance enhancement, applications of Business Intelligence techniques for
various areas of business like supply chain management, network intrusion detection,
privacy and security enhancement, anti-spamming techniques, retailing, nance and
business policy making, Competitive Intelligence, etc.
A rank holder throughout his career; he has experiences in working for high-ended
projects at the IBM Research Laboratory India and at the Tata Consultancy Services.
He has published a number of research papers in forums of international repute and
of national signicance. He is currently doing research in collaboration with world
renowned research labs of MIT, Oracle Centre of Excellence, Messagelabs, etc.
Anthony Finkelstein
Anthony is Professor of Software Systems Engineering at University College
London. He is a founder of London Software Systems a new joint research
institute established jointly by UCL and Imperial College. He has contributed to
About the Series
xv
research on software specication methods, software development processes, tool

and environment support for software development. Recent work has included
signicant contributions to work on tools for managing model integrity in software
development, specication from multiple viewpoints and requirements engineering.
His current interests area is the managing of distributed information in software
development. He is the Head of the Department of Computer Science and founder
of Systemwire a technology spinout company.
John Gtze
John is chief consultant and enterprise architect in the Danish Ministry of Science,
Technology and Innovation, where he since 2001 has been heavily involved in establishing and managing the Danish governments enterprise architecture program, as an
important part of the national e-government policy. Recent work has included policy
documents as well as academic publications on cross-governmental interoperability,
architecture frameworks, service-oriented architecture and open standards. He is
also a non-tenured associate professor at Copenhagen Business School and the IT
University of Copenhagen, where he teaches masters classes and supervises theses. He
holds a PhD and MSc from the Technical University of Denmark.
FOREWORD
We live in an interesting (in the sense of the May your life be interesting Chinese
proverb) moment of the IT industry. Finally, many of the disciplines required
to manage IT in a structured and agile way are converging together. We now
have the theoretical concepts and practical experience to align IT to the business,
manage IT as a business, dene enterprise architectures, and align IT initiatives and
individual projects (not only as far as development is concerned, but also with regards
to outsourcing, deployment, integration, governance, and the other IT concerns).
Nearly any imaginable technology IT challenges can currently be solved (or has been
solved) by some company in the industry. Companies willing to invest the time and
money and having the experience (or the luck ) to do it can truly take advantage
of IT to achieve their business objectives. And we have many failures and successes to
learn from, and to build upon.
A key architectural and technology element of this convergence is an architectural
style that became known in the year 2000 as Service-Oriented Architecture. Since
2003, this has started to morph toward an architectural style with the potential
of impacting the vast majority of IT: an Enterprise Service-Oriented Architecture
(ESOA). In its widest meaning, this term indicates the architectural style enabling
an interoperability layer reducing the costs of integration, creating a technical
and functional decoupling between applications, and supporting an enterprise-wide
rationalization of IT applications. When ESOA is fully adopted, enterprises can align
IT to their business processes and create a transitioning path toward a normalized IT
portfolio. As such, an ESOA has the potential to profoundly impact both business
and IT.
It is indeed the rst time in our industry that the IT disciplines, technologies,
approaches are coming together to this extent. And still, IT is today more complex
than ever: addressing IT this way is still beyond the capability of most organizations,
both due to the costs and to the complexity of the task. The industry still needs to
mature to bring costs and complexity down. For this to happen, among other things
xvii
xviii
Foreword
we need to see books addressing the required architectural elements. This is one of
those books.
In 1996, Oliver Sims and I were working together on a large product development
(over 500 hundred developers building an ERP product with thousands of tables).
Oliver introduced a concept he called semantic messages. A semantic message
was a message that contained not only the data that needed to be sent, but also
tags describing the data being sent. For a couple of years, we explored together
the many challenges of addressing interoperability through semantic messages to
address development, interoperability, and deployment in-the-large. We built a
simple language for describing these semantic messages. This was the rst time I had
met the concept of what I later called a strongly-tagged language. We also built
the infrastructure, patterns, architectures, modeling tools and code-generation tools,
repositories, and processes to dene these semantic messages in an architecturally
consistent way. As often happens in our industry, in time I discovered that many
other teams had been working with similar approaches in their companies.
These approaches were not standards, but rather very proprietary approaches. So
when XML came out, many people were happy to nally see a tagged language
that was being standardized and adopted industry-wide: something for which we had
unsuccessfully lobbied in various standard bodies for years. Now nally the industry
had a standard as a basis for what we believed was the best way to address many
interoperability issues.
Of course, XML is just a very basic language, alone it cannot do much: to
achieve its potential, it needed a whole set of additional standards, technologies,
infrastructures, frameworks and tools to cover the whole spectrum of what is needed
for interoperability. Today, this is reected in the many Web Services standards, and
in the many products being sold to support Web Services. But, once these basic
technical layers are addressed, any serious project needs to address, among other
things, the architectural issues. The point is not (and has never been) the technology:
the point is how to use these technologies in the various architectural choices we have
to make to address the requirements.
The industry has come a long way. Many Web Services technology providers have
come and gone. The industry is stabilizing and consolidating. The Open Source
movement has brought costs down. Second-generation architectural approaches
provide maturity models for component blueprints and models enabling faster and
more reliable software manaufacturing. Now we have much experience with Web
Services infrastructures, and, not only in specic companies but industry-wide, we
know what works and what doesnt. For the business, this promises the elimination of
costly, redundant, and proprietary approaches, and the ability to integrate applications
Foreword
xix
quickly and easily. IT is rapidly commoditizing, and ESOA helps drive value higher
and higher up the IT chain.
And nally, we start to see books not treating the problem as a simple standard
or technology problem, not focusing only on the basic Web Services bricks, but
positioning Web Services within the larger architectural perspective. The book you
hold in your hands is a ne example of this.
This book has the advantage of having been written by industry practitioners covering
many perspectives in IT: the authors together have the right mix of technology product
perspective, consultant perspective, and large IT shop perspective. The book covers
many of the most important topics: components, registries, security, management,
transactions, and events. As such, it addresses an important need: bridging the gap
between technology and architecture.
Enterprise Service-Oriented Architectures are one of the most signicant evolutions
in the IT industry in the past few years. They share the spotlight with other significant evolutions and trends. These include the wide-spread adoption of enterprise
architectures, the creation of an enterprise architecture discipline that looks well
beyond software architecture to address the many business, functional, structural
and technical aspects of IT today, the maturity of governance and compliance process
frameworks, and the application of agile concepts to all aspects of IT (including
deployment of packaged software and outsourcing). But software architecture remains
the critical prerequisite for success in IT and in ESOA in particular: this book addresses
this prerequisite.
Thanks, guys, for putting together a ne and timely book.
Peter Herzum
President, Herzum Software
PREFACE
Dont tell people how to do things, tell them what to do
and let them surprise you with their results
George S. Patton
Service-oriented architectures (SOA) fundamentally changes the way enterprise

software is developed and deployed. SOA enterprise applications evolve. The change
will morph existing software, as we currently know it away from using monolithic
approaches. Instead, SOA will enable virtualized on-demand execution models that
break the current economic and technological bottleneck caused by traditional
approaches.
Software as a service has become pervasive as a model for forward looking enterprises
to streamline operations, lower cost of ownership and provides competitive differentiation in the marketplace. SOA is not a silver bullet that will address all of the
deciencies within an enterprise related to integration, reusability or the elimination
of redundant systems. SOA can assist in these problem domains in an incremental
manner while providing guidance on business architecture as well.
Service orientation has become a viable opportunity for enterprises to increase the
internal rate of return on a project-by-project basis, react to rapidly changing market
conditions and conduct transactions with business partners at will. Loosely coupled,
standards-based architectures are one approach to distributed computing that will
allow software resources available on the network to be leveraged. Applications that
separate business processes, presentation rules, business rules and data access into
separate loosely coupled layers will not only assist in the construction of better
software but also make it more adaptable to future change.
Service-oriented architectures will allow for combining existing functions with new
development efforts, allowing the creation of composite applications. Leveraging
high-quality low-cost applications that can be procured from third party suppliers to
xxi
xxii
Preface
Figure 1. Service-Oriented Architecture

replace lower quality components lowers the risks in software development projects.
By reusing existing functions that are of high quality, it leads to faster deliverables
and increases the potential for overall higher quality.
Loose coupling helps preserve the future by allowing parts of the business and the
technology that provides support to change at their own pace without the risks linked
to costly migrations using monolithic approaches. SOA allows business users to focus
on business problems at hand without worrying about the technical constraints within
existing IT applications as they can choose to replace components incrementally. For
the individuals who develop solutions, SOA helps in the following manner:
Business analysts focus on higher order responsibilities in the development
lifecycle while increasing their own knowledge of the business domain.
Separating functionality into component-based services that can be tackled by
multiple teams enables parallel development.
Quality assurance and unit testing become more efcient; while errors using
SOA have the potential to be discovered much later in the lifecycle, the overall
quantity will be reduced to support better service level agreements (SLA) and
development specications.
Component-based approaches when used within a service-oriented architecture
can aid in becoming reusable assets for systems and software that are constructed
using proper software development techniques.
Preface
xxiii
Functional decomposition of services and their underlying components with

respect to the business process helps preserve the exibility, future maintainability
and eases integration efforts.
Security rules dened for consumers and providers based on role, function,
context and method of interface that helps solve many security considerations
within the enterprise.
Implementing a service-oriented architecture will not happen overnight nor requires
enterprises to throw existing technology investments. The challenge will be in deciding
which services are core to your business and investing in them and leveraging services
provided by others as appropriate. In all technology revolutions the prior generation
usually remains a vital part of the infrastructure but not the only part. Software
built to support services signicantly reduces the need for enterprise software; not
eliminate it. Service-oriented architecture is an economically responsive model to the
needs of IT and enables the corporation to focus more on the business process and
the applications rather than the deployment of more enterprise software.
Service-oriented architectures also provide an opportunity for an organization to
mature their enterprise architecture perspective. Enterprise architecture provides
the necessary framework for categorizing everything for the purpose of supporting
future business planning. Both business and IT executives should be able to look
at the enterprise architecture and understand the tradeoffs in using various systems,
processes, technologies and so on. SOA can be the next level of specication
and provide guidance on the best ways to realize many of the goals of enterprise
architecture.
ABOUT THIS BOOK
The great revolution in our generation is that of human beings, who by changing
the inner attitudes of their minds, can change the outer aspects of their lives.
Marilyn Ferguson
The goal of this book is to share insight gathered by industry thought leaders in a
practical easy to read manner. This book contains many leading edge examples that
illustrate how agile approaches to enterprise architecture can be applied to existing
business and technology issues. It will help one focus on how to think concretely
about enterprise architecture while providing solutions to todays problems.
Within the covers of this book, you will learn about the following topics:
Fundamentals of a Service-Oriented Architecture,
Component-Based Services,
Orchestration,
Registries,
Management,
Transactions,
Event-Driven Architecture, and
Understanding Distributed Computing.
xxv
xxvi
About This Book
Audience
This book is for every Java and .NET developer and architects of Fortune 1000
enterprises and the consultancies that service them who have the drive to spend extra
hours feverishly seeking bits of wisdom on their craft and who want to gain the latest
insights and strategies for leveraging emerging enterprise SOA disciplines for value
creation, increased business agility and strategic competitive advantage.
This book does assume that one has signicant IT experience under their belt and
have worked on projects that were both large and small; on time as well as those
which are over budget using different project management, software development and
infrastructure paradigms. This book is not for those who desire all the information
they require in a single book as this is an impossible goal to achieve; rather this is best
suited for those who want to gain insight from thought leaders and are willing to be
savage in the leap from good to great.
Finally, this book is aimed at the people who must create applications in the real world
day-in, day-out. Many of the best practices books treat each tip as the sole focus of a
chapter with no discussion of integrating it into a real application. Real applications
are tenuous at best, requiring lots of moving parts in order to function together.
Concepts and recommendations within this book are presented in context of a living
enterprise with all the places that the real world intersects with the academia of the
problem space at hand.
The hardest part of starting an enterprise service-oriented architecture initiative is
knowing where to begin. We hope that our insights will be a useful starting point for
a set of well-managed endeavors for many an architect.
What This Book Is Not!

First, this is not a book for beginners. If you are looking for introductory material on
service-oriented architectures, we recommend searching for articles on the Internet
using search engines such as Google. Second, while this book may sporadically use
SOAP, WSDL and UDDI snippets to describe SOA concepts, this book is not about
web services. If you require a book on web services, we recommend Java Web Services
Architecture published in 2003 by Morgan Kaufmann.
Many of the examples contained within this book use XML. If you do not have a
thorough understanding, we recommend the following books:
Learning XML, Second Edition OReilly.
About This Book
xxvii
Effective XML: 50 Specic Ways to Improve your XML OReilly.

The authors recommend that the readers have a thorough understanding of general
software development, design patterns, enterprise integration and network infrastructure as a foundation to truly understanding the material contained within. Some
good books are:
Enterprise Integration Patterns Addison Wesley.
Pattern-Oriented Software Architecture, Volume 1 Wiley.
Network Security Architectures Cisco Press.
Introduction to Networking McGraw-Hill.
Software Architecture in Practice Addison Wesley.
Finally, if you have not programmed using a modern language such as Java and/or
.NET, we recommend the following books as a good starting point:
Java 2 Enterprise Edition Bible John Wiley & Sons.
Bitter Java Manning Publications.
Java 2: A Beginners Guide McGraw Hill.
Beginning VB.NET (Programmer to Programmer) Wrox.
How to Use This Book

The authors have taken deliberate action to avoid lling this book with lots of code
and specication information. This book strives to provide vivid insight into the
dynamics of SOA and a strategic approach to successfully integrate technology into
business decisions and the day-to-day actions of the workforce, business partners
and the consumer to reduce waste of resource and enhance the deliverables of the
enterprise.
The goal of this book is to strive to be both practical and philosophical. Sections
of the book may discuss general principles with examples that illustrate them:
examples drawn not from trivial demonstration programs (i.e. pet stores, shopping
carts, etc.) but rather actual architectures used in production systems in industry
applications such as nancial services, supply chains, manufacturing, defense and
telecommunications.
xxviii
About This Book
Motivation for Writing This Book

There is an abundance of books on Web Services on the market many of which are
bestsellers. The vast majority of texts simply explain vendor product offerings, APIs
and very brief tutorials without providing the reader with sufcient understanding
of the problems that exist within enterprises. The authors felt it is not sufcient to
merely learn the technology and APIs but to understand the principles behind them
so that they can be appropriately incorporated into existing systems and applications.
Many books also explain the happy path to software development. Knowing what
works is good (design pattern), but knowing what doesn?t work is better (antipattern). Where appropriate, this book will outline potential pitfalls. The authors of
this book have been doing service-oriented software development before it became a
hot topic in industry magazines and journals. The authors themselves are relentless
in the pursuit of nding better ways to develop software, which relies on thinking
about problems of the past differently. Success is bred by not simply understanding
technology but also by understanding agile methods for realizing the business goal.
The author teams primary motivation in writing this book is to share our working
experience to bridge the gap between the knowledge of industry gurus and newcomers
to service-oriented architectures. Many powerful techniques lay hidden in magazine
articles, conference proceedings, vendor marketing materials and academic papers for
years before becoming recognized by the masses.
NOTE: For those who are not familiar with Agile Methods, we recommend you read
the Agile Manifesto at http://www.agilemanifesto.org.
Disclaimer
The advice, diagrams and recommendations contained within this book may be used
as your heart desires, with the sole discretion that you may not claim that you were
the author. The publisher, authors or their respective employers do not provide any
form of warranty or guarantee its usefulness for any particular purpose.
This book is 100% error free! Not! If you believed us even for a second, I have a
suspension bridge in my backyard that I would like to sell you. The author team and
editors have worked hard to bring you an easy to understand, accurate guide on Enterprise Service-Oriented Architectures. If you nd any mistakes in this book, we would
appreciate your contacting us via email at [email protected].
About This Book
xxix
This book may use from time to time examples of a ctitious organization. Any example
companies, organizations, products, domain names, email addresses, people, places, events
depicted herein are ctitious. No association with any real company, organization, product,
domain name, email address, person, places or events is intended or should be inferred.
About the Authors

James McGovern
James is the co-author of several bestselling books, including A Practical Guide to
Enterprise Architecture. He is employed as an Enterprise Architect for The Hartford,
a Fortune 100 nancial services rm. James is a popular speaker at elite technology
conferences around the globe. He is a member of the Java Community Process and
the Worldwide Institute of Software Architects.
Oliver Sims
Oliver is a recognized leader in the architecture, design, and implementation of
service-oriented and component-based enterprise systems. Currently an independent
consultant, Oliver has held a number of senior technical positions in software product
and service companies. He was a founding member of the OMG Architecture Board,
has been active in several OMG Task Forces, and has contributed to the development
of MDA. Co-author of Business Component Factory and Building Business Objects, and
author of Business Objects.
Oliver has many published articles to his credit. Olivers experience in IT and
software spans three decades, and he has accumulated wide practical experience in
a number of roles and with many kinds of system. He was one of the rst to prove
the synergy between components and distributed objects, and was chief architect
for a groundbreaking component container middleware product in the mid 1990s.
Most recently Oliver has helped IT organizations in their transition to effective
component-based high-productivity development of service-oriented systems.
Ashish Jain
Ashish is an enterprise architect with over 11 years of industry experience with
expertise in the architecture, development and mentoring of the mid-tier and
server-side components. He currently works as a Principal Architect with Ping
Identity Corporation, a leading provider of solutions for identity federation.
Prior to joining Ping Identity, he worked with BEA Systems where his role
was to assist BEA customers in designing and implementing their e-business
xxx
About This Book
strategies using solutions based on J2EE. He holds several industry certications

from SUN and BEA and is also a board member for the Denver BEA User
Group. He is also the co-author of The J2EE 1.4 Bible (ISBN: 0-7645-3966-3,
http://www.amazon.com/exec/obidos/tg/detail/-/0764539663).
Mark Little
Mark is Chief Architect for Arjuna Technologies Ltd., a spin-off company from
Hewlett-Packard, specializing in the development of reliable middleware. At HP
Mark was a Distinguished Engineer and architect working on their transation and
Web Service products. He is one of the primary authors of the OMG Activity Service
specication and is on the expert group for the same work in J2EE (JSR 95). He is
also the specication lead for JSR 156: Java API for XML Transactions. He is on the
OMGs OTS Revision Task Force and the OASIS Business Transactions Protocol,
OASIS WS-CAF and OASIS WS-TX Technical Committees, to name a few.
Before joining HP he was for over 10 years a member of the Arjuna team within the
University of Newcastle upon Tyne (where he continues to have a Visiting Fellowship).
His research within the Arjuna team included replication and transactions support,
which include the construction of an OTS/JTS compliant transaction processing
system. Mark has published extensively in the Web Services Journal, Java Developers
Journal and other journals and magazines. He is also the co-author of several books
including Java Transactions Processing: Design and Implementation and The J2EE 1.4
Bible.
ACKNOWLEDGEMENTS
When the character of a man is not clear to you, look at his friends.
Japanese Proverb
A book like this is never just the work of those whose names appear on the cover. Like
the academy awards, there are so many people we would like to thank We are
immensely grateful for all those who have encouraged us, provided practical advice,
debated the ner points on controversial topics and whose insights have honed our
own.
The authors have beneted immensely in their own professions by reading papers
from other industry thought leaders including but not limited to (in no particular
order):
Jeff Schneider
Doug Barry
Martin Fowler
Doug Kaye
Anne Thomas Manes

Peter Herzum
The author team would also like to thank other writers we have worked with in the
past and desire to work with in the future (in no particular order):
Per Bothner
Jason Gordon
Leeanne Phillips
Sameer Tyagi
Elias Jo
Nitin Narayan
Kurt Cagle
James Linn
Sunil Mathew
Alan Williamson
Scott W. Ambler
Rahul Tyagi
Yakov Fain
Lynn Denoia
Vaidyanathan Nagarajan
Vikas Sharan
Dave Hollander
Kito Mann
Finally, the author team would like to thank our editor, Robbert van Berckelaer, for
allowing our ideas to be published in a manner the community sorely needs and
most importantly our copy editor, Jolanda Karada, for painstakingly copyediting and
ensuring this book is error-free.
xxxi
xxxii
Acknowledgements
James McGovern
Best wishes to the reluctant warriors within the Armed Forces who spread freedom
throughout the planet. Prayers to the families in Palestine, Israel and other parts of
the Middle East who seek peace and those who have lost their lifes in pursuit of it.
To all of my coworkers at The Hartford who have been merciless in the support of
speed, agility and balance in our daily lives. To Democrats, who make thievery and
cowardice sound so romantic. To Republicans, who make Democrats look principled.
Regardless of land, religion or language, there is just but one God. I must thank our
creator whom has bestowed upon me many favors. I am grateful for being blessed
with a great family, doing the things I love and working for such a great employer.
To say that I am not worthy of such blessings is obvious to anyone who knows me,
which makes me all the more grateful.
Oliver Sims
Such insights as I may have developed over the years have been mainly due to the
many valued colleagues with whom I have had the honor of working. In particular,
I would like to thank Martin Anderson, Alan Boother, Roger Brown, Peter Eeles,
David Frankel, Mike Guttman, Peter Herzum, Haim Kilov, Wojtek Kozaczynski,
Maurice Perks, Dave Roberts, Mike Rosen, Trevor Sharpe, Sandy Tyndale-Biscoe,
Rick Williams, and Bryan Wood.
Ashish Jain
I would like to thank all my colleagues at Ping Identity for helping me learn everyday.
In particular, I would like to thank Darren Platt and Brian Whitney for taking the
time to share their real world experiences with me. I would also like to thank my
ex-colleagues at BEA, John Funk and Bob Webster, for reviewing the content and
their invaluable comments.
Mark Little
I would like to thank all of my colleagues at Arjuna Technologies, who have helped
to make it a great working environment over the years. In particular Stuart Wheater,
Barry Hodgson, Dave Ingham, Steve Caughey and the members of the transactions
team, past and present. Many thanks go to Professor Santosh Shrivastava of the
University of Newcastle upon Tyne, who started my career off in this direction and
has been a constant friend over the many years. Thanks to my ex-Bluestone and
Hewlett-Packard friends and colleagues, including Bob Bickel, ex-general manager of
Hewlett-Packard Middleware, Al Smith, Greg Pavlik and Jon Maron, who showed
me that the best things in life are free.
ABOUT THE REVIEWERS

Synergism is the simultaneous actions of separate entities which together
have greater total effect than the sum of their individual effects.
Buchholz and Roth
The author team owes a debt of gratitude to all of the reviewers who provided
guidance, feedback, constructive criticism, praise and encouragement throughout
the manuscript writing process. Our reviewers came with diverse backgrounds: from
people who believe in traditional processes to those who have embraced agile methods;
from those whose native tongue is English, to those who speak joyous languages such
as Arabic, Chinese, French, Hindi, Spanish, Urdu and others; from those who are
lowly developers and project managers to those who are senior executives, this book
would not be what it is without you.
The author team deeply appreciates all of the feedback received during the manuscript
writing process from the following individuals (company afliations are included
where permitted):
Argentina
Daniel J. Biondi, Regional CT, EDS Latin America Solution Centres, Buenos Aires
Australia
Shaji Sethu, Solutions Architect, Volante, North Ryde, NSW
Belgium
Fanuel Dewever, Consultant, Human Capital Management, IBM Business
Consulting Services
xxxiii
xxxiv
About the Reviewers
Robin Mulkers, Head Application Architect, Christelijke Mutualiteit
Canada
Henry Chiu, Architect/Trainer/Mentor, ObjectSoft Solutions, Ontario
Finland
Anders Aspnas, Solution Architect, Fujitsu Services OY
Germany
Stefan Tilkov, Managing Director, innoQ Deutschland GmbH, Ratingen
India
Naveen Gabrani, Software Project Manager, Computer Sciences Corporation
Rohit, S. Gulati, CISA, CSQA, Iex Solutions, Bangalore
Shivprasad Koirala, AG Technologies, Nepal
Israel & Palestine

Issac Goldstand, Owner, Mirimar Networks, Jerusalem
Pakistan
Ammad Amjad, Systems Architect, Lahore Stock Exchange
Scotland
Jon C. Ferguson, PhD, Director and Architect, Omega Software, Ltd., Aberdeenshire
Singapore
Victor Putra Lesmana, IBM
About the Reviewers
xxxv
Ukraine
Ruslan Shevchenko, Director, Grad-Soft Ltd., Kiev
United Kingdom
Max Kington, Technical Architect, Cantor Fitzgerald, London
United States
Wayne Allen, Product Development Manager, Corillian, Portland, Oregon
Matt Anderson, Senior Enterprise Architect, Great American Insurance Company,
Ohio
Adolfo Lagomasino, Architect, Lucent Technologies, Landover, Maryland
Barbara McGowin, Principal, McGowin Enterprises, Goose Creek, South Carolina
Robert W. Meriam, Chief Infosec Ofcer, Catholic Health Services of Long Island,
Melville, New York
David L. Nicol, Founder and CTO, TipJar LLC, Kansas City, Missouri
Donna M. Schaeffer, Associate Professor, University of San Francisco, California
Ravi Sharma, Senior Enterprise Architect, Systems Development Factory, General
Motors, Detroit, Michigan
1
UNDERSTANDING
SERVICE-ORIENTED
ARCHITECTURE
Mans mind, once stretched by a new idea, never regains its original dimensions.
Oliver Wendell Holmes
The idea of software modules providing services to other software modules is a longestablished approach to software design. A service has three fundamental attributes:
the services description in the form of an interface denition, a mechanism for
accessing or consuming the service by invoking its interface, and an implementation
or a provision of the service that is, the code behind the interface. Today, this
approach is being revitalized by a new and standard technology Web Services.
In the past, Enterprise Systems have been dogged with a plethora of different kinds
of interface, each tied to a specic technology or commercial product. Project teams
have had to master a whole array of different ways of invoking interfaces of binding
or tying their software to other software often at high levels of technical complexity.
And it is the process of binding software modules or subsystems together that often
presents the greatest development problems. It has been said that plumbing never
leaks in the middle of a pipe: it always leaks at the joints; and so it is with software.
Although there have been a number of approaches to this problem, there has been
a lack of effective standards that apply not only between enterprises but all the way
down to between modules or subsystems co-located on the same machine. Companies
Diagrams in this chapter Copyright Sims Architectures LLP 2005.
Enterprise Service-Oriented Architectures
have developed their own in-house standards, sometimes centered on an enterprise

bus approach. However, whenever a new piece of software was acquired, it had to be
wrapped with the in-house interface.
Produced by the World Wide Web Consortium (W3C), Web Services are a set of
standards not only for dening interfaces but also for accessing those interfaces.
Together they provide a way to tie systems and modules together. They are
increasingly being adopted by software vendors, whether the software is middleware
such as BPM (Business Process Management) products or application packages. This
provides obvious advantages for enterprise IT organizations. However, many still
see web services as being for inter-enterprise use only. This is probably due to the
word web in the title. But nothing could be further from the truth. Unless the
standards groups that control the evolution of web services prevent it (a hugely
self-defeating proposition!), the same standard can be used within the enterprise
and across a wide technical scope. Sometimes it does make sense to not develop
web services. In situations where you have full control of both service and client
there are often considerable savings to be made (both in development effort and
communications latency/bandwidth) by avoiding web services and adopting a more
streamlined integration approach. Some of the web services middleware now available
even provide local optimization for access between software modules in the same
address space, making this less relevant from a performance perspective for in-house
consumed services.
But although it is the arrival of web services that has put the service-orientation
into Enterprise Service-Oriented Architecture (ESOA), there is much more to service
orientation than just a technology standard for interfaces. Service orientation in the
enterprise context is not just a neat technology that is also an international standard.
Service orientation is a pervasive mindset a way of thinking about things that
can help transform IT responsiveness and agility. Thus the business denes services
it wants to provide to customers and partners. IT denes services for and to the
business. IT develops or buys application modules that provide services to other
application modules, and applications to other applications. But the thinking can be
usefully extended further, even to organizational units within IT, so that they provide
services to each other. That is why ESOA is so much more than just web services.
Service orientation concentrates architects minds on provision of value, whether
from one module to another, or from one discipline to another, or from one part of
IT to another. Web service standards and technology, although perhaps the catalyst
for ESOA, are by themselves merely the icing on the cake. And who wants icing
without the cake when they can have both?
In this book we aim to show how to make a head start on baking a fully-iced cake.
We focus on the main areas where service orientation makes a difference, and also
Chapter 1: Understanding Service-Oriented Architecture
on the major architectural concerns that are affected by service orientation. We start
with an introduction to web services in the context of enterprise architecture. The
following chapters address component-based services, service orchestration, service
registries, security, service management, transactions, and event-driven architecture.
We hope that readers will nd how enterprise architecture can be re-vitalized not only
by the introduction of web services technology but also by the way a service-oriented
approach can make a desirable difference to enterprise architecture.
Introduction
This chapter presents the major aspects of a service-oriented architecture for the
enterprise. To be effective, such an architecture must address more than the technical
design of services: it must also provide a design for IT agility in the face of increasing
time-to-market and responsiveness pressures.
Enterprise Service-Oriented Architecture is centered on three major concepts:
Services offered over the web and within the enterprise using XML-based
standards both for the specication of invokeable services and for the message
ows involved in requesting and delivering those services. This is often referred
to as just service-oriented architecture (SOA).
Implementation of services using component-based software engineering
(CBSE) architecture (see Chapter 3), which not only addresses structuring
new applications, but also embraces such aspects as legacy wrapping, workow,
and Business-to-Business (B2B) specications.
An architectural, procedural, and organizational mindset that is service-oriented,
and which can merge the web services technology and CBSE potential into a
synergistic whole that can be hugely productive.
The rst of these concepts is becoming well known. Some services will be provided
by suppliers outside the enterprise. However, most enterprises know that competitive
advantage depends on their core systems and processes. They also think it foolhardy
to cede control of these, and more importantly the valuable in-house business domain
knowledge, to a third party. Hence the core services will continue to be provided
in-house, either though in-house development, outsourcing, or through packages.
It is sometimes thought that SOA means building applications by composing thirdparty services. However, someone somewhere has to build the code that provides
the services. A key question often overlooked is, how are the implementations of
business-oriented services best designed, structured, and built? The second concept
provides the answer to this question, and is addressed separately in Chapter 3.
Underlying these more technical considerations is the pressing need for a major
improvement in exibility and timeliness of systems. The third concept that of a
service-oriented mindset has developed as an important part of the response to this
need.
The chapter is structured into four parts as follows:
Section 1 (Introducing Service-Oriented Architectures) shows how a single architecture based on mature CBSE can address a wide range of capabilities including
Business-to-Business (B2B), Business-to-Customer (B2C), Business-to-Government
(B2G), workow, Business Process Management (BPM), and legacy wrapping, and at
the same time can provide the scalable enterprise services needed. In addition, we show
how a Service-Oriented Architecture can establish better separation between major
enterprise components, and enable inherently more adaptable back end applications.
CBSE itself is addressed separately in Chapter 3.
Section 2 (Service-Based Collaboration) examines how the basic service orientation
concepts can enable effective intra- and inter-enterprise collaboration. The key to
collaboration is federation a concept that enables loosely-coupled but effective
integration through higher-level services. Federation as the basis for collaboration
and integration can enable much more rapid business evolution and responsiveness
than other approaches, such as providing a new packaged or purpose-built application
for each new service requirement, or re-engineering two existing services into a single
combined service implementation that also provides the new higher-level service
required.
Building an Enterprise Service-Oriented Architecture (ESOA) is not easy. It is true
that some impressive immediate gains can be made, but the challenge is to ensure
that those early gains become the foundation of a truly productive and responsive
service-oriented system. Without such a foundation, the danger is that after a few
years all that has been created are yet more rigid and inexible stovepipes serviceoriented legacy! The ESOA that positions you not only for the present but also for
the future rests on a rm architectural basis that delivers:
Separation of technology platforms from the business services offered, and from
the business logic that implements those services.
Flexibility and responsiveness so that the services offered both within and outside
the enterprise can respond fast to the evolution of the business.
Section 3 (The Platform) deals with technology platforms that supports business
services, and, more importantly, its separation from those services. Such separation
is crucial in enabling business services to evolve at the speed of the business, rather
than being constrained by technology change.
Section 4 (Transition) briey discusses a process that can address the key aspects of
a transition to service-orientation. Moving to an agile ESOA environment is not a
simple task. A major inhibitor is the effort and management focus needed to plan
and take the rst steps towards the cohesive holistic environment required. This work
may well include some organizational aspects.
Of course, to be effective, and to truly enable business agility, there must be clear
traceability between business requirements and system implementation. Otherwise
the services provided may well not be those the business requires and needs. What
is needed is to connect business requirements directly with the kind of ESOA
discussed here, The next chapter discusses an approach for providing a near-seamless
progression from business requirements to service implementation.
1. Introducing Service-Oriented
Architectures
First we provide an introduction to Web Services and the place they can play in
enterprise systems. The aim is to provide an overall understanding of the technology
rather than a tutorial on its detailed application. A high-level model of our enterprise
system context is also presented to aid the discussion. Second, we examine the extent
to which web services can be used within an enterprise system, and conclude that they
can. This means that a single worldwide standard for services can be used not only
between enterprises, but also within them, even within a single address space! All parts
of an enterprise system can be provided with the same standard technology, fronting
legacy systems, message buses, new systems, business processes, and workow. Thus
the current Tower of Babel surrounding interface denition and invocation can be
conquered, and a new level of simplication becomes achievable.
1.1. Web Services

Web services have been standardized through the World Wide Web Consortium
(W3C). The specications mainly use XML as their alphabet.1 The reason for their
success over the past few years is that the specications are:
platform-independent,
effectively, a single platform-independent type system,
international standards,
relatively simple,
freely available, and
supportive of loose coupling.
These standards provide a highly effective lingua franca for inter-system communication. The two standards of concern2 for this chapter are at the heart of web services,
and are:
WSDL (Web Services Description Language), and
SOAP (Simple Object Access Protocol).
A third standard, also used in web services, is the Universal Description, Discovery
and Integration registry (UDDI).3 This provides for service publishing and discovery.
Before describing these three standards, let us rst describe our context through a
high-level and much-simplied model of an enterprise system that provides external
web services to other organizations.
1.1.1. Enterprise IT and Web Services

Web services are often thought of as an inter-enterprise concern. Later, we will show
how this is a very constraining view. For the time being, however, in order to illustrate
what, technically, web services look like, we retain this context. The main parts of
an enterprise IT system that provides and implements web services are illustrated in
Figure 1, and are as follows:
1 As a standard for carrying not only data but also the names of that data, XML provides for a degree
of resilience in the face of interface creep; however, most standards do not make use of this capability,
insisting on a rigidity that is often, in our opinion, somewhat over-specied.
2 Another W3C standard XML Schemas is important for the design of XML structures.
However, further description of this standard is beyond the scope of this chapter.
3 Although primarily designed for use with web services, the UDDI standard is not from W3C
it is from the Organization for the Advancement of Structured Information Standards (OASIS see
www.oasis-open.org).
The Core Enterprise System is the set of custom applications and bought-in
packages that form the heart of the IT system and that run on distributed servers
and/or mainframes. Security and scalability are of major importance here. The
Core Enterprise System is protected by rewalls and other security mechanisms.
The code that implements the core business logic behind web services, and also
manages core data assets, is here.
The Web Services Integration Broker is the set of middleware responsible for
handling B2B and B2G interactions with external systems.4 This includes storage
of the collaboration specications and their execution, and also invocation of
services provided by the Core Enterprise System.
Web/Application Servers handle user and B2C requests from browsers, both
within the enterprise and outside. User sessions, often long-running, are also
managed here. The sessions make requests for services to the Core Enterprise
System. For scalability reasons, these requests are typically one-shot requests,
such as update order or get stock details.
A Process Engine manages automated business processes and workows, and
is also responsible for storage of specications and their execution. There may
be more than one process engine. For workow, such engines typically provide
a user interface so that users can see and manipulate their work list, and so
that the work items can be managed. A process engine that addresses EAI will
often provide numerous adapters for a range of packaged applications, while a
Business Process Management engine may well provide complex peer-to-peer
protocols for collaborations with third parties systems.
User Systems are typically PCs on users desks, and are included here for
completeness. Some user systems access the core enterprise system directly using
a client/server style of interaction.
The typical interactions between these parts of the IT environment are shown by the
dotted arrows.
Now let us look in more detail at the technology of web services WSDL, SOAP,
and UDDI.
4 For an excellent white paper on the internals of this, see Cummins (2002). Broker function is
almost always provided within commercial Web Services products.
Figure 1. An IT View of the Enterprise and Its Partners5
1.1.2. WSDL and SOAP

WSDL or Web Services Description Language is a way of dening an interface to
a service. It is the external view seen by the developer of the service. The vast
majority of the time, developers will use a graphical IDE to interpret WSDL to
display it in a user-friendly manner. SOAP (Simple Object Access protocol) denes
the protocol on the wire for making requests and receiving responses. The
creators of WSDL decided to use the then-existing SOAP standard as the protocol
for accessing web services dened by WSDL. Together they form the basis of web
services the denition of services and the subsequent invocation of them.
This section illustrates how web services work through a highly simplied example.
Consider a B2B service that allows products to be ordered. The service species that
a request containing customer number and product number should be sent, upon
which the product will be ordered, and the purchase order number will be returned.
Error conditions are not considered in our example.
5 The gure mentions B2B and B2C. B2G and Business-to-Employee (B2E) have been omitted
for clarity, but should be assumed to be beside B2B and B2C respectively. B2E in particular is starting
to exploit mobile, wireless, and telecommuting trends for external access by employees, especially since
user systems are becoming more than browsers with rich client and smart device technologies.
If the interface were written using CORBA IDL,6 the order placement function
might look like this:
interface Ordering {
void placeOrder(in string CustNum, in string ProdNum, out
short PONum);
...
};
However, in the web service world, the interaction is described using WSDL (often
pronounced wizdle by practitioners), and the WSDL le is stored by the enterprises
Web Services Integration Broker.7 A simplied version of the WSDL corresponding
to the preceding interface denition is as follows:8
WSDL fragment for an Ordering Service
<definitions name="OrderingService"

<message name="placeOrderRequest>
<part name="custNum" type="string"/>
<part name="prodNum" type="string"/>
</message>
<message name="placeOrderResponse">
<part name="PONum" type="int"/>
</message>

<portType name="OrderingPort">
<operation name="placeOrder">
<input message="placeOrderRequest"/>
6 CORBA is the OMGs Common Object Request Broker Architecture. IDL stands for Interface
Denition Language.
7 This repository is indexed using another web standard, UDDI (Universal Description, Discovery
and Integration), whereby the service dened in the WSDL le can be found when potential users of
the service do not know its web location. In our example, we assume the location is known, and so
UDDI is not used.
8 WSDL examples in this chapter are based on WSDL version 1. At the time of writing, version 2 is
in preparation. Version 2 introduces new terminology and a different approach to packaging the various
elements of a service denition. However, the overall discussion here will not be materially affected by
the new version.
10

<output message="placeOrderResponse"/>
</operation>
...
</portType>

<binding type="OrderingPort" name="OrderingBinding">
<soap:binding style="rpc" transport="http"/>
<operation name="placeOrder">
</operation>
...
</binding>

<service name="SalesOrderService">
<port binding="OrderingBinding">
<soap:address location="http://.../ordering.asp"/>
</port>
</service>
</definitions>
The purpose here is not to provide a full explanation of WSDL, but merely to
illustrate the kind of thing it is. Note, however, that the placeOrder service is
handled by two separate messages one incoming from the external system to
the system providing the service (placeOrderRequest, and one outgoing from the
service provider to the requesting external system (placeOrderResponse).
The WSDL le constitutes a denition of the service provided at run-time. This
includes the services location (the URI to invoke the service), and the input and
output messages (if any) that will ow in a service execution as SOAP messages. In this
way, WSDL abstracts access to the underlying application data and processes there
is no specication whatsoever of the service implementation itself. The link with the
service implementation is specied, in the service-providing system, quite separately
from the WSDL service denition. Thus there is an absolute separation between the
interface (the service denition) and its implementation.This is a crucial aspect of
Web Services: not only are interface and implementation logically separated, but they
are also physically separated. It is precisely this that enables WSDL interfaces to be
provided for a very wide variety of service implementations, from Java on an app
server to COBOL on a mainframe all accessible from a huge variety of clients. And
it is largely this that make WSDL such an effective universal standard.
11
The (simplied) SOAP messages that ow across the wire during a service execution
could look something like this:
SOAP messages on the wire

<Envelope>
<Body>
<placeOrderRequest>
<custNum>AB123</custNum>
<ProdNum>XY-238/5<ProdNum>
</placeOrderRequest>
</Body>
</Envelope>

<Envelope>
<Body>
<placeOrderResponse>
<PONum>845273</PONum>
</placeOrderResponse>
</Body>
</Envelope>
A system that requests a service typically does so through some application that
sends the appropriate SOAP message. Creation of the application code that does
the mapping of (say) Java objects to SOAP messages, and the actual invocation, is
handled by tools. This code is often called a proxy the code that represents in the
client the actual service implementation in the service-providing system.
Just as tools generate proxy code at the requesting end or consumer end-point,
so the code that invokes the implementation of the service at the service provider
endpoint can also be generated. This code receives SOAP messages and maps them (in
various ways according to the particular web services integration broker middleware)
to the native language (e.g., Java) code that implements the service. Responses follow
the reverse path.
But how are services found? In a world of ubiquitous services, there has to be a
standard way for nding services that can be used. The UDDI standard (Universal
Description, Discovery and Integration) provides the basis for this.
12
1.1.3. UDDI
UDDI (Universal Description, Discovery and Integration) is a standard for registering
and searching web services within directories yellow pages as it were containing
web services. UDDI is comprehensively addressed in the Registries Chapter, and
only a brief overview is provided here. UDDIs main function is to provide for the
discovery of services. First, a service provider uses UDDI to store the who, what,
where, and how of a service that is, the publisher, service description, location of
the service, and the interfaces to access the service. Using these elements (as well
as registry-specic service categorizations) potential consumers of web services can
search for services. Then a potential consumer can use UDDI to locate an appropriate
service. This can be done by a human using a browser, or programmatically. Generally,
design-time discovery of services is a manual process performed through a browser. A
consumer of a service will iron out with the publisher details such as the suitability
of the service (does it meet enterprise requirements for availability, disaster recovery),
and register to use the service if authentication is required. Finally, the consumer can
connect to and use (interact with) the service. SOAP is used for messaging (with
other messaging standards also being allowed for).
An additional strength of UDDI is run-time binding. Web service clients should
be designed to cache the endpoint of the web service they access. Should a
communications failure occur, the client should query UDDI and retrieve the
endpoint for the service. It may well turn out that the reason for the failure is that
the URL of the service has changed, and hence the service access can be re-tried. In
this way, web service providers can move their web services (for example to perform
routine hardware maintenance, failover within a cluster architecture, or disaster
recovery) while minimizing impact on consumers.
The UDDI standard denes interfaces to a directory that are themselves web services
described by a WSDL. Web page interfaces to UDDI execute those UDDI web
services based upon human input through the browser, and display the results of
those web services.9
Although there is some provision made within the UDDI standard for some limited
semantics to describe a service, within the context of an enterprise UDDI is probably
likely to be used by humans to browse for services. In other words, it may turn out
to be more useful as a repository of services available than as an automated discovery
facility.
9 Examining UDDIs API documentation, schema and WSDL can serve as an introduction to more
complex web services and their use. For example, see http://www.oasis-open.org/committees/uddispec/doc/tcspecs.htm#uddiv2. See also http://uddi.org/.
13
Having introduced the technical elements of web services, we now examine how a
developer would use them. Again, the intent is to provide an overview rather than a
tutorial.
Creating and Accessing a Web Service

Within the enterprises IT organization, a developer rst denes a service using a
WSDL tool, and stores the resulting WSDL le in a repository managed by the
Web Services Integration Broker. This is shown as step A in Figure 2 (the developer
is shown at a user system: a developer is a user too of a system providing the
development environment). Part of this task is identifying the module within the
core enterprise system that will actually deliver the service.
Figure 2. Creating a Web Service

To access the service, a developer in the external enterprise:
1. Uses a tool to retrieve the WSDL le.
2. The tool then generates a proxy for the service in the programming language
of choice. The proxy is a programming language class that provides a friendly
interface, in the programming language of choice, for the developer to invoke
the service. Generated code inside the proxy handles SOAP formatting based
on the WSDL specication and communication with the system providing the
service. The proxy (and the application containing it) is installed on the external
organizations system.
14
3. When the application runs, it calls the proxy, which does everything necessary to
map the native language request made to it to SOAP requests that go across the
wire to access the service.
4. When a request is received by the enterprises Web Services Integration Broker, it
is mapped to a specic service provider (the implementation of the service the
thing that provides it) in the core IT system. This could be a legacy application,
an EJB, or some other application. Once serviced, a response is sent to the
Integration Broker (not shown in the diagram).
5. A SOAP response is sent back to the requesting system.
With proxies and adapters in mind, our picture of enterprise IT is now as shown
in Figure 3. The adapter effectively provides the route between the incoming SOAP
message and the service provider. The WS Broker and the service provider together
make up the service. But Enterprise SOA (ESOA) is not only about the design
of the service interfaces and the use of web standards such as SOAP and WSDL.
It is not only about the internal structure of the Web Service Integration Broker
(not discussed in this chapter). ESOA is also about the design and structure of the
internals of the service, and the extent to which service orientation using web services
technology can be usefully applied elsewhere within the enterprise IT system.
Figure 3. Web Service Proxies and Adapters
15
1.1.4. The Beginnings of Enterprise Service

Orientation
Consider the service interface provided by the Web Services Broker Adapter in
Figure 3. Now consider the service provision module in the same gure. This module
also has an interface of some sort otherwise it could not be called. Other parts of
the enterprise IT picture also have their own interfaces workow, B2B as a service
to internal parts, EAI systems that integrate legacy and/or packaged systems, BPM
systems, etc. There are in fact many different interfaces that developers must deal
with. Can we do anything about this tower of Babel?
Well, consider the advantages of web services listed previously. There is no reason
why these should not apply inside enterprise IT as well as outside. The result would
be a single kind of interface, using the same technology, for all internal systems that
provide a service. Many existing systems would need to be wrapped of course. And
performance would have to be considered. However, the potential advantage of a
single interface type, that maps to many programming languages, is a huge simplier
for enterprise systems. It is like a common hub. Indeed, considering the number of
other technologies that go with web services such as message queuing a single
interfaced type is almost a must a highly compelling simplier. And simplication
means effort reduction, which means cost reduction and/or faster response to business
needs.
Figure 4 shows web service interfaces (the gray [green in e-book version] lollipops)
not only on the B2B collaborations (our WSDL/SOAP example was an extremely
simple example of a B2B service), but also on the core system modules that provide
or implement services. In addition, web service interfaces are also wrapped around
the process engine so that process instances (including workow and BPM instances)
can be kicked off using a web service interface.
Note that this also reduces the variety of invocation mechanisms. In Figure 3, the
order placement service provided by the core enterprise system is invoked using quite
possibly four different mechanisms, for example:
CORBA from the Web Service Integration Broker;
MQ from a BPM or workow instance;
JMS from the web server;
COM from a PC using client/server.
Each of these has its own programming model, and this is often visible in the
applications making the requests. Wrapping all of these mechanisms with web
16
Figure 4. Internal Web Service Interfaces

services not only provides simplicity for the application developers, it also separates
the communications and messaging infrastructures from applications. This simplies
evolution of the service without impacting other applications.
One signicant benet is that the debate around synchronous and asynchronous
messaging models can now be divorced from the communications middleware, and
brought back to where it rightly belongs in the business developers scope. For
occasionally, it is required that a service should be invoked asynchronously with
respect to the invoking code. It has been said that the only way to do this is with
an asynchronous communications mechanism, such as a MOM (Message-Oriented
Middleware). It has been further stated that CORBA (for example) is inherently
a synchronous mechanism. However, proxy design can cater for either kind of
communication. Further, a proxy can be designed to provide either or both a
synchronous and asynchronous interface for the developer using the proxy. Hence the
developer may invoke a service either synchronously or asynchronously with respect
to his code, without being concerned about the precise communications mechanism.
So far we have treated the core IT systems as if they merely implemented services.
But what about their own internal structure? For this, we turn to component-based
development. Not the hyped components of third-party off-the-shelf plug-n-play
modules, but the much more realistic and valuable modules of mature CBSE
(Component-Based Software Engineering), which have been service-oriented from
their beginning (some time before the term came into general use). CBSE in the
context of the Service-Oriented Enterprise System is discussed in the next chapter. For
17
the time being, we assume that web services have CBSE-based implementations which
access the appropriate corporate resource. This assumption also addresses granularity
and dependency management issues for modules that implement a service.
1.2. Enterprise Service-Oriented

Architecture
Now we bring the two main threads web services and CBSE together. The result
is an enterprise SOA that applies to both web services made available externally and
also to core business component services built or specied for internal use. Figure 5
illustrates this. The business components within the Core Enterprise System area have
web service interfaces that is, they are specied using WSDL and invoked using
SOAP. It may appear that performance will be unacceptable when components are
co-located. Co-location can also have the potential of lack of support of distributed
transactions, security context propagation/trust and other important concerns. Good
middleware will optimize access, so that components in the same address space will be
automatically invoked directly rather than via SOAP messages. 10 (If the middleware
chosen does not provide such optimization, then of course performance considerations
may well dictate a dropping-back to possibly multiple different interface technologies.
In addition, groupings of tightly-coupled applications can be wrapped.)
Figure 5 also shows wrapper components that wrap legacy and/or packaged
applications for either process or data access. The wrapper components look like
components from the outside, but their internal implementation consists of whatever
is needed to access the legacy or packaged application, plus the relevant part of that
application itself.
The result is a single way of providing and invoking service interfaces, whether
externally or internally.11 In addition, all the advantages of CBSE accrue. Indeed,
Enterprise SOA has some unique advantages:
10 It is an ancient dictate of middleware, going back at least to the mid-1970s that remote/local access
should be transparent to the application developer, with local access being optimized. The introduction
of local interfaces in J2EE seems to have ignored this rule, and in some eyes was a giant leap backwards
in the annals of effective middleware. However, server middleware is available today that does observe
the rule.
11 Some might say that services of different granularity require that they be implemented using
different invocation or binding technologies. This was a reasonable position prior to the advent of
Web Services technology. However and so long as middleware suppliers provide for local optimization
of service invocations it is difcult to see how this position remains generally viable into the future.
Certainly in certain proprietary implementations of service and component concepts in the 1990s, the
advantages were reasonably well-proven for a wide range of situations.
18
Figure 5. Enterprise Service-Oriented Architecture

Provides, through the use of web service interfaces, a single interface and service
access design, which is independent of the underlying platforms;
Provides a single type system for interactions across and outside the enterprise;
Provides a clear architecture for the internal implementation of services;
Is tailor-made for use of model-driven development, such as the OMGs
Model-Driven Architecture (MDA see (OMG1)) strategy or Microsofts
Domain-Specic Language (DSL) developments (Greeneld and Short 2004).
Separates much more clearly the business logic in code, business rules, workow, or B2B collaboration specications from the underlying middleware
which includes integration subsystems, communication subsystems, component
containers, etc.;
Simplies the whole enterprise development environment.
In conclusion, Enterprise SOA is a concept whose time has come, and which will
form the foundation for future enterprise systems. Perhaps CBSE and Web Services
are indeed a marriage made in heaven.
19
2. Service-Based Collaboration
through Federation
The ability to have services collaborate to provide new services is an important part of
Enterprise Service-Oriented Architecture (ESOA). Without this, the various services
are islands silos tomorrows legacy. In order to achieve service-based collaboration,
our ESOA must address how services are federated such that they deliver the
collaborations required. A collaboration can then be said to be achieved through a
federation of services.
2.1. A Federation Is
Federal/federating/federation pertaining to a system of government in which several
States form a unity but remain independent in internal affairs; an association of
largely independent units. (COD 1982)
But what, in our context, are the units being associated? Answer: services. We
are talking about federations of services. Federation is therefore about creating an
environment where common services are made available for use in conjunction with
others.12
Let us start with a typical situation faced by many enterprises today. Figure 6 shows
three legacy systems (X, Y, and Z) that have been given a service interface (the
gray [blue in e-book version] lollipops) through wrappers, which could use EAI
technology. The term service interface refers to a programmatic interface whose
description conforms to the WSDL standard, as described previously in Section 1
of this chapter. The wrapper is implemented using the approaches described in the
same section.
Now assume the enterprise wishes to deploy a service that is a combination of the
existing services. It can do so by dening a business process (using one of the many
tools on the market) that rstly provides a service interface, and secondly is largely
implemented through invocation of the existing services provided by the legacy
systems. Process A in Figure 6 gives an example of this approach. And what we have
here is a federation: Services provided by applications X and Y are federated by
Process A to provide a new service.
12 If services are federated, this must imply that service providers are also federated. Federation of
providers can be implicit or explicit, and there is a range of governance issues that may apply such
federations. However, in this chapter we do not address these issues.
20
Figure 6. Service Federation
There are three things in this picture worthy of note:

1. Apps X and Y are not modied in any way.
2. Current services are unaffected.
3. The service provided by App Y is re-used by a second federated service (Process
B).
There are also three things wrong with the picture:
1. The legacy and packaged applications are often monolithic: although they may
contain many useful smaller-grained services, they are often not usable because
the application designers did not provide ways of invoking them from outside
the application.
2. Because of the monolithic nature of the applications, the service wrappings are
limited in their scope for service evolution.
3. The function made accessible by the applications may or may not map to
the function the business needs, and hence further federation opportunities are
limited.
21
What is happening in Figure 6 is that, while there may be impressive advantages in

providing services outside the enterprise, in a short time it will become apparent that
the drawbacks of legacy systems lack of exibility, high maintenance costs, etc.
will leak through to service provision. In effect, the enterprise will be exporting the
drawbacks of monolithic systems to their customers. When the business evolves, while
the service interface can be rapidly changed due to current technologies, the provision
of those services the applications cannot. A good service, valued by customers,
can become a poor service over a relatively short time by being unable to evolve at
the same (or faster) rate of a competitors service. We are in danger of exporting the
symptoms of our current legacy integration problem outside the enterprise slow
response to change, high cost of maintenance, possible poor performance, etc.
What we have in Figure 6 is a set of new legacy services! For example, a sales order
placed through Process A in Figure 6 is unlikely to be processed in a single ACID
transaction. This means that it may well not be committed. Handling situations
where compensating transaction must be launched in order to undo a unit of work
are well-understood. Propagating errors back to requestors is also well understood.
Providing for both in a federation over networks and across disparate systems is
another level of complexity. However, even were this to be achieved, conveying the
correct semantics of an error across a federation is a major challenge. Even within a
single system, we often cannot provide understandable error messages to the human
being who needs to know what has happened! In our example, somewhere there will
be a human being who is interested in the fate of the sales order even if it is only
as a monitor of exceptions in a fully-automated supply chain. How will that person
know whether or not the order was actually placed?
So service federation is not just about gluing together what is there already; it is
about seeing federated services as the basis for future systems architecture, from
medium-grained to very large-grained services. When we said that federation is
about creating an environment where common services are made available for
use in conjunction with others, we omitted the context, which is that of exible,
evolvable, low-maintenance service provision, so that the businesss time-to-market
and responsiveness goals are met.
22
What is a Service?
We have referred to services, service interfaces, and service implementation.
Figure 7 shows a simplied UML13 model of these concepts (cardinalities have been
omitted, since they are quite complex, and the model is, we believe, sufcient for
current purposes without them). The topmost class is the concept of a service instance.
This consists of a service interface, through which the service is programmatically
invoked or requested at run-time, and a service implementation, which is the
code that implements and provides the service. The lower three classes are the
development-time denitions and specications that, when put into production in
the run-time, offer and provide a given service.14
Federation means that a service implementation may be nothing more than an
invocation of another lower-level service, although federation will normally involve
at least a minimal amount of business logic and/or data mapping. Such a service is
often referred to as a choreography, since it causes the underlying services to dance,
as it were, to the tune of the federating service. However, in addition to choreography,
and unless the federation is entirely within the IT facilities of a single enterprise, there
will also be important additional issues such as security, data model semantics, and
trust.
Figure 7. Anatomy of Service

13 Unied Modeling Language see OMG2 (2003) for the formal UML version 1 specication.
UML version 2 is nearing nalization at the time of writing (see OMG3 2004).
14 In those cases where services are dynamically bound based on run-time discovery, then the interface
denition is a run-time artifact for the service consumer as well as a development-time artifact for the
service provider.
23
2.2. Federation and Mature CBSE

For many organizations, the legacy problem is possibly the most signicant barrier
to evolving modern application architectures. However, while it is occasionally
intractable, many EAI products enable a programmatic interface to be provided. The
key question is what should these interfaces be? This is addressed in the next chapter,
where we show how business elements can be identied and mapped one-to-one to
service interfaces, such that the interfaces provided by application-level code reect
the important business areas of interest. In other words, the interfaces that wrap
legacy should be modelled on this basis, so that they map directly to business needs.
In addition, Section 1 argued that new or re-factored applications should be
modularized according to mature CBSE principles. Such modules, implemented
as components, and aligned with business elements, can very easily be provided
with service interfaces. While a mature component approach offered this capability,
relatively few organizations have succeeded in achieving it. This has primarily been
because:
The initial focus for CBSE was on off-the-shelf markets in components. This
focus has not proved fruitful, as the lack of such markets today shows. Another
strand of CBSE focused on components as a better way of modularizing, and
this focus has been far more fruitful (see Chapter 3) if less widely-known.15
Focus on component technology per se. Effective component design and build
does not need component container middleware, although such middleware
makes the job considerably easier. But component container middleware alone
does not produce well-designed systems that meet the goals of CBSE.
Lack of focus on architecture, especially dependency management. Mature
CBSE requires several things to be combined into an effective design and
build environment, all informed by a clearly dened architecture which shows
how application systems built in conformance with that architecture will
meet the businesss needs for responsiveness and time-to-market. Lack of such
architecture, together with lack of processes that support a direct and clear link
with business requirements, has resulted in approaches to service-orientation that
have delivered new monoliths with rather pleasant service-oriented interfaces
that are slow and difcult to evolve over time.
15 The academic world seems to have focussed on CBSE as providing a market of off-the-shelf
components. One professor recently said to me that CBSE is dead, and service-orientation is the new
thing, and applications will be built simply by invoking appropriate services, regardless of where these
are or who implemented them. He seemed to have forgotten that a service must be implemented by
something somewhere, else it is like a chain letter!
24
Component middleware not only leaves technology glue to be built by most

if not all organizations making signicant use of them, but they it also tends to
have built-in design biases (such as EJBs rigid session/entity dichotomy) that
can sometimes make effective design more difcult.
Lastly, but probably most importantly, an inability (for varying reasons) to make
the organizational changes necessary. We will re-visit this question in Section 4
of this chapter.
Making mature CBSE viable has probably mostly been driven by what now can
be seen as a systemic service orientation. This not only applies to architecture, but
also to organization. This involves organizing the development environment in a
way that rigorously separates implementation and provision of the development and
run-time infrastructure from that of service denition and implementation. The
infrastructure organization provides services to the service development group. Thus
effective CBSE is to some extent dependent on a service approach being taken to
the IT organization itself (as recommended by the product line approach (Clements
and Northrop 2002), and also by the component factory approach (Herzum and
Sims 2000).
On the architectural level, mature CBSE requires that components should be as
autonomous as possible. However, they are designed from the start to cooperate
with other components to deliver a solution. This drives stringent dependencymanagement approaches in the architecture. One of the more important dependency
management schemes is based on the mediator pattern (Gamma et al. 1995). The
business element approach described in Chapter 3 helps with this through the
realization that business elements are naturally organized in this way. That is, there
is a spectrum within which collaboration of autonomous components happens; the
spectrum stretches from entities with very few business rules attached to them other
than basic validation through to high-level processes.
2.3. The Federation Spectrum

Figure 8 illustrates this spectrum applied to a simplied core business componentbased system that provides Invoice Management services. An Invoice Manager
component contains the main invoice-related processes and/or procedures such as
create invoice, print invoice, update invoice, etc. Important entities such as Invoice
and Customer contain the business rules and validation associated with the entities
themselves, irrespective of which process invokes them. Such an entity may make
use of bottom-level utility components, such as an address database manager, or
Address Book component. Utility components tend to be commodity in nature;
that is, if the enterprise could buy them off-the-shelf from a third party, it would.
25
Figure 8. Federation in the Core Business System

Processes, however, tend to provide competitive edge, and hence are likely to be built
in-house, or be based on a package that allows extensive customization.
A component higher in the spectrum can invoke any of those beneath it. This is in
effect a use of the mediator pattern in the large. The dependencies create an open
directed acyclic graph.
Now consider a second system, providing Order Management services, as shown in
Figure 9. Here we see re-use of the Customer and Currency Book components. Since
Customer is dependent on Address Book, this also appears.
Figure 9. Simplied Order Management System
26
These two systems could well be deployed, as shown in Figure 10. Hence these
components must be designed such that they provide services to any appropriate
requestor or consumer. A key part of this approach is rstly that the components
map to business elements as dened (or better discovered) in the requirements phase
of development, and secondly that each component is highly service-oriented such
that they can collaborate with others in providing business solutions. Thus in the
evolution of mature CBSE, there has been a shift from reuse of components as
pluggable modules of software technology to business-oriented reuse and business
collaboration of business services the components themselves being the service
implementations.16
Figure 10. Federation with Common Components

Finally, notice that the Invoice Manager and the Order Manager components provide
service interfaces. In this way, the collaborations that they choreograph are in
themselves large-grained components, which provide service interfaces to higher-level
components. But what are these higher-level components? They are the business
service implementations that that make use of lower-level processes. For example,
a customer order fullment service could span order placements and invoice creation
as well as other things such as delivery scheduling. The execution of a given instance
could take several weeks to complete.
One way of implementing this might be to modify existing services such as Invoice
Manager, so that the new process is embedded within them. However, this would be
a gross breach of encapsulation, and would be exceptionally poor design. Much better
16 Students of the history of this evolution can nd an early application of components as
business-oriented pluggable modules in Sims (1994).
27
is to develop a higher-level service implementation that invokes the lower-level

ones.
Figure 11 illustrates a Fulll Order service (probably implemented using BPM-type
tools) that makes use of the services provided by the lower-level Invoice Manager and
Order Manager services. These in turn make use of the even lower-level entity services
provided by Customer, Order, Invoice, etc. In effect, each service implementation is
a combination of a process or set of business rules plus specic usage of lower-level
services. The term often used is that higher-level components choreograph those at
a lower level. This is the essence of federation.
Thus we can see a loose hierarchy of processes. At the bottom, the processes
become procedures and algorithms, and indeed at the limit morph into data access
components, whose main function is to hide or separate the specic details of data
access. For example, the processes within a Customer or Order entity service
implementation are typically the sets of algorithms and business rules to do with
validating the internal contents of a Customer or Order (for example, order header
must have more than one order line, and each order line must refer to a valid
product).
Figure 11. Federation Using a BPM Component
28
2.4. The Spectrum as a Service Taxonomy

The spectrum then is partly data through process, but more importantly it is a
spectrum of business scope. Hence the main difference between different levels in the
spectrum is the business scope addressed by the service. This implies that the greater
the scope, the larger the granularity of the resulting federation. Service granularity
means the amount of process (and consequential state change) you get from a single
invocation of a service.
Figure 12. A Service Taxonomy

Figure 12 suggests a service ontology based on these concepts. While not a rigid
classication, it has proved useful in separating concerns, and in ascribing implementation technologies. Thus higher-level services are implemented by peer-to-peer
collaborations, or ow processes such as business process ows or workow. These
levels are typically implemented by business process management systems or workow
systems. Tools17 in these areas tend to provide for high-level denition of processes
17 Higher-level processes can be categorized into B2B collaborations, ow-type business processes,
and workow. However, this categorization is partly driven by the past evolution of tools. Tools in
this area tend to implement various developing standards such as BPML (Business Process Modeling
Language) and BPEL4WS (Business Process Execution Language for Web Services). These standards
are currently in considerable ux, and it is not yet evident which will win out in the longer term. The
tools and middleware that enables business processes to be specied, and provides for their execution,
29
(the service implementation), such that the service implementation is interpretive.

Lower-level service implementations tend to be much more algorithmic in nature,
and (with todays technology at least) tend to be implemented in programming
languages such as Java, through formal IT development processes, and running on
middleware providing for high transactions rates and scalability (such as component
container and transaction processing middleware).
This gure shows a spectrum of services, both internal to the enterprise or
organization, and also external services. Lower-level services are federated by higherlevel services.
Figure 12 may seem to suggest that a given process should be placeable at one of the
ve levels shown. However, useful sub-levels can be dened. For example, working
in a number of industries, we have seen that the bottom three levels in Figure 12
can be expressed as eight or so useful levels (Herzum and Sims 2002). Furthermore,
these levels tend to be the same within an industry area. For example, Supply Chain
Management systems have core processes that can be sub-divided into two layers:
core business processes (such as Inventory Management) and support processes
such as credit checking. Core entities can be sub-divided into contracts (e.g. sales
order), trading parters (e.g. customer), basic concepts (e.g. price, or inventory), and
materials (e.g. item or product). Finally, core utilities may be classed as books (e.g.
addresses) or support items (e.g. notes).
Higher-level and broader-scope services can also have multiple layers, and work is
on-going as to whether there are useful industry-wide categorizations.
Finally, the spectrum could be called a spectrum of instability, since the higher
the level, in general, the more frequently the processes change. We prefer the term
spectrum of exibility, to reect where high exibility and ability to rapidly evolve
are vital aspects of architecture technology selection. The lower levels, down to the
essential entities of the business, are relatively more stable, although specic sets of
business rules can sometimes change frequently. This argues for maintenance of those
sets of business rules such that new versions can be plugged into a component.
Further discussion of how this can be done is beyond the scope of this chapter.
Thus in a given situation, ow and collaboration services will appear at a number of
levels. What is important here is:
are often called Business Process Management Systems (BPMSs). Workow systems are addressed by
the WorkFlow Coalition (WFC), and have broadly the same overall characteristics as BPMSs. At the
detail level, they are quite different, and currently products on the market tend to be either BPMSs or
Workow Systems. It is not yet clear whether in future they will coalesce into a single kind of product,
although we see no a priori reasons why they should not eventually be able to come together.
30
Each is service-oriented, business function is provided through federations of

services.
Services are designed and implemented according to mature CBSE principle for
responsiveness and agility in the face of business evolution.
The SOA framework for federation is equally applicable for service use internal
and external to the organization.
These characteristics provide for organizational adaptability.18
In summary, enterprise SOA (ESOA) is the basis for enabling federation through
good modularization (components), clear identication of business elements that are
service-oriented and whose implementation will be components, architecture that
embraces the exibility spectrum, and, at the technical level, denition of WSDL as
the means to dene and invoke services. The latter factor is often overlooked, but is
of considerable importance. As mentioned in Section 1, WSDL scales very well. It is
applicable to all levels of the service spectrum. And it is a widely-accepted standard,
supported by a large and growing number of products and tools. For the rst time
we have a standard for all service interfaces, that can apply from the lowest-level
service to a majority of required standards.
Note that WSDL is not the sole web service denition language out there,
http://hinchcliffe.org/archive/2005/05/10/215.aspx
2.5. Federation Example

The various levels of service can be federated in many different ways. Figure 13
illustrates one example, where three enterprises (B, C, and D) collaborate in a virtual
enterprise (A in the gure). We can see at least four federations (the gure does not
show all the service interactions or federations):
Enterprise A provides a business process ow service (Provision Customer), whose
implementation is a federation that includes, at one of its parts, invocation of a
B2B collaboration process service (Contract Negotiation) among B, C and D.
Ds part of Contract Negotiation involves (when the contract has been agreed) a
federation of existing business ow services, one of which is Fulll Order. This
in turn
18 A good exposition of how service orientation can affect organization is to be found in the rst
several chapters of Harmon et al. (2001).
31
federates several core process services, one of them being Order Manager,
which in turn
federates several core entity services.
Figure 13. Enterprise Collaboration

Thinking service orientation at the enterprise level leads inexorably to thinking in
terms of systemic service orientation, both within the enterprise and outside. In this
section, we have discussed federation of services, where the federations themselves
become higher-level services in their own right but which are implemented by
well-modularized and accessible components, each of which provides its own services
and also will probably federate other components of lower granularity. This is the
essence of ESOA. Granularity is to do with business scope, not in the sense of the
scope of enterprise that needs to know about it, but in terms of managerial ownership.
The important consideration is that separations between levels of service federations
are not a technology matter but are formed by business drivers and concerns.
The goal for ESOA is nothing less than a worldwide mesh of collaborating services,
allowing each enterprise to focus on its core competence. A key part of this is
the concept of applying the same model internally, so that within an enterprise
32
each subsidiary, division, region, location, and even each department has its core
competence published as a service.
But we cannot just build each service implementation as a new stovepipe. We need
to bring different services together as and when required, without touching the
implementation of those services. The process of doing this is called federation.
This section has shown, based on concepts developed in earlier sections, how
services that are federations, and how federations of services, can be achieved. But
achieving service federations alone is not sufcient. We also need to change and
evolve them and introduce new services without being tied by the underlying
technology. Thus we must consider how to separate the business-level services and
service implementations from their underlying technology base. We need to provide
a platform that appears to business developers as substantially more stable and unied
than before, while retaining our ability to evolve the many underlying technologies
at their natural pace. Thus we need to consider the platform that supports an ESOA.
3. The Platform
ESOA is quintessentially about satisfying enterprise business requirements in a
responsive and efcient way. A major inhibitor to this is that the business changes and
evolves at a quite different rate than the rate of change in underlying software and
hardware technology. Why is this an inhibitor? Because most current approaches to
application software development fail to separate business logic needed to implement
services from the platform that is, the set of DBMSs, application servers, GUI
infrastructures, BPM engines, messaging systems, communications stacks, system
services such as logging, conguration, naming, etc., that underlie the business
services, and upon which they run.
Look at any business-level application code today, and the chances are you will nd
all sorts of technology code buried inextricably with the business logic, from GUIdriving code to thread and transaction management code. This means that changes
to business function can drag in technology concerns, and changes to technology
impact business logic. And this is not some minor techie problem. It means is that
every time the business needs something to change, or needs some new function,
the business logic developers are immediately immersed in software technology, and
vice-versa. Figure 14 illustrates the effect: not only do both business and technology
changes result in modications to the IT systems, but also each change is one of both
business and technology, thus doubling the effect of each change! By the way, this
applies to new function too, not just changes. And what if we need to move to a new
33
platform, or we are developing services for two or more platforms? It is imperative

that we shield our service implementations as much as possible from the impact of
underlying software technology, whether from changes to that technology, or from
the need to run services on several different platforms.
It has been said that model-driven development approaches as in the OMGs
Model-Driven Architecture (OMG) will remove this problem. However, if a model
generates not only business logic but also the technology glue, then we have certainly
simplied one part of the problem, but have introduced another the problem of
maintaining an increasing amount of glue-generating code! The argument for clear
separation of these two major aspects business and technology, in both models and
code remains.
Figure 14. Development Churn

In this section, we take a top-down view of the technology platform and describe
an approach that protects, as much as possible, the business services and their
implementations from the evolution and change that must occur as platforms mature
or change.
3.1. The ESOA Blueprint

How do we get a handle on separating software technology function from software
business services? Well, the A in ESOA stands for architecture. This is not there
just to pad out a snappy acronym. It means that service-orientation at the enterprise
level must be clearly designed, and that design must be visible. In other words it must
have an architecture. But architecture can mean many things. Here, we use it in
the sense of a blueprint that applies to a range of different applications, all built
according to the same structural principles. When a number of different applications
34
can all share the same structure, and where the relationships between the parts of the
structure are very similar, then we have what might be called an architectural style
(Hubert 2002). Much the same concept is also called a product line (Clements and
Northrop 2002), or part of an approach (Herzum and Sims 2000). An architectural
style can be expressed through what is sometimes called a blueprint (also called
a metamodel). Such a blueprint becomes the expression of the architecture in
ESOA. It is a detailed design for:
The structure of the modules to be dened and built by business developers
(including their dependencies and statefulness), how they relate to each other,
where the services are offered, and to whom (internal, external, etc.).
The transparencies enjoyed by business developers that is, the extent to which
software technology is hidden from them.
A specication of how extra-functional challenges such as scalability, buildability,
performance, etc. will be addressed.
One part of such a blueprint is the model of what a component is (see Chapter 2).
A further example of what the blueprint addresses is the distribution part of
a distributed system that is, the logical distribution tiers. These are areas of
responsibility or aspects of business logic as seen by the business developer. They
are separate from, but mappable to, the physical distributed system. For reasons
beyond the scope of this chapter, we prefer a four-tier model to the often-used
three-tier model. Briey, the tiers are:
User the specication of the user interaction with some device (including
screen layouts).
Workspace the business logic involved with managing a users session. This
tier provides services to the user tier.
Enterprise the business logic involved in providing enterprise-level services to
authorized requestors, and also the business logic inherent in business objects
the business-level resources that business processes depend upon. This tier
provides services to the workspace tier, and also to other areas such as B2B
collaborations
Resource the business logic required to access resources such as data (or
printing). This tier provides resource services to the enterprise tier.
Figure 15 illustrates these four logical tiers, and shows how they can be mapped to a
variety of physical systems.
35
Figure 15. The Four-Tier Distribution Model

Figure 15 is fairly high-level, and illustrates the distribution concept that is part of
the set of core concepts within an ESOA.19 However, the objective for the ESOA
blueprint is to dene things, based on these core concepts, at a sufcient level of
detail that the same business-level code can be mapped to several different physical
system structures, and to differing software technology platforms. This requires two
things:
A concept of development that rigorously separates business logic from the
technology infrastructure, and also explicitly addresses mapping of platformindependent designs to the necessary platform-specic artifacts. This is exactly
the province of the OMGs new strategy, Model-Driven Architecture or MDA
(OMG1 2003).
A concept and implementation of platform that enables the separation to be as
complete as possible. Of course, it can never be completely separate: if it were,
19 Note that this approach separates the orthogonal concerns of distribution of logical areas of
responsibility (workspace, enterprise, etc.) from those of separation of logical business responsibilities
(process, entity, etc.). It is not uncommon to nd these two concerns conated, resulting in business
processes being placed in a PC or in web server session logic, leaving only entities or business objects
in the shared server. This makes provision of business services difcult, since they either have to be
duplicated on the server, or an important part of the system must be re-engineered to separate them
properly.
36

then business logic could not run! But we can push the boundary of the platform
upwards so that separation becomes as clean and as complete as possible.20
The key to separation is to dene a virtual platform for business developers that
is deterministically mappable to a number of real platforms. Figure 16 illustrates
this. Note the additional blueprints (designs) for the virtual platform and for the
real platforms. For example, components built according to the business service
blueprint could be mapped or transformed to J2EE or a CORBA Component
Model (CCM) implementation or .NET, all of which support the component
concept. It can also be mapped to some transaction processors such as IBMs CICS.
The mapping or transformation (or partial transformation) would produce business
service components that are of the specic form supported by the target platform.
Figure 16. The Virtual Platform

Now this all seems like a huge task. But it is not. Most IT departments have
skilled people who produce bits of code, or add-ons to development tools, that hide
complexity and make things easier for application developers. A run-time example
of glue is a logging service (it is surprising how many real platforms do not
provide this, or provide one whose function needs to be expanded in some way).
A development-time example of glue is a script that takes an analysis-level model
and transforms it into the beginnings of a design-level model. In each case, what is
really being done is to dene a virtual platform and to provide for the glue to map
that higher-level platform to the real COTS (commercial off-the-shelf ) platforms.
20 MDA denes platform as follows: A platform in general is a set of subsystems/technologies
that provide a coherent set of functionality through interfaces and specied usage patterns that any
subsystem that depends on the platform can use without concern for the details of how the functionality
provided by the platform is implemented. (OMG1 2003, pp. 23).
37
However, the blueprints generally only exist in the minds for the glue providers, and
are often lost when the project ends. Figure 16 suggests that this process, so common
in so many projects across the industry, should be formalized and applied explicitly
within the enterprise so that they are not lost and re-invented project by project.
Section 4 presents a process whereby this problem can be xed.
3.2. Current Status for ESOA Platforms

A mature ESOA platform actually comprises two important platforms: the runtime platform, and the development platform used by business developers to build
service components that will run on the run-time platform. However, despite the
wide range of commercial off-the-shelf (COTS) products that are available today,
implementing an ESOA is difcult. There are three reasons for this:
1. COTS products are point solutions, not ESOA solutions;
2. COTS products are general-purpose; and
3. IT development is (typically) project-based.
3.2.1. Point Solutions

IT development organizations today are faced with a huge choice of middleware
and development tools. Somewhere on the market, you can nd everything needed
to implement an ESOA. The main constituent parts of EOSA run-time platform
include such things as an application server (component container middleware), Web
Service support, DBMS, and a user interface infrastructure. The function required
includes:
Web Service denition and publishing;
Optimization mechanisms so that web services can be used where required
within a single address space;
XML handling;
Transaction processing and Concurrency support;
Event management through a notication service;
Workow and B2B support;
User Interface infrastructure;
38
DBMS;
Caching mechanisms.
And for each of these, there are a number of sub-functions needed for example,
effective transaction processing needs thread and connection pooling, and event
management needs queuing.
The development platform includes such things as web service denition tools,
compilers, repositories, GUI design tools, and modeling tools. Again, sub-functions
are needed, such as the ability to interchange artifacts among the various tools.
The overall problem is that, although everything required is available, it is not
available in a single integrated product. This results in high levels of complication
across the whole development environment. From the CIO/Chief Architects view,
all you can buy from vendors today are big construction kits, where you often have to
make up your own assembly and operation instructions. There are lots of specs and
instructions for the individual parts, and much advice on sub-assemblies assuming
you understand all the parts. The main areas of complexity are:
Technical complications throughout the development environment. Addressing
these requires high levels of scarce skill for many development projects. The
result is low effective skill levels applied to many developments, with the
inevitable poor quality and re-work.
Rapid change in and evolution of software technologies. This results in high
levels of technology churn, and a disinclination to install new or upgraded
products. When technology is changed, this has a severe knock-on effect on
the business function portfolio. This is exacerbated by business evolution and
change (naturally) being out of sync with technology change.
Lack of focus on an ESOA architectural style.
Thus we have a situation where a great deal of work is needed to turn the collection
of products into an effective ESOA platform.
3.2.2. General-Purpose COTS Products

The reason there are holes in the COTS products available today is that producers
have not appreciated the concept of architectural styles. Hence they have had
to produce very general-purpose platforms, suitable for many architectural styles.
However, it is the concept of architectural style that enables a platform to be very
high level. Implementing services in a distributed enterprise system can done with
a particular architectural style, and hence the platform for that style could be much
39
higher-level than that provided by the general-purpose COTS products. There are
some indications that this may just may be changing. But do not hold your breath
for the next two years.
3.2.3. Project-Based Development

The third impediment to effective ESOA implementation is a strong project
orientation within the development organization. Of course, this does not mean
that projects per se are an impediment. However, where each development project
has signicant authority to choose (some of) its own technology code or products,
and also to build its own glue, then it can be extremely difcult to achieve a sensible
separation between software technology and business function. This is because in
such an environment:
There is little funding for efforts outside the project, and so:
There is limited re-use of common artifacts, whether software technology
or application modules, across projects; rather, there is often a great deal of
re-invention.21
Such organization means that expertise and artifacts are not shared across projects,
and so valuable learning and knowledge is dissipated at the end of each project. This
adds to the cost of each project, but such costs are not visible because the cost basis is
project-oriented, not service-oriented. Figure 17 illustrates this.
Now one of the striking things about IT organizations is that they are often relatively
impervious to the business pressure for dramatic time-to-market improvements. This
is because IT organizations are ghting on two major fronts:
1. Time-to-market pressures from business stakeholders, and
2. Dealing with rapid and wide-ranging technology churn.
And this when much current application development and design thinking is
optimized for yesterdays environment, characterized by OO and fat client design
plus rapid development processes such as RAD and XP. To this has been added
distributed system thinking characterized by thin or rich clients and app server
technologies. Finally, the underlying software technology is often researched and
21 By re-use I do not mean harvesting something, then adapting it, then testing it, then teaching
people about the adaptation, etc. I mean taking some asset and using it unchanged in a project.
Experience shows that re-use of this kind is a perfectly viable proposition but probably not in a
tradition project-based organization.
40
Figure 17. Project Orientation

adopted on a per-project basis, with little other than operating systems and major
middleware products being used as a common infrastructure across projects.
Changing to more productive development environments while facing such pressures
is difcult, and implies organizational as well as design paradigm changes. The major
inhibitor is the lack of an end-to-end cohesive product that directly aims at lifting
productivity an enterprise productivity platform (EPP). While few organizations
are able to meet the challenge of building an EPP as a marketable product, many can
move in the right direction using existing products combined with a tight focus on
harvesting their skills, and the assets that those skills have produced, to provide an
in-house equivalent. The objective is to share assets across projects, and have a single
way of lling the holes in the existing product providing the run-time platform and
the development environment. Without such a move within IT, business stakeholders
are likely to continue to be disappointed.
3.3. Filling the Holes

We talked about the prevalence of glue code, and the need to take an enterprise view
of the effort in producing it. To summarize this, Figure 18 shows the requirement at
the top, what is available in the middle (showing the holes), and the way that glue lls
the gaps at the bottom. From the business service developers point of view, where
much ingenuity and time must be applied to providing the business logic (processes,
procedures, algorithms, data handling, etc.), the less software technology the better.
However, much glue function is built from the bottom up, with important aspects
left to the business developer in case theyre needed. With a clear blueprint, the
41
things that are needed can be pre-dened, and a great deal of software technology
can be hidden much more effectively.
Figure 18. The Importance of Glue

In other words, there is an urgent need to view the glue as creating a virtual platform
for service developers. Space prevents detailed discussion, but two examples can
illustrate the kinds of concerns involved. Both look at providing simplicity and
transparency to business service developers in ways that are specic to the ESOA
architectural style, but independent of underlying platform technology. The two
examples involve invoking services at the client end, and building a GUI front-end.
3.3.1. Invoking a Business Service

First, we look at an example of the kind of glue code often provided for business
developers on an ad hoc basis by friendly colleagues working in the same project. The
example is one where a business developer needs to invoke a Place_Order service
(say from a PC or from a web server). The information he or she has is the order
data (in a variable order_data), and a key the string Place_Order to the URL
that identies the WSDL that is the effective interface to the service implementation.
The code the developer might write to place the order is as follows, where the code is
actually simple pseudocode, and lines beginning // are comments:
// Get the URL of the web service
url = wsdl_location_service.find("Place_Order");
// Bind to web service whose WSDL is at the specified URL
42

web_service = wsdl_system.bind(url);
// Create a proxy to use to access the web service;
order_proxy = proxy_service.makeproxy(web_service)
// Now invoke Create Order:
error = order_proxy.create(order_data);
Notice the three functions used by the developer: wsdl_location_service, wsdl_system,

and proxy_service. At least one, probably two, and perhaps all of these are helper
functions written locally to assist the business developer, and not provided by the web
service COTS product being used. The friendly colleagues have, on an ad-hoc basis,
picked out the parts that they believed could do with some help, and have provided
it in a way that reduces software technology complexity to some extent.
Now let us take a top-down view, and ask, if I am a business developer, what might I
prefer to code here. A good answer is:
// Find the "Place_Order" service:
bus_service = system_service.find("Place_Order");
// Invoke the service:
error = bus_service.create(order_data);
Experience suggests that this simplication is not only eminently do-able, but also
better than the following even briefer and still viable approach:
// Invoke the Place_Order service:
Error = bus_service.invoke("Place_Order","create",order_data);
Support for either of the above two approaches is more complex to write, even if it is
much more helpful to the business developer. However, this is the kind of thing that is
dened in the blueprint, written once, and provided as a standard aspect of the virtual
platform for all business developers. It clearly separates business logic from software
technology since this form can be mapped within the virtual platform to a number
of different technologies not only different web service COTS products, but also,
other non-web technologies such as CORBA. It can also provide for performance
enhancements.
3.3.2. User Interface

The user interface is an area often assigned to junior programmers to implement.
This is a mistake. The user interface is where the system actually delivers benet
to the enterprise. The less usable it is, the less benet. In addition, UI code is
particularly difcult to write well, requiring much low-level technical detail that is
technology-specic. For example, an HTML-dened UI is quite different from one
43
using Swing. However, it is possible to construct a UI framework part of the

virtual platform that not only simplies things a great deal, but can also shield the
business developer from technology differences. Such an infrastructure includes such
capabilities as:
Navigation schemes;
Widget standards;
Internationalization concerns;
Performance framework (so that the users do not experience the sticky mouse
syndrome, nor do they wait for server access when performing trivial operations
such as tabbing off a eld).
Today, XML is starting to be applied to the UI. In the mid-90s, I helped build a
component container middleware product which used a pre-cursor to XML as the
only way to dene GUIs. A layout tool produced XML-like denitions, which were
interpreted dynamically. This worked well on 16MHz processors running Windows
3.1, and on slightly more powerful PCs running Windows95. Performance was not
a problem. Neither was UI building (Eeles and Sims 1998). Today, this approach is
being developed using XML: XUL (XML User interface denition Language) (XUL)
is an interesting approach to simplication, as is Microsofts forthcoming XAML
(Extensible Application Markup Language pronounced Zamel, to rhyme with
camel (Petzold 2004)). Here is an example of a simple XUL panel denition
<xul>
<window title="Nexaweb Example" >
<button text="Button 1"/>
<button text="Button 2"/>
</window>
</xul>
Given an appropriate run-time container, this can dynamically produce a window
containing two push-buttons. We can expand this to make the push-buttons do
something:
<xul>
<window title="Simple Example Window" >
<button text="Button 1" oncommand="/myapp/button1click.jsp"/>
<button text="Button 2" oncommand="/myapp/button2click.jsp"/>
</window>
</xul>
There is already work being done to produce a layout tool, so that you do not
have to write raw XML. However, much of the work visible on the web is still a
44
point solution. Our experience in this area has taught us that a holistic system-wide
approach is required to integrate this effectively into an overall system where data
shown on user interfaces is obtained from a large shared database. For example,
consider a user getting a partial list of entities (e.g. Customers) from the enterprise
database, and double-clicking on one item in the list to open a window showing
that customers details. The function involved in this can be generalized to cover any
entity, so that all the developer has to do for a given entity type is to specify the
display format of the list, and of the panel where details will be shown.
3.4. Summing Up the Platform

COTS platform products tend to be point solutions, and are not integrated with other
products. It is a signicant job to integrate them, and to provide some support for
business service developers. Where that support is not provided, then the integration
task is placed on the business developers, and this overhead is sometimes repeated in
each development project. When support is provided, it is often ad hoc, and useful
glue is often not captured for re-use elsewhere.
Mature ESOA suggests doing this work once, capturing it, and re-applying it across
service provisioning projects. This involves consciously providing and maintaining an
ESOA virtual platform. While this is unlikely to increase overall costs over a period,
and should reduce them - perhaps signicantly, the cost prole within a development
organization will certainly change.
This does not mean a big bang approach. Proven techniques exist to evolve towards
an EPP for ESOA. The major inhibitor is the effort and management focus needed to
plan and take the rst steps towards the cohesive holistic environment required. This
means that in-house efforts to create an EPP cannot provide instant gratication
indeed, it may be six months from the start before benet is seen.22 The solution
to this is a focused transition program to move from ITs current state to the ESOA
virtual platform and EPP.
The next section addresses just such a process for transitioning to an ESOA virtual
platform, and the organizational implications of such a move.
22 The book Business Component Factory (Herzum and Sims 2000) was based on an early proprietary
Productivity Platform that was highly successful in the late 1990s.
45
4. Transitioning to ESOA
The previous sections have discussed the nature of web services, scaling up to
federations or collaborations of services, and separation of business from platform
concerns. While all of these are necessary for truly effective ESOA implementation,
by themselves none are sufcient. Applied together, they are mutually supportive and
highly synergistic. However, making them so is not just a matter of throwing them all
into the ring and hoping that something good will emerge. Rather it needs a focused
process to weld them together, and an organization that supports their application
rather than hinders it. Indeed, experience suggests that the greatest inhibitor to
effective ESOA implementation is lack of an appropriate organizational structure.
A big bang approach is almost certain to fail. This leaves evolution and so we need
an evolution process aimed at enabling the IT organization to make the journey or
transition to the desired goal. This section rst discusses the goal of ESOA, and also
ways of articulating the vision at the start of a transition towards the goal. Second,
we briey discuss how moving to ESOA involves organizational changes, so that the
core concepts are supported by the organizational structure rather than inhibited by
them. A process for achieving the transition can usefully follow the general approach
described by Guttman and Matthews (1998/1999); further description of such a
process is outside of the scope of this book.
4.1. The Goal

We assume that the overall business goals, and the supporting objectives set by the
business for IT, have already been clearly stated, and have resulted in a determination
to move to a service-oriented environment. The goal then is agile provisioning
of enterprise services, and avoidance of legacy service creation. Agile means
responsiveness and exibility; that is, not only can new or changed business function
be deployed rapidly in terms of minimal impacts on existing services; it also means
that the development organization can produce those new or changed services quickly.
Two key factors are instrumental in achieving this goal:
1. A set of core concepts that show the technical feasibility of the goal. These should
be articulated in sufcient detail as to enable people to say, Yes, this can work,
theres no magic. The previous sections of this chapter have introduced the
essential core concepts.
2. A process for achieving the goal that is, for applying these core concepts in the
development organization.
46
Taking this as a model, the rst thing to do is to articulate the vision of the desired
end state. This includes:
The core architectural concepts, which together dene the technical vision.
Separation of concerns along product line principles (Clements and Northrop
2002), to produce an ESOA Productivity Platform (EPP).
There are several ways to articulate the vision, including a PowerPoint presentation
with accompanying text, and a scenario of what things will be like for developers
and project managers (an example is provided later in this section). In addition a
target programmers model is also highly useful. This is typically dened early in
the transition process, and comprises target component code and target development
procedures. The target code is the code involved in a sample service implementation,
written with an eye to eliminating as much software technology as possible. The code
is annotated to identify how things work, and what facilities will be required of the
EPP to enable this code to run. The target developer procedures should be written
with an eye to eliminating as much work as possible from the development process
itself. This will typically assume integrated tools and an effective repository, thereby
identifying further function required in the EPP.
4.2. Separation of Concerns

Focusing on the development organization, there are four major areas of concern:
1. Service Design and Implementation;
2. The EPP;
3. The ESOA blueprint; and
4. Development Goals.
When these areas are well designed, teams can work at their own pace, based on their
resource/need balance, without overt interference. This is a critical success factor.
Figure 19 shows how these areas relate to each other.
The Service Provisioning area is responsible for the agile design, development, and
delivery of service-oriented application systems to the enterprise, including not only
new systems, but also such things as (the business logic involved in) EAI, and in
business process denition and testing.
Service Provisioning is the customer of the EPP area, which provides it with the
required high-productivity build and test environment. To meet his goals of agility
47
Figure 19. Organizational Separation of Concerns

and responsiveness, the manager of this area is motivated to demand the most
productive level of EPP as is possible. It is no bad thing if the Service Provisioning
area is empowered to obtain EPP artifacts from other sources than the EPP. This
keeps the EPP area on its toes, as it must satisfy the requirements of the Service
Provisioning area, which will be motivated by the demands of the enterprise for
responsiveness, time-to-market, etc. in other words, agility.
In turn, the EPP area is the customer of the ESOA area, which is responsible for the
core concepts and for architectural blueprints. They produce architectural deliverables
tailored to the needs of the EPP area. In other words, they provide an architectural
service. The manager of the EPP area is motivated to demand the most productive
mechanism for the delivery of architecture. For example, a 2000-page architecture
manual is much less use than a set of well-designed and tested UML proles. It is no
bad thing for people working within this area to be regularly seconded to the EPP
and Service Provisioning areas.
Finally, the Development Goals area, often a sub-group reporting to the CIO, is the
management function that:
Maintains the other three areas focused on the goals of the development
organization (which in turn must support the goals dened for IT by the
enterprise); and
48
Ensures that the appropriate provider/producer dynamics are maintained

between the other three areas.
It might be argued that other elements such as Process, Software Engineering,
Project Ofce, and Facilities (networks, operating systems, email, etc.) are also of
key importance; and so they are. However, experience suggests very strongly that
organizing such that the four areas discussed have clear and overt presence is a
vital and necessary aspect of achieving agile service-orientation goals. In addition,
experience also suggests that the other essential elements can be accommodated quite
happily alongside or within the four major areas shown here.
We deliberately use the term area for these four elements. This is because precise
organizational structures can vary signicantly. But however the actual organization
might be arranged, it is vital that these four areas are made very evident as autonomous
units, and that they are managed as such, with the usual management prerogative of
authority over resource allocation.
5. Summary
We started this chapter discussing web services, which can provide an international
technology standard for enterprise IT. Accessible not only from outside the enterprise,
this technology is also widely usable internally even within the same address space
on a single machine. So now all interfacing problems are solved? Well, no. Web
services provide the potential for a single software interfacing standard throughout
enterprise IT. But the realization of that potential must be consciously designed or
architected.
This chapter has outlined the major aspects of such an enterprise architecture based
on service orientation. Starting with the nature of the web services technology, we
have indicated how it can be applied systemically across the enterprise, how services
can be federated, and how business logic can be separated from underlying platform
technology. And we have briey discussed how a project or program to evolve to an
ESOA can be planned. In the next chapter, we look at the implementation of services,
and also discuss an approach to lessening the divide between business requirements
and the IT systems that deliver solutions to meet them.
2
COMPONENT-BASED
SERVICES
Success is neither magical nor mysterious. Success is the natural consequence of

consistently applying the basic fundamentals
Jim Rohn
Recently, someone (who should have known better) said of service-oriented architecture, SOA means that we dont have to build applications any more all we have
to do is invoke other peoples services. Thus a service implementation is merely a
collection of service invocations, and the services invoked are implemented by other
service invocations, and so on ad innitum. Like a successful pyramid or chain letter
scheme. Would that it were so easy!
The concept of a service can be usefully seen as comprising three parts:
Service denition dened by a service provider, and seen by a service consumer.
Service implementation code that, when executed, performs the business logic
required to provide the service. (The term code is intended to cover not
only application code but also specications or denitions that are interpreted
by an engine as occurs with Enterprise Application Integration (EAI), or
Workow, or Business Process Management (BPM) products.)
Service execution an instance of a service being provided and consumed at
run-time.
Diagrams in this chapter Copyright Sims Architectures LLP 2005.
49
50
The idea that SOA can work without a service implementation is absurd. Clearly
there has to be an implementation somewhere.
Chapter 1 dealt primarily with the service denition and execution. This chapter
addresses the implementation of a service. The implementation may well consist of
a choreography of other services, and they themselves may be implemented by yet
further services. Thus there can be a chain of subsidiary services involved in the
implementation of some top level service.1 Wherever in this chain there is some
business logic, then someone somewhere must dene and implement that logic. This
applies not only to process denitions that are interpreted by a BPM (Business Process
Management)2 or an EAI (Enterprise Application Integration) infrastructure, but
also to code written by application developers. In the latter case, as well as dening
the business logic, someone must also decide how to structure the software that
implements the service.
In the past, poor design structure has resulted in monolithic and opaque code that
has been costly to evolve, and where traceability of business requirements is lost. In
implementing services, it is essential to avoid creating tomorrows legacy. This chapter
proposes that services are best implemented through component-based software
engineering (CBSE).3 CBSE is much more than merely using component container
middleware. It is a design discipline whose core focus is modularization based on
the natural modularization of the problem domain. It is also a design discipline
that not only includes the best of previous software structuring practice, but also
allows for crucial architectural concerns such as granularity, distribution, exibility,
and rigorous dependency management to be addressed in a coherent way. A design
discipline such as this is an essential pre-requisite for development agility, applying
as much to out-sourced work as to in-house. Finally, we shall see how a unied
component concept cannot only address code structure, but can also encompass
BPM and EAI denitions.
The chapter is structured as follows:
First, the nature of mature CBSE is described.
1 In this chapter, we ignore the possibility, which could at least theoretically eventuate in an open
and trusted worldwide chain of dynamically-discovered and invoked services, of a loop!

2 In this chapter, and unless specied otherwise in the text, we use the term BPM to apply to
both business process denitions and to workow denitions. The difference between the two is often
held to be that workow includes steps that are performed by humans, whereas business processes do
not. This differentiation is useful when considering infrastructures provided by BPM and EAI vendors
versus those provided by Workow vendors. From an architectural point of view, on the other hand,
there are many similarities between the two.
3 CBSE is sometimes referred to by the term Component-Based Development or CBD, although
CBD has also been applied to other arguably less formal approaches.
Chapter 2: Component-Based Services
51
Second, we dene what is meant by the word component, building upon the
UML version 2.0 denition (OMG3 2004), which is very different from that
of previous UML versions.
Third, we consider the major architectural concerns inherent in mature CBSE,
and suggest by brief examples how these concerns can be addressed in a way
that exploits and expands CBSE.
Finally, to be effective, and to truly enable business agility, there must be
clear traceability between business requirements and system implementation.
Otherwise the services provided may well not be those that the business requires
and needs. Section 4 discusses an approach that can deliver a seamless progression
from a business requirements denition to service implementation with little
information loss.
1. Component-Based Software
Engineering (CBSE)
There has been much market hype about components being things bought from
third-party suppliers, so that application development becomes merely the assembly of
third-party components. However, given the almost complete lack of interfacing and
composition standards, together with the lack of denition about what a component
was and what it was not such hype was never very likely to bear fruit, and
indeed in the enterprise application area it has not (although there are some useful
technology-level components, mostly in the GUI area).
Components were also seen by many as being a technology thing (COM, EJB,
CORBA Component Model), where key technical characteristics have been tight
coupling (generally static) of low-level programmatic interfaces and synchronous
RPC-style messaging. This contrasts with the dynamic binding (loose coupling)
provided by such technologies as SOAP and WSDL, with their ability to invoke
services across different technology platforms, and support for several interaction
styles from RPC to asynchronous loosely coupled collaborations. (However, as
we shall see in section 2.3 Network-Style Interfaces, thinking of components as
merely tightly-coupled technology artifacts and hence inappropriate for SOA is to
dramatically miss the point!)
52
In short, CBSE has come to be seen as yesterdays hype, with little to show for it other
than some useful technology. However, if ever there was a case of the baby being
thrown out with the bathwater, this is it!
1.1. Understanding CBSE

Over the past ten years or so, CBSE has taken two quite different paths: rst the
market in components approach that has failed to bear much fruit (see Section 2.2
Federation and Mature CBSE of the previous chapter), and second a mature software
design discipline. The second approach (for example, see Herzum and Sims 2000;
and Hubert 2002) is a quite different animal from the unlamented market in
components hype of the 1990s. This approach, having been continuously developed
since the late 1980s, might be called mature CBSE.
Mature CBSE is primarily a way of thinking about application design (and then
applying that thinking of course). How you think about design is probably the
most important design choice of all. For example, forty years ago many application
designers thought of an application as a kind of written report, an algorithm that
started at the beginning (of the coding pad), went on until it reached the end, and
then stopped. When this structure did not match the problem domain that the
application was trying to implement, the result was reams of spaghetti code.
The history of software can be viewed as a search for better ways to think about
application design structures. First there was the idea of splitting applications into
separate jobs, each implemented by a separate batch program. In the 70s came
insight into the program internals, so that each program was split into modules,
based on its function. Inside each module, code was organized according the four
main structured programming concepts.4 This structure was well suited for the batch
systems of the time, but did not suit the event-driven systems that started to become
prevalent in the 80s.
Object orientation, introduced in the 80s, promised much. The concept was right
modules of code each of which implemented a clear thing, and which were eventdriven. The idea was to represent in software the things of the problem domain, for
example things such as customer, order, trade, etc. But at this point, something
went grievously wrong. OO was implemented through programming languages.
But a programming language addresses the inside of a module how you code
the internals. The result was that the objects of the problem domain disappeared
inside an amorphous module which itself did not represent anything recognizable
4 The four main structured programming concepts were: Sequence (of code statements), If-Then-
Else, DoWhile, and Case.
53
Figure 1. Applications vs. Components
by domain experts.5 Some modules were a whole application. We were back to

the single-module application of the 50s and 60s! This is illustrated in Figure 1,
where the important6 concepts in the problem domain are seen at the rst stages
of the development lifecycle (labeled Analysis here), after which, if going down
the traditional application track, they start to become successively buried in the
opaque deliverable which is an application in one or more anonymous modules.
Mature CBSE, on the other hand, renes the domain concepts and delivers them as
software modules that are invokeable through their interfaces, and which can plug
autonomously into the platform.
Now there is nothing wrong with the concepts behind OO. What went wrong was
that these concepts were not implemented by sensible middleware it was all left
to the application programmer. What a difference it would have made if the major
software objects had each been a separate module containing more minor objects,
each major module relating to an important and real thing in the problem domain,
5 Ivar Jacobsen once said that in the problem domain he could see objects; during analysis and
design the objects were all there; but once the programmers got hold of them they disappeared, and
never returned, being buried invisibly in applications. (Private communication to the author in 1993).
6 Few development processes address the issue of granularity in the business domain. Hence there is
often a problem with using vague terms such as important and major. Examples of major things
are customer, contract, and sales order taking process; examples of minor things are product
number, balance-on-hand, and address. Minor things are contained in major things. Section 4 of
this chapter shows an approach to formalizing this.
54
and independently pluggable into a system. But such pluggability needs something
to plug into a software socket provided by the middleware platform.
But which platform? Here we are not talking about the technology platform (J2EE,
.NET, CICS, etc.) but rather the virtual platform described in Section 3.1 of the
previous chapter. There is no point in building a component if a large part of that
component consists of bits and pieces of technology code. All that does is to increase
the surface area of the component increases the complexity of the plug that is
a necessary part of each component, and which plugs into the socket provided by
the virtual platform. A major reason why component technology platforms do not
deliver the goods as far as truly pluggable components are concerned is that they leave
far too many aspects to be handled by the developer of the component. For example,
such things as transactions, concurrency, activation, and access to platform services
such as logging, error management, publish/subscribe, and so forth.
But, one might argue, all such code can be generated, especially with the growing
availability of tools implementing the various avors of model-driven development
(for example, implementations of OMGs MDA, Microsofts Domain-Specic
Language (DSL) initiative, IBMs EMF, to say nothing of the many modeling
tools available today that provide for code generation). However, there are problems
with code generation. First, technology code generated for one component (for
example code providing for transactional behavior) may conict with code generated
for another, which means that these two components often cannot co-exist within
a given system. So much for pluggability and interoperability! We need platforms
that provide high levels of technology services, plus a standard policy conguration
approach, where each component is, on deployment, congured for a dened policy.
Second, the more code generated, the more must be re-tested and re-deployed when
something in that code changes. If all components contain signicant fragments
of platform-related (non-business) code, then changes to the platform can result in
massive re-generation, re-testing, and re-deployment.
All these points lead to a clear conclusion that raising the level of the platform
to as high a level of abstraction as possible is much more than a nice-to-have: it
is a necessity. Part of that high level of abstraction will be interfaces to platform
services that allow for exibility as those interfaces evolve which they will. The
key is to slow down the rate of impact on the business-level components. These
too are evolving as the business changes. An effective virtual platform will help a
great deal with this. Instead of both technology and business changes impacting
55
business-level components, in the main only business changes will impact them.
Technology changes will affect mainly the virtual platform.
Components, then, are objects writ large, with appropriate middleware to support
them. Components realize much of the original vision of object orientation.
Components are language-neutral and can be written using either OO or procedural
languages.7 Components are also coupling-agnostic; that is, they may be tightly or
loosely coupled.
But wait a minute consider a system such as supply chain management. A count
of the software repository would show thousands more probably tens of thousands
of OO classes. Does that mean tens of thousands of components? Well, no. As
indicated previously, a component is normally a collection of classes. An important
CBSE concern is dening which classes belong to which component (remember that
a component is not only a deployable module, it is also seen as a big class in its
own right). So, for example, a Sales Order component could consist of an Order
Header class, an Order Line class, a collection class (an instance of which would hold
the Order Line instances), and several smaller classes such as Currency, Order Value
Calculator, etc. CBSE is also concerned with identifying and scoping the components
in a system. So our supply chain management system could end up with perhaps
300500 components, each consisting of a number of classes.
Let us now dene what a component is in more detail.
2. A Component Denition
What is a component? There have been many descriptions and denitions of
component. A useful summary of various aspects, including component vs. class,
and component as a type or as an instance, can be found in Atkinson et al. (2002).
Over the past several years, there has been a developing consensus as to what a
component is, culminating in the OMGs UML2 denition (OMG3 2004). This is
quite different from that given in UML1, and Section 2.1 reviews the new denition.
However, the OMGs denition deliberately applies to many different kinds of
component, including low-level technology components such as GUI components.
For ESOA, it is useful to rene the OMGs denition, and a denition of an
7 Although not available today, probably due to the justied lack of interest in procedural languages,
in the mid-90s there was middleware available that supported mixing components each written in a
different language all in the same address space.
56
enterprise component is presented in Section 2.2. One of the key differences

between a UML2 component and an enterprise component is the nature of the
latters interface. Section 2.3 discusses the network-style interface provided (and
required) by enterprise components.
2.1. The UML2 Component

The OMG has recently standardized version 2.0 of the Unied Modeling Language
(UML). An important aspect of UML2 is its much-expanded component concept.
Whereas UML version 1 saw a component as essentially a build-time and deploymenttime module, UML2 has hugely enlarged the concept, and has extended it across
the development lifecycle, so providing standard support for what many architects
involved in mature CBSE have been modeling for some time now using their own
UML proles. A UML2 component is a three-part construct consisting of:
A type;
One or more interfaces (both required and provided); and
A realization.
The standard (OMG3 2004) states, The component concept addresses the area of
component-based development and component-based system structuring, where a
component is modeled throughout the development lifecycle and successively rened
into deployment and run-time. A component denes its behavior in terms of
provided and required interfaces. As such, a component serves as a type, whose
conformance is dened by these provided and required interfaces (encompassing
both their static as well as dynamic semantics). Figure 2, a slight simplication of
the diagram in OMG3 (2004) illustrates the three-part nature of a UML component.
Figure 2. The UML 2.0 Component
57
A component, then, is a kind of class. However, this is not the class you would
use in normal modeling. It is a special kind of class, dened by UML2, and whose
properties include:8
Ability to have an internal structure and ports (a port is a distinct interaction
point both externally with the environment, and also internally to internal
parts).
May be designated as having its own thread of control, or of executing within
the context of something else.
May specify which signals (such as events) can be handled
Behavior that can be fully or partly described by the collaboration of owned or
referenced classiers. (A classier is essentially an abstract supertype within
UML that is a type, that denes a classication of instances (a set of instances
that have features in common), that can include features of various sorts, and
that can be sub-typed.)
Based on this supertype, UML 2.0 denes a component as follows:
A component represents a modular part of a system that encapsulates its contents and
whose manifestation is replaceable within its environment. A component denes its
behavior in terms of provided and required interfaces. As such, a component serves
as a type, whose conformance is dened by these provided and required interfaces
(encompassing both their static as well as dynamic semantics). One component may
therefore be substituted by another only if the two are type conformant. Larger pieces
of a systems functionality may be assembled by re-using components as parts in
an encompassing component or assembly of components, and wiring together their
required and provided interfaces.
A component is modeled throughout the development lifecycle and successively rened
into deployment and run-time. A component may be manifest by one or more
artifacts, and in turn, that artifact may be deployed to its execution environment.
A deployment specication may dene values that parameterize the components
execution.
Note that the realization of a component can include other components. This allows
for various granularity strategies to be applied. In addition, it is expected that a
component that does not include other components will be realized by a number
of classiers, and a good example of a classier is a class. An example of realization
specications are shown in Figure 3, which is a modication of an example in OMG3
(2004) and illustrates a wiring diagram of components. This is a new kind of
diagram introduced in UML2.
8 For full details, see Section 8 Components of OMG1 (2003).
58
Figure 3. Component Wiring and Component Internals (UML2)

Four components are shown here: Store, Order, Product, and Customer. Store provides
an OrderEntry service, and its realization comprises three other components, Order,
Product, and Customer. Store provides an interface called OrderEntry, and requires an
interface of type Account. Store delegates implementation of its OrderEntry interface
to the Order component via its CreateOrder interface. The Order component requires
OrderableItem and Person interfaces. Product provides an OrderableItem interface,
and Customer provides a Person interface. In addition, Customer requires Account,
which is delegated to Stores required interface.
Note that Product and Customer should be designed so as to provide services to other
components. This is what autonomous means, and is really a design statement.
That is, Product and Customer (and Order and Store for that matter) should be
managed during their design and construction on the basis that they are autonomous,
that their services may be used by a number of different components, and that they
are in the business of providing services to those as yet un-named components.
Experience suggests that (at least when working with the kind of virtual platform
discussed in Section 3 of Chapter 1) the effort to design a service-oriented component
autonomously is no greater than designing one that turns out to be constrained to
work only with one calling component.
But do we really want to design, say, a Customer component, that might be a core
entity (see Section 2.4 of Chapter 1), so that its services are available on the web
to all and sundry? Of course not! We design such a component so that its services
are used by other components executing within a domain that is behind rewalls
59
etc. This means that the services provided by each component should have a dened
scope of visibility within the enterprise as well as outside.
Figure 3 also shows another realization that of Order. Four classes are shown
is realizing Order: Order, OrderHeader, LineItem, and Collection. Order has the
UML-provided stereotype focus, showing that the Order class provides the Order
component with its essential type. Other classes are auxiliary (this stereotype is also
provided by UML) and show that they play the role of the supporting cast as it were
within the component, assisting the Order class to fulll its role as the focus within
the component.
In this way, the dichotomy of type is resolved. In a business information model,
one may well expect to see the classes Order, OrderLine, possibly OrderHeader, and
also many others such as Customer, Address, etc. Domain experts will assert that
there is a concept Order that has within it the concepts of Order Header and Line
Item. But there is also the concept Order that both provides business logic and also
choreographs the other classes involved. So we have Order as a group that ignores
any internal arrangement and choreography of classes, and we also have Order as the
choreographer and holder of some of the business logic. This is the dichotomy. UML
provides for (and so resolves design difculties associated with) this dichotomy. The
component model element (a type) is used to represent the concept Order, and the
component realization is used to show the choreographing Order class and also the
other classes involved without which Order would not be an Order!
Finally note that there is nothing visible in the diagram to show whether Store
physically composes the other three components. Typically, components are not
physically composed; rather Store refers to them.
Now let us return to SOA. The key aspects of the UML2 component that are really
useful for service-oriented architecture are that it:
Is a type;
Can be visible and tangible throughout the development lifecycle and also in
the run-time;
Can be built and deployed autonomously (an example of autonomous
deployment is updating a system with a new version of the component);
and
Is intended to be combined and composed with other components with which
it collaborates to deliver solutions.
60
The UML2 component goes far in providing what is needed for service implementation. Based on the UML2 component, we can dene an Enterprise component
the kind used in building enterprise-strength service-oriented systems.
2.2. The Enterprise Component

An Enterprise Component is an extension of the UML2 component that exhibits all
of the following characteristics:
Represents and encapsulates a single business concept (process or entity) that
provides one or more business services through its interfaces;
Is of medium to large granularity;
Has one or more well-dened network-style service-related interfaces;9
Is intended to be managed at run-time by a middleware container which
provides the run-time technology socket into which the component plugs.
In other words, it enjoys the separation from technology concerns via the kind
of virtual platform described in Section 3 of Chapter 1;10
Addresses the scalability and distribution challenges of enterprise distributed
systems;
Can publish events and can subscribe to them (assuming the availability of a
notication service);
Can be managed at design time by a modeling tool that can be said to provide
a development-time socket that is, it knows about components, and can
import a component model (or design) into a larger system model that
comprises collaborating components;
Supports both synchronous and asynchronous messaging, synchronicity being
with respect to the ow of control within the component, and not to the
underlying communications stack. In this way, message synchronicity is completely separated from underlying messaging middleware. Although middleware
9 An enterprise component will also have an administrative interface which can provide for aspects
such as systems management and monitoring. These are not considered further here.
10 With such a virtual platform, the design of an Enterprise Component may ignore a whole host
of technology factors such as transactions, event management, conguration, activation/passivation,
systems management, and provision of any required technology-oriented interfaces. Late in the
development cycle, the code required for implementation of technology bindings, and also of policies,
can be generated or merged (perhaps using aspect-oriented weaving techniques) into the component.
61
supporting this and other transparencies has existed in the past,11 providing
this transparency today requires glue to be added to the available middleware.
This capability can be implemented in proxies that are typically generated.
The underlying invocation will be a Web Service invocation as described in
Section 3.3 of the preceding chapter.
A component also has an internal implementation or realization. This realization
appears both at design time and at build-time. The detailed realization of a component
can vary widely in nature. However, it is useful to distinguish between two main
forms:
1. A programmed realization that is, the internals are coded using a programming
language such as Java or C# (or even a procedural language such as COBOL
in some circumstances!), or by an interpreted language such as Python. The
number of classes in a component implementation typically varies from around
ve to as many as one hundred and twenty or so. An example of a programmed
component is a server-side component implemented with EJB technology and
written in Java. A programmed component is generally intended to be run on
component middleware such J2EE or .NET, and its creation requires IT technical
and management skills.
2. A declarative realization that is, the internals take the form of a declarative
script, such as a BPM process denition (usually dened using a GUI-oriented
tool, with the script itself being stored by the tool as an XML document). The
code that executes the script is an interpreter provided by middleware (such as
BPM middleware). Examples include workow and B2B denitions. Production
of a declarative component does not always require IT technical and management
skills.
It is important to understand that an Enterprise Component is primarily a design
and application code construct, and is mapped to underlying middleware technology
as far towards the end of the development lifecycle as possible. In addition, it assumes
a reasonably high level of virtual platform. Thus it is ideal as the design vehicle for
systems built according to the emerging concept of model-driven development such
as OMGs MDA.
2.3. Network-Style Interfaces

A network-style interface is an interface that can be invoked from other address
spaces in the same system, or from geographically remote systems, or even from
within the same address space. Invocation from within the same address space
11 See Sims (1994).
62
requires middleware that provides optimization of the invocation. There are two
important initial constraints on an interface that makes it a network-style interface:
References passed as operation parameters must be references to other
components.12
Instances of classes are always passed by value, never by reference. That is, a
reference to something that is not a component, but which must be passed as a
parameter (for example, an Order Header class), is always passed by value.
These two constraints make it much easier to avoid the object spaghetti that resulted
in some early distributed object implementations where designers assumed that all
objects could be distributed, and ended up with dense webs of dependencies in the
form of object invocations that owed across networks. Systems built this way were
quickly found to be unworkable due to seriously appalling performance: iterating
an OOPL collection instance across a network is not a good idea! And, of course,
maintenance of such systems, with multiple cross-network low-level dependencies
was in many situations quite impossible. The lesson is that some important techniques
in classical object-oriented design and implementation, although good for code that
runs entirely in a single address space, do not scale.
A network-style interface is also loosely-coupled in the following ways:
Data carried on a request to an enterprise component should carry associated
metadata (data labels or tags). Experience over the past twenty years or so
has shown that use of metadata in messages can be exploited in some very
valuable and innovative ways. These include better management of interface
creep, and a number of ways of providing generic code that dynamically adjusts
its behavior based on the metadata provided to it. Prior to XML, this was
done using proprietary mechanisms. Today, service orientation assumes that the
messaging mechanism will be web services using XML as the data format. This
means that advantageous exploitation already identied, but not widely-known
because of the lack of standardization of the data formats, now becomes available
to service-oriented architectures (although they do not seem to have yet been
picked up by the Web Services community).
Note that this is a design not an implementation statement. Loose coupling
can be implemented using web services, and for example J2EE component
12 Strictly speaking, this statement implies that there is a concept of reference scope, where any
given kind of reference has a specic scope. For example, a URL (for a web service for example) can
have worldwide scope; a CORBA IOR has scope within a given network of interconnected CORBA
systems (which may be extended by, for example, the COM-CORBA inter-working standard); a C++
object reference has validity at most within a given address space in a single computer.
63
container providers are beginning to provide for this. However, there will be
times when tighter coupling is needed. In this case, a network-style interface
can be implemented using a more tightly-coupled interface technology. The
opposite is generally not true.
Middleware implementations of network-style interfaces will normally provide
some means of optimizing invocations if the target component is co-located
that is, in the same address space as the invoking code. If this is not available,
then either appropriate glue code must be provided (quite a tricky task), or
compromises must be made.
One such compromise is to assume the existence of network-style interfaces until
build time, when tightly-coupled interfaces are substituted. With MDA, such a
decision could be parameter-driven, with the necessary code being generated.
Can be invoked regardless of the underlying communications infrastructure.
That is, the component designer/programmer does not need to know over what
communications channel the component will be invoked. Thus a component is
insensitive as to whether an invocation is carried by a language call from within
the same address space, by RMI, JMS, CORBA, HTTP/SOAP, etc.
Can be invoked across asynchronous or synchronous communications stacks.
Is insensitive as to whether an incoming message was originally sent synchronously or asynchronously with respect to the invoking code. Providing this level
of transparency with current COTS middleware will require additional glue
code, but the advantages in simplicity for the application developer are great.
Some of the glue code may well be resident in the component, but such code
can be generated.
Can operate as a port. That is, can accept data without there being an operation
name. This provides for Enterprise Components that model business processes
to be implemented as BPM specications.
It may seem that these kinds of capabilities will result in an unwelcome overhead
for a component. However, aside from optimizations and function provided in
middleware, an enterprise component is large enough in terms of number of classes in
its implementation to be able to handle some overhead. Remember that a component
(specically a programmed component) is a kind of small program. Indeed, when
compared with a class, an enterprise component is best thought of as an adult in the
software world responsible for its own fate (thus, for example, when deactivated by
middleware for garbage collection reasons as opposed to shutdown an enterprise
64
component should be able to decline deactivation. This would be specied through

policy rather than coding.
This mind-set, when well supported by the virtual platform, can provide for further
technology transparencies, and so make life simpler for the application developer. As
a useful piece of serendipity, the nature of the Web Services model provides support
for this concept. For example, there is no concept in Web Services, as there is in some
component middleware, of a client having to create the component with which it
wishes to communicate.
Finally, circumstances often exist which prevent full implementation of enterprise
components according the description given above. Experience suggests that getting
as close as possible, and being clear about where the boundaries are, still gives
signicant benet. For example, in a web-based distributed application, application
artifacts that run on a web server and a browser are such things as Javascript, ASP
pages, servlets, and HTML GUI panels. While the separations of concern, and the
design mindset of CBSE and the enterprise component concept are still very useful
and can be applied to some extent, it is not appropriate for example to try to provide
a servlet with all the attributes of the network-style interface. Hence common sense
compromises based on the virtual platform which will depend on the COTS
middleware capabilities are always appropriate.
3. Component Granularity
This section discusses the often thorny issue of component granularity. There are a
number of approaches to handling component granularity, and we choose to describe
the approach taken by Herzum and Sims (2000). The advantages of this scheme are:
Denes three clearly different kinds of enterprise component that range from a
component consisting of a small number of classes up to an entire application;
Provides for a component design scheme that is platform independent, and that
can be mapped to a number of different platforms;
Integrates scalability and dependency management aspects;
Extends across the distributed system from GUI to database;
Addresses re-use directly, and clearly denes what is intended to be potentially
re-usable and what is not;
65
Provides for a direct linkage from a requirements model (see Section 4 of this
chapter);
Denes a granularity level that is of immediate assistance as a viable unit of
project management;
Can be applied partly or wholly depending on circumstances.
The granularity scheme presented here assumes the presence of a virtual platform as
described in the previous chapter. This means that components will largely implement
only business-related concerns, and for present purposes we assume that this is wholly
the case. In reality there will be some non-business software technology that must
exist within components and that will be visible to application developers; however,
such technology as there is does not invalidate the assumption.
Since one of the levels of granularity includes distribution, we begin by describing a
distribution architecture that provides four logical distribution tiers, and that can map
to a number of physical tiers. Section 3.2 then presents the four different granularities,
and nally Section 3.3 discusses the main aspects of dependency management.
3.1. Distribution Domains and Tiers

The previous chapter (in Section 3.1) introduced a four-tier model, where each tier
is a logical area of concern, as follows: user tier, workspace tier, enterprise tier, and
resource tier. This section briey presents a conceptual background the big picture
which denes distribution tiers and domains, and then describes each.
3.1.1. Looking at the Big Picture

The four-tier model actually consists of two linked two-tier models, each based on the
concept of an inner or core tier, and an outer or external tier. This is shown in
Figure 4, where a system is seen as comprising a number of logical distribution tiers.
A distribution tier is a logical area of responsibility in the distribution dimension of
a system.
At this point, we should re-iterate a crucial point: the distribution tiers being discussed
here relate to application code only, assuming a high-level virtual platform. That is,
they are tiers as seen by the application developer. While a similar tier analysis can be
made for middleware, and is then of use to middleware developers, this aspect is not
considered further here.
66
Figure 4. Logical Tiers and Domains

Since a distribution tier is a logical construct, it can be mapped to one or more
physical tiers (for example an application server, a web server, a browser, a PC). It can
also share a physical tier with other logical tiers. A distribution tier can have many
components placed within it. Finally, inasmuch as a distribution tier is the result of
an architectural separation of concerns, the concept can also apply (very benecially)
to non-distributed applications.
An application code can be seen as having two different kinds of logical responsibility:
Core Tier: a logical distribution tier that realizes business function rather than
the handling of external devices such as GUIs or DBMSs. It is the result
of the architectural separation of concerns between managing externals and
implementing business logic. It can be seen as the business core of a system.
A core tier may be accessed by several outer tiers, and occasionally by none. In
other words, the core tier provides services to zero or more outer tiers.
Outer Tier: a logical distribution tier whose responsibility within the system is
to manage, dene, or access a resource such as a user interface, a data base, a B2B
link, and so forth. It handles external protocols and data formats, and maps
them to/from the internal canonical form(s) dened for the core tier. An outer
tier may have trivial business logic embedded in it (for example, maintaining a
total of amounts entered on a GUI).
Logical Distribution Tiers can be combined to form a Domain. For example, the
Application and Data tiers of the traditional three-tier model might be seen as making
up a server domain. A Domain is a signicant logical part of a system that has a
67
specic responsibility within the system, and that is the focus of interest across many
viewpoints. It is also a technical management domain; it is an area of the enterprise
IT system that is technically managed as a unit. Examples are the business-related
application software artifacts on a users workstation, and the set of address spaces
and systems within which an ACID13 transaction is supported. Chapter 7 discusses
transactions in some depth.
Figure 4 shows the relationships between tier and domain concepts. The term
invokes in the gure means some form of programmatic invocation, such as a Web
Service invocation, a CORBA or RMI call, a message sent via a message queuing
middleware, or sometimes an intra-process transfer of control. A Domain consists of
one Core Tier and usually one or more Outer Tiers, although sometimes a Domain
has no Outside Tiers.
It can be shown that the three-tier model is in fact a core and two outer tiers, all
within a single domain. But the essence of enterprise distributed systems is that there
are two kinds of domain in the end-to-end distribution area, and a third domain that
addresses BPM. Let us rst look at the two end-to-end domains, and the tiers within
each.
3.1.2. Distribution Domains and Tiers

The two end-to-end distribution domains are the user-workspace domain and the
enterprise resource domain, as shown at the bottom of Figure 5.
The User Workspace Domain (UWD) is responsible for support within the wider
system for a single person using a user interface, and consists of a User Tier and
a Workspace Tier. The User Workspace Domain accesses one or more Enterprise
Resource Domains in order to access data (read or write), or other system resources
that are also shared by more than one person.14
The Enterprise Resource Domain (ERD) is responsible for provision of services to
multiple concurrent authorized requestors, and also for the integrity and protection
of those resources. A given Enterprise Resource Domain is dened by the scope of an
ACID transaction running in that Domain, (possibly over several physical systems).
Hence a given system may include more than one Enterprise Resource Domain. If so,
13 ACID is a transaction processing acronym that stands for Atomicity, Consistency, Isolation, and
Durability the required attributes of a recoverable transaction.
14 For example, in a rich client situation, the UWD normally resides on the PC. In a browser plus
web server situation, the UWD normally maps to the browser plus the one or more sessions on the
web server that are dedicated to a single user.
68
then responsibility for coordinating transactions lies outside the Enterprise Resource
Domains.
The nature of requests for services made to an ERD from another Domain can be an
important scalability factor in distributed systems. Where a single service request is
handled within a single ACID transaction, scalability is signicantly enhanced. If a
transaction is held open over several requests, then the system scalability reduces. The
architecture should enforce this single shot ERD access pattern. The ERD provides
one or more services. When one is requested, it runs in a single transaction.15
This rule maps well to a common business rule. That is, the business requires some
change of state to be either completed or not done at all, and is not interested in any
intermediate state, and wants the change to be made as fast as practicable.
Figure 5. The Four-Tier Model

Within each of these domains are two tiers, making the four-tier model shown in
Figure 5. The tiers are:
1. The User Tier is an outer tier that provides user interaction logic through a set
of user interface components such as windows, panels, menus, and prompts (or
15 This rule implies that even queries (read-only) will needlessly run in a transaction. For some
systems, this will incur an unacceptable overhead. In this case, the architecture should be changed to
show read-only requests running non-transactionally.
69
their voice equivalent). It depends on the Workspace Tier for its source and sink
of data, and for its management in terms of user sessions.
2. The Workspace Tier is a core tier that provides a users model, and contains
user session logic and user-side representations of processes and information.
This tier consists of a set of components that communicate with components in
the enterprise tier.
It is possible that non-re-usable components are in this tier too, components that are
really to specic to one application can be potentially re-used elsewhere.
3. The Enterprise Tier is a core tier that provides enterprise-level business function
services, and is also responsible for protecting the integrity of enterprise resources
at the business logic level. Services can be either process-oriented (which typically
provide the service interface required by, for example, web services), or entityoriented (which typically provide entity services for the process services and
are typically not exposed outside this tier). The Enterprise Tier depends on
the Resource Tier for access to data and other resources (such as high-volume
printing).
4. The Resource Tier is an outer tier that provides components for access and/or
updating data and other resources. The Resource Tier provides persistent storage
services to the Enterprise Tier, and accesses mechanisms for storing data in some
manner, such as a standard relational DBMS, or an object-oriented database. It
may also provide access mechanisms to legacy system data.
Finally, the User Workspace Domain sometimes has its own persistence tier (not
shown in Figure 5). This can occur when it is important to persist the state of a
user session when the user logs off (for example, to go home at the end of a day
when a unit of work is only partly complete). Thus the User Workspace Domain can
sometimes have three tiers user, workspace, and user-resource; or, in other words,
presentation, application, and data.
3.1.3. The BPM Domain

One additional domain completes the picture. This is the BPM domain, which is
usefully seen as orthogonal to the User Workspace and Enterprise-Resource domains.
This is because a single workow denition, for example, may invoke both workspace
and enterprise tier components. The BPM domain often has only one tier a core
tier. It may be argued that an outside tier is necessary, in that adapters (to legacy
70
systems or to COTS application packages) are needed. However, we assume that such
adapters are effectively part of the infrastructure, and that application developers need
not be concerned with them. This is not always the case; business-oriented developers
may need to dene data transformations to do with integrating data from different
adapters. In that case, an outer tier which might be called the data transformation
specication tier, can be dened. Again, workow requires some form of work list
visible on a users screen. However, these are typically provided by the workow
product, and as such are not the concern of the application developer. The BPM
domain is discussed in detail in Chapter 5.
Having discussed distribution aspects, we now move on to consider a useful scheme
of component granularity, one level of which will use distribution concepts.
3.2. Granularity Scheme

A service-oriented architecture must dene clear levels of granularity for service
implementations. Since services are assumed to be provided by components, then this
means that component granularity must be dened. Without this, it becomes difcult
to dene how components are composed, and how the various scalability aspects are
taken into account. Several different granularity schemes have been proposed; here
we briey present the scheme described in Herzum and Sims (2000) (and used with
some variation in Combine (2003)) which denes three levels of granularity within
the user-workspace and enterprise-resource domains:
1. Distributed Component (DC);
2. Business Component (BC); and
3. Application Component (AC).16
The relationships between these three are as shown in Figure 6. The most coarsegrained is the application component, which consists of a number of business
components, which in turn are composed of distributed components.
16 This level of granularity is called a System-Level Component in Herzum and Sims (2000). Here
we use the term given to this concept by Combine (2003).
71
Figure 6. Three Levels of Component Granularity
3.2.1. The Distributed Component (DC)17

A Distributed Component or DC is an enterprise component (as dened previously)
that is typically implemented by platform component models such as EJB, CORBA
component, COM component,18 various BPM denitions, and even a set of CICS
transaction programs. The DC is the smallest level of granularity, and is responsible
for the implementation of services. EDOC (OMG2 2004) uses the same term,19 but
Combine (2003) refers to it as a Service Component. In general, most component
architectures provide for this level of granularity.
In spite of its name, a DC exists in a single distribution tier. Put another way, a
given distribution tier consists of a number of DCs. Figure 7 illustrates how this
works, where round-cornered boxes are DCs of various kinds. Enterprise component
architectures often dene different kinds of DC; the kinds in the gure derive from
Herzum and Sims (2000), where EDC for example means Enterprise-tier DC,
17 When rst mooted as a concept at a higher level of abstraction than code, the DC was named
distributable component, reecting its network-style interfaces, together with the idea that it could
be moved to a different machine and (pace performance) would still work correctly. However, it turned
out that distributable component was much more difcult to use in conversation than distributed
component, which in turn is usually shortened to DC.
18 CORBA Component middleware is now becoming available; for example, see
http://openccm.objectweb.org/.
19 It was the intention when EDOC was being dened to include a full specication of a distributed
component. However, for various non-technical reasons, the specication was not completed at the
time EDOC was standardized.
72
Figure 7. Different Kinds of Distributed Component

this being further divided into process and entity DCs. Other authors give them
different names for example, Harmon et al. (2001) call this a Process Controller.
The user tier in Figure 7 illustrates how the artifacts of a DC and hence a logical
distribution tier can at run-time be distributed over more than one physical tier.
The process panel which is a GUI panel relating to a process that is deployed in a
browser (probably dynamically generated from the denition on the web server). For
example, this might be an order processing workbench, which would embed other
panels relating to entities (entity panels not shown in the gure). The Process UDC,
which Harmon et al. (2001) call a view controller, is located on the web server.
Hence at run-time, the user tier artifacts appear on two physical tiers.
In summary, a DC can have a number of quite different implementations. A user-tier
DC could consist of a C# module and a panel denition (emerging XML-based
technologies for GUI panel denition, such as Microsofts XAML (Rector 2003), are
of interest here). An enterprise tier DC could be implemented as an EJB, and a BPM
domain DC could be implemented as a workow specication or B2B collaboration
script.
The several different ways that a DC can be implemented illustrates a really important
aspect of the DC concept: it is a design concept that maps well to a wide variety of
different technologies. Each DC provides services, some local, others of very wide
73
scope. In addition, the DC concept is sufciently closely specied20 to provide for

rich design models at the MDA Platform-Independent Model (PIM) level, with the
concepts preferably presented through a UML prole in an appropriate modeling
tool. Finally, a DC is large enough to be able to carry some overhead to help provide
transparencies for the application developer. For example, mapping from/to XML
can be handled at run-time by generated code within the DC, so that the application
developer deals only with normal classes in the programming language being used.
3.2.2. The Business Component (BC)

Consider a system that has no concept of, say, customer (a rather articial
supposition, but useful for exposition). Now ask the question, what artifacts would
be required to add Customer to this system? We might immediately say a Customer
DC. But of course, there is much more than that:
Customer GUI panels for users to list, inspect, and perhaps update Customer
records;
A Customer DB schema; and
Pass-by-value objects that contain Customer data.
The list could go on. The important point is that the implementation of the
Customer concept could well require artifacts in each of the four distribution tiers.
A business component or BC is an enterprise component that implements all
and only a given business concept (process or entity) in a system. As such, it also
provides for the business concepts distribution throughout the system, across the
four distribution tiers described previously. It also maps well to business concepts
in a requirements and/or business model, and the benecial implications of this are
discussed in Section 4 of this chapter.
Distribution Aspect
A BC consists of at least one DC, and it must have at least one workspace tier or one
enterprise tier DC. Of course, it may have both, and it sometimes has more than one
DC in a given tier. Each DC fullls the responsibilities of its tier within the BC.
An example of a BC consisting of only an enterprise tier DC is a calculation engine
that it implemented as a DC. For example, in ERP systems, a price calculator, often
very complex, may be produced as a BC whose realization is a single enterprise-tier
DC.
20 For further information, see Sims (2002).
74
A BC is not required to have to have all four tiers. However, its tiers are always
adjacent tiers in the four-tier model. That is, while one BC may have all four tiers
(user, workspace, enterprise, resource), or perhaps only two for example, enterprise
and resource only, it will never consist of tiers that are not adjacent. For example, a
BC consisting of only user and resource tiers would be invalid.
The interfaces of a BC are provided by (delegated to) one of the DCs in the enterprise
tier, and optionally one of the DCs in the workspace tier.
Business Function Aspect

A BC can represent both process and entity business concepts. Thus an application
consists of a number of business components that collaborate to deliver application
functionality across the distribution tiers. In turn, a DC within a BC is either a
process or an entity DC based on what the BC is. It is often useful to divide entity
components into two layers: entity and utility. A utility BC is one that provides
widely-used business services to both entity and process BCs. Examples include
Address Book, Number Generator, and Currency.
A key consideration in component design is to minimize dependencies so that
components can be composed and re-composed (re-used) as freely as possible. A
component architecture should feature specic interaction patterns that minimize
dependencies, such as the mediator pattern. The BC internals and also its external
relationships with other BCs are the obvious place to apply dependency management
patterns.
Service Orientation
A BC provides business-related services through its interfaces not only at the required
server (enterprise tier) level, but also, where advanced user interaction designs are
being applied, at the workspace tier level as well. Enterprise services are provided by
process BCs to BPM components and also to external requestors. Entity services are
provided to process BCs and also to BPM components.
BC Benets
The BC concept does a number of important things, and in particular provides ve
valuable benets:
1. Link with Requirements: It provides a direct link with the business requirements
model which knows nothing about distribution tiers (see Section 4 of this
chapter).
2. Project Management Unit: As a cohesive collection of DCs, the BC is an ideal
unit of project management, and valuable and meaningful metrics can be easily
derived. This makes for much more accurate prediction than is often the case.
75
And when application development is mainly done through outsourcing, the

BC, or an assembly of BCs, is an ideal module to outsource. Otherwise,
outsourcing, while cost-effective, may reduce its usefulness by building yet more
silos that are not interoperable, nor very maintainable through lack of effective
modularization to say nothing of re-use.
3. Lifecycle Continuity: As it moves through its lifecycle, a business component
is realized by different artifacts at different stages. At analysis time, the artifacts
comprising the business component will be UML models, and documents
(specications). Design time will add detail to these artifacts, so that code can
be produced. The code will consist of some DCs, plus other artifacts such as
GUI panels and DB schema denitions. A component architecture will dene
what completeness means at the end of each stage, denes the stages, and shows
how traceability is provided. Thus a BC encapsulates all development artifacts
within it. Its life starts with the business concept and is rened throughout
its development. Indeed, with appropriate support, the BC can actually run
as soon as it is conceived (through a standard administrative interface). This
approach has been called component metamorphosis (Hubert 2002) and is
extremely attractive, especially when combined with a capability for immediate
prototyping in order to gain quick and early feedback from subject matter experts
(see Pawson and Matthews 2002).
4. Domain knowledge: The BC also acts as a focus for domain knowledge within an
application development group. And since the BC extends across the distribution
tiers, there is not only shared domain expertise across tiers, but also within a small
team (say, two to six people) there is some cross-familiarization in the various
technology concerns found across tiers from user interface through transactions
to data base concerns. All of this can be hugely useful in terms of skill growth.
However, it does assume a good virtual platform and a separate organization that
provides the platform.
5. Unit of Ownership: Re-usable software artifacts are valuable enterprise assets, and
must therefore be owned by a manager within the IT organization. However,
to avoid one manager looking after ten thousand individual assets which is not
humanly possible, asset ownership must be structured, such that ownership of a
large asset implies ownership of embedded assets.
The BC is an ideal unit of asset management, since it owns all of the assets that
go into the DCs that make up the BC, and among these there will normally be a
number that are re-usable. Without a larger owning construct, such re-usable assets
are often difcult to categorize, and can quickly become lost. Examples of smaller
assets that are usefully owned by a BC are proxy classes required to access the
DCs in the business component, and also business data types (pass-by-value classes
76
that are used as parameters in operation signatures) that are dened for the provided
interfaces. For further detail, see Sims (2001).
The value of the BC as a unit of asset ownership is enhanced by the development
lifecycle aspect, where a BCs lifecycle must also be managed, from the start of
its development, then into operational use, through subsequent evolution, to its
retirement perhaps many years later.
A full treatment of the richness of the business component concept is beyond the
scope of this chapter, but is presented well in Herzum and Sims (2000). Other authors
also suggest a similar concept. Richard Hubert, for example, talks of a component
having both client and server personalities, and says, These personalities exist to
cleanly encapsulate and denote the two design partitions inevitably required of any
component if it to be distributed. (Hubert 2002, p. 87)
3.2.3. The Application Component (AC)

Business components collaborate to provide a business solution. The assembly of
BCs that together provide such a solution effectively make an application. When
this assembly is deliberately produced to be a component in its own right, with
one or more service-oriented interfaces, then the assembly is called an Application
Component.
We can dene an application component as an assembly of collaborating BCs that
provides a dened set of business services. This concept is similar to EDOCs
Application Component (OMG2 2004). A legacy system wrapped with a serviceoriented interface looks like an AC to its clients.
Figure 8 shows an example of an AC. It consists of ve BCs, each having up to
four tiers, and together providing a simplied Invoice Management service. Business
function layers (process-entity-utility) are shown vertically, and within each BC, the
distribution tiers are shown as a stack of slices. User tiers are gray (blue in e-book
version), and resource tiers are light grey (yellow in e-book version). Interfaces for the
workspace and enterprise tiers are shown. Some of the business components have less
than four tiers, and this is quite normal.
This AC might be called the Invoice Management component.21 As with most ACs,
its interface is dened as being the interface of the topmost process BC (Invoice
21 In any architecture, naming conventions are important. This architecture chooses to call process
components X Managers, application components X Management, and entity components just
X. Hence Invoice Management, Invoice Manager, and Invoice. This does not necessarily mean that
somewhere there is a Customer Manager. If managing a Customer requires no other component, then
77
Figure 8. Application Component Composed of Business Components
Manager in the example); that is, a union of the Invoice Managers workspace tier
and enterprise tier interfaces. The enterprise service interface that is, the interface
that would be published in a UDDI repository is almost always the enterprise tier
interface only. (See Chapter 6 for a more detailed discussion of UDDI.)
In order to show how the tiers within each business component interact, we can
atten the gure out, the result being shown in Figure 9. Each BC is shown as
a gray (blue in e-book version) rectangle. Within each BC are DCs implementing
its various tiers. The DCs are named by their tier, so that, for example, the DC
implementing the enterprise tier of the Customer BC is known as the Customer
Enterprise-Tier Distributed Component or just Customer EDC for short. Other
naming conventions are possible, of course, but this one, although a little acronymheavy at rst glance, has proven simple for developers to use once the concepts have
been introduced.
Black solid arrows show intra- and inter-BC invocations. The assumed component
architecture allows intra-BC invocations between workspace and enterprise tiers;
some architectures disallow this, routing all invocations through the top-level process
BC.
the Customer business component can manage itself. Actually, this is unlikely for customer in most
businesses, since normally at least a credit check is required as part of the add new customer business
process. See Herzum and Sims (2000) for a discussion of naming conventions, and also for a discussion
of styles of entity components.
78
Figure 9. Invoice Management Application Component
Figure 9 does not repeat the process/entity/utility layering, although it is more or less
implicit in the positions of the BC.
Finally, with regard to Figure 9, it is worth saying something about the DCs in the
user tier. In the gure, they each appear to presents a different panel to the user. This
is certainly not the intent. The various panels are normally presented as panes and/or
pop-up windows within a window that presents a tool appearance.
Several views of the essential component model are needed. The diagrams above
show component-centric views. A common additional view is that of the data base
schemas. Another is the set of user workspace domain components, and yet another
is a view of the set of enterprise resource domain elements. Such views are normally
provided through development tools.
Finally, note that the user and workspace tiers in an AC constitute its user-workspace
domain. It is not unusual for an AC to provide for a number of different user
experiences such as browser, PC, PDA, and phone. In this case there will be a separate
user-workspace domain for each experience, although it is quite possible that many
of the same workspace tier DCs will be used in each domain.
79
3.3. Dependency Management

A focus in a component architecture on dependency management is essential.
Without this, component interactions can become highly complicated, so slowing
business evolution as changes take much longer to make and test. Two important
dependency management patterns used in BCs and ACs are:
1. Inter-tier interactions; and
2. Business function layers.
3.3.1. Inter-Tier Interactions

Some of the constraints on inter-tier communication have been noted previously.
Figure 10 illustrates a set of constraints that have been found useful. Essentially each
tier (or more precisely, the DCs in each tier) provides services to the tier on the left.
The BPM tier provides services to both workspace and enterprise tiers, and vice-versa.
That is, BPM components (process or workow specications) can be kicked off by
DCs in either the workspace or the enterprise tier.
Figure 10. Inter-Tier Interactions

Note that Figure 10 could be part of a UML prole. Loosely speaking, a UML
prole is a kind of template, or conceptual model. However, a UML prole not
only denes the concepts to be used; it can also be applied by a modeling tool so
that an application developers model is constrained to use the concepts expressed in
the prole.22 For example, the developer can be constrained to design components
and component interactions that conform to the prole. Tools can then check the
22 One example of such a tool is Objecteering from Softeam (www.softeam.com). The latest version
of this tool provides not only normal UML modeling, but also a prole building tool.
80
correctness of the model as far as proper use of the proled concepts is concerned.
This is an excellent way of delivering architecture to the application developer. When
a prole denes scalability patterns and constraints, then the result is that run-time
implementations whose structure is generated from the design models are likely to be
inherently scalable, and also conform in all respects to the modularization and re-use
standards dened by the prole. This approach is an important aspect of the OMGs
MDA (Model-Driven Architecture) strategy (OMG1 2003).
3.3.2. Business Function Layers

It is useful to separate the orthogonal concerns of distribution tiers and application
layers. A distributed system has a number of logical tiers which are essentially IT-related
responsibilities. A business has a number of logical business-related responsibilities
that we call layers. These responsibilities are, briey, process and entity.
Some architectures conate these two quite different concerns. Thus we see architectures that suggest a user tier, a workspace or session tier, a business service tier
(process), and a business object tier (entity).
However, there is more or less universal agreement now that process components
should be separated from entity components, and that the pattern governing their
collaboration should be the mediator pattern (Gamma et al. 1995). This is a very
useful dependency management pattern, and is an example of a pattern developed
in the context of object-oriented design that does indeed scale. This pattern is well
established, being formalized in EJB (session and entity EJBs) and similarly (although
richer) in the CORBA Component Model.
However, business function layering is richer than merely a separation into two
parts. Business function is more a spectrum, from process to entity. Of course, this
spectrum applies not only to business components (and hence to DCs) but also to the
BPM domain, where the separations are sometimes considered to be continuous, and
sometimes discrete. However, here we focus on the three component granularities
described previously, based on the architecture described in Herzum and Sims (2000).
This rstly separates the layers into three categories of process, entity, and utility. It
appears that, by industry sector, these can usefully be further subdivided.
Within an application component, the process-utility layers play an important part
in the architectural separation of concerns, and contribute signicantly to reducing
complexity through dependency management through application of the mediator
pattern. This is done by having process BCs mediate (or choreograph) entity
BCs. Utility components, however, provide services both to entity and to process
components.
81
Figure 8 illustrates this pattern, with the process BC at the top invoking entity
components in lower layers. A formal model of the constraints inherent in
this approach would show a directed acyclic graph of invocation dependencies.
Essentially this says that Process Components use Entity Components and/or Utility
Components. Entity Components can use Utility Components, and nally Process
Components can invoke other Process Components when the process layer is further
sub-divided. BPM components operate at a higher layer than the process BC, which
is how application components are federated.
An Entity component is responsible for managing a specic set of data, and can
be thought of as owning that data. It provides entity services to other (typically
process) components, and protects the integrity of the data as far as enforcement of
business rules directly to do with that data is concerned (for example, validation).
A signicant advantage of entities owning their data is the reduction of dependencies
between data access mechanisms and schema on one hand, and business logic
on the other. Often a traditional application that needs to access, say, customer
data, will do so directly using SQL. This creates a web of dependencies between
the database and multiple applications. The entity component approach helps to
address DB schema dependency management. In the enterprise tier, entity DCs
effectively provide a service-oriented component structure to otherwise monolithic
process-oriented services.
A utility BC is a kind of entity that is of widespread use across many business
functions. An example could be an address list containing addresses for customers,
suppliers, business partners, and employees. Another example might be a currency
book that handles currency conversions, currency arithmetic, and provides exchange
rate information.23
4. From Requirements to Design

An important aspect of ESOA is that the services provided, and the components that
implement them, should map as directly as possible to the needs of the business.
Then, when the business evolves, it is very quickly clear what parts of the IT
system are affected and which new or changed services and components should be
provided. However, current methodologies often do not provide this (although some
23 An example of this is the FbcCurrency Module in the OMG Currency Specication (OMG
document formal/00-06-29).
82
go part-way towards it), nor do they help with structuring already-dened services.
The resulting impedance mismatch can cause loss of information and traceability.
Without a solution to this problem, it is hard to see how any software initiative,
by itself, can hope to make business more agile, responsive, or IT become more
productive so that it can meet time-to-market, exibility, and service-orientation
goals. Failure in this area means that much of what IT delivers is likely either to not
do what the business needs IT to do, or to result in service silos. Either way, the
result is loss of agility.
So we need to be able to specify business needs with minimal information loss
while crossing the bridge between business rules and structures to IT algorithms
and modules. Many current requirements analysis methods provide mapping of the
business rules to IT algorithms. However, those IT algorithms usually end up buried
in code modules that bear little resemblance to business structures. Real and effective
traceability becomes very difcult; the bridge weakens, sometimes to the point of
collapse. But such traceability is particularly important for ESOA, where proper
identication of the components that provide the services is the key to agile IT.
The approach described in this section, called the Business Element Approach,24
provides for changes to business processes or rules being quickly and unambiguously
be traced to one or more specic components. In addition, new business structures
and algorithms are analyzed such that the structures dened in the requirements
activity become the start of the IT design model.
We start by suggesting a particular approach to requirements. Then we show how
the resulting business-oriented model can be directly implemented by the kinds of
software component discussed previously.
4.1. Requirements
The objective of the requirements activity25 is to dene precisely what services the
planned system should provide, how they are carried out, and what business features
it should exhibit. The requirements activity ends when all relevant business questions
have been answered. Therefore, it is not some high-level scoping activity (although
24 This term is due to Taylor (1995), and his approach is further described in Hubert (2002).
Extension of this work was done within the COMBINE project (Combine 2003), and also after that
project completed.
25 We use the generic term requirements as the name of the activity that denes precisely what the
business wants and needs of IT. This activity goes by many names in the industry, including business
modeling, requirements, requirements analysis, scoping and denition, etc.
83
that is part of it); rather it is a signicant part of the system development effort, and
goes into substantial business detail. However, it does not include IT system design.
The main elements of the process are as follows:
Dene scope, vision, and goals;
Dene the business requirements in terms of business processes and resources
used or created by those processes (resources are sometimes known as business
objects, or entities);
Dene the business elements that the IT system will implement; and
Dene user interaction requirements.
It should not be assumed that this list implies a sequence. More importantly, following
description of business elements should certainly not be interpreted as implying a
waterfall-style process. Such is certainly not the case.
4.1.1. Business Elements

A business element is a process, a resource, or an organizational unit. These three are
inter-dependent: organizations perform specied processes; a process requires both
an organization to perform it and resources as input or things produced; resources are
assets of an organization, and are used or produced by processes. For current purposes
we focus only on process and resource BEs.
Informally, we can say that a Business Element (BE) is an identiable chunk of the
business that is seen as important by business people.26 A BE can be process-oriented
or resource-oriented. Examples are Sales Order Service, Invoice Management,
Customer, Booking, and Sales Order.
But how are these BEs identied? With experience, many BEs rather jump out at one,
and indeed some candidate BEs can often be suggested at the start of requirements
based on experience in other organizations in the same industry. However, there is a
set of heuristics that enable candidates to be identied based on process and resource
models.
4.1.2. Processes and Resources

The Business Element approach requires that we dene clearly:
26 A precise denition of what important means in this context is beyond the scope of this book.
However, Tyndale-Biscoe et al. (2002) provides a useful exposition.
84
Business processes, each of which has one or more steps, and where each step
is, where appropriate, further rened as a process in its own right. A process
provides a service (possibly an internal service), and the steps dene how the
service is provided. A bottom-level process is often a set of steps that constitute
what might be called a procedure or algorithm.27
Resources28 used by or produced by steps, and which provide limited-scope
services to processes.
Processes may be captured in text within use case models, or more formally by
UML classes or action states. We can identify a number of different kinds of process,
differentiated by business-related factors. Of relevance here is an immediate process.
An immediate process is one that is required to complete as soon as possible, and
whose intermediate states are of no concern to the business in that they are not
required to be remembered after the process has completed. An immediate process
is performed autonomously, with no intervention from a human. It typically denes
a service that would be provided by the core enterprise system (see Section 1 of the
previous chapter). Other kinds of processes would map to BPM specications.
Resources are captured in a business object model or an entity model. Examples
are Customer, Balance, Address, and Order Line. Given models of immediate
processes, and resources, we can dene the set of business elements that are within
the scope of a given project.
4.2. Business Element Analysis

Three different kinds of business element can be derived from resource and process
models:
1. Resource Business Element (RBE) the important resources in the business. An
RBE is a candidate for implementation in the eventual IT system as a Business
Component.
2. Service Business Element (SBE) a set of related services provided by the business
and implemented by a business process. An SBE is a candidate for implementation
in the eventual IT system as a process-style programmed Distributed Component.
27 There may be a number of kinds of resources, such as entities, performers, and actors. Here we are
concerned only with those resources that are entities.
28 Some business modelers prefer not to analyze down to this level while in a business requirements
activity. Others prefer to do so, arguing that business modeling only stops when, within the scope
selected, there are now more questions about the business.
85
3. Delivery Business Element (DBE) a dened grouping of service and resource

business elements that provides a coherent set of services related to an organizational unit. A DBE is a candidate for implementation in the eventual IT
system as a Distributed Component (implemented using such technologies as
EJB, CORBA, .NET, etc.).
Note that, although the word element implies a single thing, a business element is
in fact a group of things.
4.2.1. Resource Business Element (RBE)

A Resource Business Element (RBE) encapsulates a group of resources. The group is
focused around a particular resource that is real and independent in the business
domain. Other resources are auxiliary, inasmuch as they support the primary
resource in some way (for example, a Customer resource may be supported by a
Customer Balance resource).
Real: A real resource is one that is both used and understood by subject matter experts
(SMEs). It is not abstract. For example, in a manufacturing business, customer,
address, and invoice line item would probably be real, while legal entity,
location, and collection member might not. That is, an SME would assert
that while a customer is a common everyday concrete thing, a legal entity is
not (although there might be agreement that, hypothetically, it would be a good
super-type of customer).29
Independent: An independent resource is one that can be talked about by SMEs
without rst saying to what it belongs. That is, its scope is implied and understood,
and is probably that of the business or of an important organizational unit within
the business. For example, in a manufacturing business, a customer probably does
not have to be qualied, so it is independent. Address or Balance, on the other
hand, would have to be qualied to be meaningful, for example, customer address
or supplier address. Of course, this is context-dependent; for example, addresses are
often kept separate from the resources they relate to, especially in distribution-oriented
organizations.
Looking at the business resource model in this way, certain resources can be identied
as real and independent, and are called focus resources, while the others are called
auxiliary resources. These terms follow UML semantics (see [UML]).
29 For a full discussion of this topic, see under Trading Partners in Herzum and Sims (2000,
p. 463).
86
Figure 11. Sample Resource Model Showing Focus Resources (Fragment)

Identication
The following heuristic can be used to identify resource business elements (RBEs):
1. Consider each resource in the business resource model, and ask whether it is
real and independent. If it is, then it is the focus resource for a RBE. Other
resources are auxiliary resources.30
2. Navigate the relationships (including upward generalizations) from the focus
resource, to and through auxiliary resources, stopping when another focus
resource is reached. Note the auxiliary resources encountered. Those auxiliary
resources, plus the focus resource, form the rst-cut denition of a group called
the focus group. Note that this is not an entirely automatic process; some
judgment on where to stop is sometimes needed.
3. When all RBEs have been identied, create an RBE relationships class diagram.
This is essential for comprehension of relationships between RBEs.
Example
The RBE example starts from a business resource model (or business object model).
Figure 11 shows a highly simplied sample resource (or business) model. Let us
assume that Employee, Customer, and Order are identied as the focus resources.
Following the identication heuristic to group auxiliary resources results in the three
focus groups illustrated in Figure 12.
30 The terms focus and auxiliary are UML stereotypes, and have the same semantics here.
87
Figure 12. Focus Groups
Figure 13. Resource Business Elements
When these are disentangled, three distinct resource business elements emerge, as
shown in Figure 13. The result is a signicant simplication of the original resource
model, and no information is lost. Indeed, some has been gained, since the resource
model can now be shown in its RBE view, which presents only the important
resources. One of the interesting simplications is that the various relationships have
also been grouped, leaving only one inter-RBE relationship. Finally, note that some
88
auxiliary resources appear in more than one RBE. This is not surprising. However, a
given focus resource appears in only one RBE.
Before leaving subject of RBEs, it can be seen that each RBE is represented in Figure 13
as a UML2 component. This illustrates well one of the intents of UML2 components
that they should apply throughout the development lifecycle including the
business model, or Computation-Independent Model (CIM) in the OMGs MDA
parlance. In this case, we can use the components concept of type and also it is
concept of realization. Some business modelers may think this use of the UML2
component concept is wrong that a business model should not show the least
indication of something that may be misinterpreted as eventual implementation. In
that case, a suitable model element for an RBE could be a UML package.
4.2.2. The Service Business Element (SBE)

A service business element is a collection of immediate steps as discussed previously
in Section 4.1. The reason for grouping the steps (and their sub-steps) is that a given
organizational unit is often responsible for the set of processes (steps) surrounding
a Resource BE. These steps are often closely related in terms of the kinds of things
they do, and together provide a coherent service. In effect, we apply the tested
modularization principles of high cohesion and low coupling to these chunks of
the business.
Identication Heuristic
To identify Service Business Elements (SBEs):
1. Identify the highest-level immediate steps in the business process model. These
are services provided by the core business systems. The name of each step is
probably of the form verb-noun. Many of the nouns will probably be the
names of RBEs, for example Supplier, Shipment, Contract, Schedule, Invoice,
or Product.
2. Group the steps by RBE. It is likely that each group will include the CRUD31
lifecycle of the RBE.
Many groups will involve more than one resource BE. For example, a Sales Order
group could include steps that not only direct the lifecycle of a Sales Order in
some way, but also update Inventory and Customer Balance. However, the focus
of the group is often a specic resource (such as Sales Order).
Some groups may involve only one RBE. In this case, it is possible that the step
forms part of the responsibilities of the RBE rather than the SBE.
31 CRUD: Create, Read, Update, Delete.
89
3. Iterate on the next level of immediate steps in the business process model. If no
further SBEs are found, then the immediate steps at this level of iteration are
candidates for the business logic within the SBE, or possibly even within an RBE
(such as validate data provided). If there are no lower-level immediate steps
found, then stop.
Each SBE will normally map to an organizational unit, and is a candidate for
implementation as a process component. The various verbs in the names of the
immediate steps are candidate operations in the interfaces of that component.
Subsidiary immediate steps associated with each higher-level immediate step typically
constitute the process that provides the service indicated by the higher-level immediate
step.
This identication heuristic groups immediate steps such that the groups service
business elements make sense in business terms, and not only provide the basis
for a service-oriented enterprise system, but also exhibit best practice modularization
along high cohesion and low coupling principles.
SBE Identication Example

Suppose the following services (top-level immediate steps) were dened in the process
model:
Amend Customer Record;
Handle Order (place a new Order);
Remove Employee;
Record New Customer;
Amend Existing Order;
Cancel Order;
Hire New Employee.
Many of these steps will have subsidiary immediate steps. For example, the Amend
Customer Record and Handle Order steps could well have the following subsidiary
steps:
90

Service
Subsidiary Immediate Steps
Amend Customer Record
Validate customer details provided

Review credit limit by running a Credit Check
Record Customer details
Check for relationships with other Customers,
update where necessary
Send the standard your details changed letter
Validate data submitted
Check Customer is valid
Calculate value of Order
Check for credit
Allocate inventory
Create Back-Orders where necessary
Create the Order
Send an Order Acknowledgment
Handle Order
Based on the example above of seven services (top-level immediate steps), the
identication and grouping heuristic could produce the following (incomplete)
SBEs:
SBE
Service
Subsidiary Immediate Steps
Customer Service
Amend Customer Record
Validate customer
details provided
Review credit limit
Record Customer details
Check for relationships
Send the standard letter
Validate data submitted
Order Service
Record New Customer

Handle Order
(place a new Order)
Employee Service
Amend Existing Order

Cancel Order
Hire New Employee
Remove Employee
Check Customer is valid

Calculate value of Order
Check for credit
Allocate inventory
Create Back-Orders
where necessary
Create the Order
Send an Order
Acknowledgment
91
As can be seen, each top-level immediate step provides a service (e.g. Amend
Customer record). Each service in turn will usually consist of a set of subsidiary
immediate steps.
4.2.3. Delivery Business Element (DBE)

Specic subsets of Service and Resource BEs are used by an organizational unit in
the business to deliver some required capability. For example, an Order Service SBE
plus several RBEs such as Customer, Product, and Sales Order are used to deliver
order management capabilities to a Sale Order Processing department in the business.
Such specic groups can be seen as providing dened services to other parts of the
organization, as well as to customers and suppliers. The grouping itself usefully forms
a third kind of business element, which we call a Delivery BE, since it is this group
that delivers functionality for the business. Consider: an RBE can do nothing by
itself (no process), and likewise, an SBE by itself is pretty useless (no resources!). The
Delivery BE is the set of assets that delivers value.
A Delivery Business Element (DBE) is a grouping of Service and Resource Business
Elements that together deliver a business solution to a business problem, and which
provides services to requestors.
A given DBE often corresponds to a major responsibility of a department or larger
organizational unit in the business. For example, the services and resources involved
in a Sales Order Management DBE would reect the major responsibilities of the
Sales Processing Department in a business.
A simplied example of a DBE is shown in Figure 14.
Identication
Delivery Business Elements can be identied by nding the top-level SBEs. A toplevel SBE is one that is not a dependant of any other SBE. For each top-level SBE,
identify the set of BEs that this SBE needs to function correctly. That is, identify
the dependent BEs of the top-level SBE. The resulting set, plus the top-level SBE,
comprises the Delivery Business Element.
Hint: In small-to-medium projects, there may only be top level SBEs. In very small
projects, there may indeed be only one, and hence the DBE will correspond to an
application.
92
A BE diagram can be constructed as BEs are identied. Applying the mediator pattern
for usages by one BE of other BEs typically results in a directed acyclic graph (as
illustrated in Figure 14). The DBE will normally be a subset of this diagram.
Example
Figure 14 illustrates the dependencies between two SBEs (Order Service and Pricing)
and three Resource Business Elements. This set of business elements together
comprises the Order Management delivery business element. Note that the UML2
component is used in the diagram to represent business elements. Only the type
part of the component is used interfaces and implementation detail are normally
added later during the IT design phase. This illustrates well a realization of the
UML2 objectives for the component concept that it should apply throughout the
development lifecycle including the requirements phase.
Figure 14. Delivery Business Element
4.3. Mapping to Components

A BE captures an important business concept. CBSE denes a set of structuring
concepts that are ideal for implementation of BEs in an IT system. This is because
little or no transformation is required between the BEs and the technical design
model. That is, each BE is mapped one-to-one to a candidate component in the
design model. This means that traceability between components in the IT system
and the business is direct and visible to business people. Each business element maps
isomorphically to a specic kind of component, as follows:
93
Service BE to Process Business Component: Each immediate step provided

by a service BE is mapped to an operation on a process business component in
the IT design. Thus immediate steps become services provided by the system
(for example, Modify Customer record).
Resource BE to Entity Business Component: Again, the mapping is direct:
each resource BE can be (largely automatically) transformed into the beginnings
of an entity business component, complete with focus and auxiliary classes. As
with the process component, the transformation has no information loss.
Delivery BE to Application Component: Earlier in this chapter we showed
how process and entity components collaborate, and a simplied example of a
sales order management application component was given. This collaboration
maps extremely well to a Delivery Business Element, and the transformation is
exceptionally simple. Dependencies between BEs map to operation invocations
between components.
In MDA terms, the business elements comprise part of a CIM (ComputationIndependent Model), and can be (largely automatically) transformed into the
beginnings of a component PIM (Platform-Independent Model). Because no
structural changes are made, the transformation is simple, and has no information
loss.
The component model is then rened, behavior added, etc., to move through the
development lifecycle to an operational IT system.
In other words, we have a clear and unambiguous bridge between the business and
the IT system, with one-to-one mapping.
This, then, is the essence of the bridge between business requirements and the IT
system. Extra-functional requirements (such as scalability and security) are typically
handled either through the architectural features of the component model (e.g. the
legitimate interactions between tiers) or as part of the virtual platform.
In this way, traceability between components in the IT system and the business is
direct and visible to business people; and reverse traceability is similarly isomorphic
and visible. Just as SBEs choreograph the use of RBEs and sometimes other SBEs, so
process components choreograph the usage of other components. Each component
provides clear services; for example entity components provide entity services to
process components, which in turn provide published enterprise services both to
business users and to requestors outside the business. Because each component is
designed and built autonomously, capturing autonomous business elements, then just
as business elements in the business can be re-used, so can the components. There
94
is as much re-use of components as there is of business elements. Further technical

re-use is also possible within the IT domain. For example, auxiliary classes within
components can often be re-used in other components.
5. Summary
In summary, process business elements can be said to be the modules of the business.
Mature CBSE gives us a technology whereby business elements can be captured on
a one-to-one basis. This means that the IT system that supports the business is
structured along the same lines as the business.
Now suppose we provide systemic event management in the IT system, so that each
component can publish business events as they happen. It then becomes possible to
think of installing probes into the message stream which can not only measure the
ow rate between given components, but also can interrogate the message content
itself (remember that the message context is XML).
Ability to probe messages for their semantic content? This starts to sound like a kind
of business nervous system! And indeed, some enterprises today are beginning to
think along these terms. Why? Because the concept of a business control room for
real-time management of organizational units is one that has been around since the
1970s (e.g., Beer 1979). Then, it was not only the technology that was lacking, it
was also the set of software structuring concepts. Today, enterprise service-oriented
architecture can provide both the technology and the software structuring concepts.
Hence ESOA may well provide the basis for a situation where the business models,
or CIMs in MDA-speak, may be become a live model of the business, in the same
way that a railroad control center provides a live model of a railway system. And
the design models or MDA PIMs (platform-independent models) could become
the main focus and tool for synchronized business and IT agility and evolution.
The naked objects initiative (Pawson and Matthews 2002) provides the basis for
technology whereby even partly-built components can be automatically visualized on
a GUI. But here we stop: further extrapolation and exploration is beyond the scope
of this chapter.
3
ORCHESTRATION
None of us is as smart as all of us.

Anonymous
Even before the advent of Web services, an increasingly large number of distributed
applications were constructed by composing them out of existing applications.
Enterprise Application Integration (EAI) techniques grew up from the realization
that no one infrastructural technology (e.g., CORBA or DCOM) will ever be
adopted by all of the software industry. Furthermore, although sourcing a solution to
a problem (large or small) from a single vendor is possible in the short term, in the
long term it is often the case that a corporate intranet will be running systems from a
variety of vendors, not all of which will be able to interoperate. Large multi-national
corporations often evolve through acquisitions of smaller companies who may have
different infrastructural investments. We have often heard the statement that Its
easier to interoperate with a different company than to talk to different divisions
within the same company. Therefore it should come as no surprise to learn that
large-scale applications are rarely built from scratch; rather they are constructed by
composing them out of existing applications.
Providing solutions that enable disparate (heterogeneous) technologies and applications to communicate is extremely important. Without them, a companys
infrastructure would either not be able to grow (leading to islands of isolation) or
would be at the mercy of a single vendor. For several years EAI solutions have made
it possible to compose an application out of component applications in a uniform
manner, irrespective of the languages in which the component applications have been
written and the operating systems of the host platforms. Unfortunately, most EAI
95
96
platforms offer solutions that are not interoperable with one another. Web services
offer a potential solution to this important drawback.
The resulting applications can be very complex in structure, containing many
temporal and dataow dependencies between their constituent applications. An
additional complication is that the execution of such an application may take a long
time to complete and may contain long periods of inactivity (minutes, hours, days,
weeks, etc.), often due to the constituent applications requiring user interactions.
In a distributed environment, it is inevitable that long running applications will
require support for fault-tolerance and dynamic reconguration: machines may fail,
services may be moved or withdrawn and application requirements may change.
In such an environment it is essential that the structure of applications can be
modied to reect these changes. In general, composite applications are increasing in
importance as companies combine off-the-shelf and homegrown Web services into
new applications. Various mechanisms are being proposed and delivered to market
daily to help improve this process. New fourth generation language development
tools are emerging that are specically designed to stitch together Web services from
any source, regardless of the underlying implementation.
A large number of vendors are starting to sell business process management, workow
and orchestration tools for use in combining Web services into automatic business
process execution ows. In addition, a growing number of businesses nd themselves
creating new applications by combining their own Web services with Web services
available from the Internet supplied by the likes of Amazon.com and Google.com.
These types of composite applications represent a variety of requirements, from
needing a simple way to share persistent data to the ability to manage recovery
scenarios that include various types of transactional software. Composite applications
therefore represent a signicant challenge for Web services standards since they are
intended to handle complex, potentially long-running interactions among multiple
Web services as well as simple and short-lived interactions.
Workow systems have been around for many years, pre-dating Web Services
and SOA. The core concepts behind the notion of workow are not tied to any
specic implementation environment. Therefore, in the following sections we will
examine workows from an architectural perspective before describing the Web
Services Business Execution Language (WS-BPEL) (WSBPEL), which is the current
contender for the title of workow standard for SOA.
Chapter 3: Orchestration
97
1. Workow and Business Process

Management
A business relationship is any distributed state maintained by two or more parties,
which is subject to some contractual constraints previously agreed by those parties. A
business transaction can therefore be considered as a consistent change in the state
of a business relationship between parties. Each party participating in a business
transaction holds its own application state corresponding to the business relationship
with other parties involved in that transaction. During the course of a business
transaction, this state may change.
Traditional methods for integration of these business transactions typically involve
embedded logic inside of functionality-oriented IT applications. The development,
testing and deployment effort required to change these applications make integration
and process changes very costly and complex. To address these issues, proprietary EAI
and Business Process Management (BPM) products emerged to abstract integration
and process automation. These software products (workow systems) liberated the
integration and process tasks from the underlying functional IT applications so they
could be more effectively changed, managed and optimized.
Workows are rule-based management software that direct, coordinate and monitor
execution of tasks arranged to form workow applications representing business
processes. Tasks (activities) are application specic units of work that may involve
multiple services or components. A Workow schema (workow script) is used to
explicitly represent the dependency between the tasks and in many ways may
look like a traditional programming language, with branch statements, conditional
executions, etc. The structure of many workow management systems is based on
the Workow Reference Model developed by the Workow Management Coalition
(WfMC). Figure 1 depicts the Reference Model.
This model provides for the manipulation and execution of workow instances
(interfaces 2, 3 and 5) as well as for the denition and management of workow
schemas (interfaces 1 and 5). According to the base workow model, a workow
application is modeled as a collection of tasks. A task is the unit of activity within a
workow application; typically you can equate a task with an interaction on a service
or services; for example, booking a ight on an airline.
The structure of the workow application is expressed by the interdependencies
between its constituent tasks. A dependency could be just a notication (temporal)
dependency (shown by a dotted line in Figure 2, indicating that t2 can start only
98
Figure 1. WfMC Workow Reference Model
Figure 2. Inter-Dependency Tasks
after t1 has terminated) or a dataow dependency (shown by a solid line, indicating

that, say t3, needs to be notied of the availability of input data from t1).
A task is typically modeled as a having a set of inputs and a set of outputs. The
execution of a task is triggered by the availability of an input(s) message. A task can
terminate producing one of a set of output messages. The outputs will, if required,
99
be propagated to other tasks as sources of input. In addition, a task can be composed

from other tasks to form a compound task.
A workow system allows the specication of task dependencies and their interrelationships and then controls the execution of that workow specication. Typically
this specication will be maintained in a persistent manner such that failures of the
workow system or individual tasks will be recoverable to ensure that the required
specication is executed to completion.
For example, let us consider a workow application that involves processing a
customers order for a book from an online shop. It may be modeled as a compound
task processOrderApplication which contains four constituent simple task instances:
paymentAuthorization, checkStock, dispatch and paymentCapture. The relationship
between the tasks is shown in Figure 3.
Figure 3. Process Order Application Example

To process an order, paymentAuthorization and checkStock tasks are executed concurrently. If both complete successfully then dispatch task is started and if that
task is successful then the paymentCapture task is started. Obviously you may want
to structure tasks slightly differently in some situations, e.g., run paymentCapture
before dispatch, or run them concurrently. However, for our illustrative purposes this
structuring is sufcient.
Note that although to dene these business constraints (interactions) requires precise
knowledge of the interaction protocols involved, the publicly visible exchange pattern
of each of the parties involved, it does not require internal implementation specic
details of the various parties. For example, if we return to the process order example, it
100
is not necessary to understand how the checkStock or dispatch tasks are actually
implemented: as a user of their services all that is required is that they conform to
some pre-agreed contract.
This is an important point and one which is exemplied by the overall Web services
model. Web services are specically about fostering systems interoperability. What
makes Web services so interesting is the fact that the architecture is deliberately
not prescriptive about what happens behind service endpoints Web services are
ultimately only concerned with the transfer of structured data between parties, plus
any meta-level information to safeguard such transfers (e.g., by encrypting or digitally
signing messages). As we have already seen, this gives exibility of implementation,
allowing systems to adapt to changes in requirements, technology etc. without directly
affecting users. Furthermore, most businesses will not want to expose their back-end
implementation decisions and strategies to users for a variety of reasons.
Some workow implementations allow the internal structure of a compound task
to be modied (sometimes dynamically, as the workow executes) without affecting
the tasks which supply it with inputs or use it for inputs. In this case it would be
possible to change the payment and stock management policies, for example, causing
payment capture even if the item is not presently in stock, or the addition of a task
which could check the stock levels of the suppliers of the company and arrange direct
dispatch from them.
1.1. Intra-Enterprise Workows

As you can imagine, workow systems are extremely useful in controlling the
execution of processes that form an application, both within a single domain, e.g., a
corporate intranet as well as between domains, e.g., business-to-business interactions.
At present it is true to say that the majority of workow systems are typically being
used to glue together services and components within a single domain, but that is
likely to change. Many examples of this exist today, including the purchase order
scenario we have already considered. For example, some large software organizations
model their issue-tracking systems on what is often referred to as the trouble ticket
scenario: a bug or problem is identied; it must be recorded; the record must be
checked for accuracy; from a single instance of a problem, the underlying cause
is identied; a resolution is identied, which must be communicated back to the
original party with the problem.
Trying to model either the purchase order or trouble ticket scenarios manually is
obviously possible, since these kinds of applications existed before there was anything
identiable as a workow system. However, it requires a lot of skill and effort on behalf
101
of the application programmer, especially when issues such as reliability, fault tolerance
and adaptability are concerned. For example, at some point during its execution a
long running application is likely to encounter changes to the environment within
which it is executing. These environmental changes could include machine failures,
services being moved or withdrawn, or even the applications functional requirements
being changed. Trying to implement this for a bespoke application is hard enough
without sacricing exibility. Luckily, some of the more advanced workow systems
provide mechanisms that will allow workow applications to change their internal
structures to ensure forward progress can be made.
1.2. Interoperability Concerns

Whilst workow systems and standards have been around for many years, there has
remained a signicant problem with these systems: interoperability. Despite the fact
that a standard for non-Web services workow systems exists (the WfMC standard),
it is written in such a way that many different, un-interoperable implementations
are possible, which can still claim to be compliant with the standard. The problems
this can cause are obvious and difcult to resolve without help from the individual
vendors, whose initial responses are likely to be to try to persuade you to move
to a homogeneous system. For example, we know of at least one major aircraft
manufacturing company that in the late 1990s had seven different workow systems
that were compliant with the WfMC standard and yet could not talk to one another!
With the advent of Web services, the hope is that this will change.
In the following sections we will look at the efforts going on in the world of
Web Services to provide a standard for workow denition and interactions that is
intended to lower the entry barrier for implementers as well as users.
2. The Business Process Execution

Language (BPEL)
Standard interaction protocols such as XML and SOAP are all well and good, but the
full power of Web services as an integration platform can only be realized when there is
a standard process integration model. In July 2002, BEA, IBM, and Microsoft released
a trio of specications designed to support business transactions over Web services.
These specications, BPEL4WS (BSPEL), WS-Transaction (WSAA, WSBA) and
WS-Coordination (WSC), together form the basis for reliably choreographing Web
102
services-based applications, providing business process management, transactional

integrity and generic coordination facilities respectively. We will look at WSTransaction and WS-Coordination in Chapter 7, so do not worry if this chapter is
rather brief on those specications.
The value of BPEL4WS is that the orchestration and renement of the individual
processes that go to make up a business application is critical to an enterprises
viability in the marketplace. Those businesses whose processes are agile and exible
will be able to adapt rapidly to and exploit new market conditions. In the rest of
this paper we will talk about BPEL4WS (Business Process Execution Language for
Web services). It is important to know that in May 2003, IBM, Microsoft, BEA and
partners submitted a version 1.1 of BPEL4WS to OASIS under the WS-BPEL (Web
Services Business Process Execution Language) Technical Committee, and it is this
version we will concentrate on. However, it is worth remembering that as we have
already seen, a standard does not necessarily imply interoperability.
Note that it is beyond the scope of this chapter to be able to present a tutorial of BPEL
(which is still evolving through the standards process anyway). There are several good
books and tutorials accompanying BPEL software that can dedicate much more space
to this subject than we can in this chapter. However, it is our intention to give a sound
architectural grounding in the concepts and techniques involved in understanding
and using the language. As such we will cover the fundamental aspects of the BPEL
language from a requirements perspective, but will have to omit some of the syntax
elements and capabilities.
BPEL is an example of a workow scripting language for specifying business process
behavior based on Web services. The language can be used to formally dene the
behavior of business processes and their interactions. Processes in BPEL export
and import functionality exclusively using Web services and the language is entirely
XML-based. As we will see, because of the cross-industry support, BPEL looks like
providing the standard integration model and language for BPM. It has the potential
to commoditize the capabilities provided by the old workow and proprietary EAI
and BPM solutions. This is extremely important because it should allow integration
tasks to leverage existing (legacy) investments in EAI and provide an integration path
between different vendor implementations.
Unfortunately, although one of the original goals of the initial WS-BPEL specication
was to foster interoperability between different implementations and to provide for
fault-tolerant tasks through the use of transactions, during its progression through
the OASIS standardization process these aspects have been removed. It is likely that
true interoperability between heterogeneous BPEL implementations based on the
nal OASIS BPEL 1.0 specication will be difcult to achieve.
103
2.1. Relationship to XPath

The default language for expressions in BPEL is XPath 1.0 [XPATH], although
the specication allows other query languages to be used. This means that basic
computations constrained by XPath can be performed as part of an activity. BPEL
supports the following distinct kinds of expressions:
Boolean valued expressions where the evaluation results in a true or false result.
Such expressions are used to manage process control ow.
Deadline valued expressions where the evaluation results in a string which is
compatible with the format of either XML Schema date or dateTime types.
Such expressions are used in timed wait and pick activities with a xed deadline,
as we will see later.
Duration valued expressions where the evaluation results in a string which is
compatible with the format XML Schema duration. These expressions are
used in timed wait situations with a relative deadline.
General expressions where any of the XPath types (string, number, or Boolean)
can be the result. Such expressions are used for assignment and may use different
operators (e.g., <=, +, *, etc.) depending on the result type.
In addition, BPEL also denes a number of other extensions in the BPEL namespace
http://schema.xmlsoap.org/ws/2003/03/business-process/. The bpws: is associated
with this namespace and when it is used in the rest of this chapter you should remember
this.
2.2. Variables
During the course of a business process it is likely that application data will have to
be updated or inspected. BPEL provides the variable construct for this purpose.
A BPEL variable is a typed data structure which stores messages associated with a
workow instance. As with any workow, the state of the application is a function of
the messages that have been exchanged and variables are used to maintain this state.
Variables provide the means for holding messages that state of the business process.
The messages held may be those that have been received from partners or are to be
sent to partners. In addition, they can also hold data that are needed for holding
state related to the process but are not exchanged with partners. Variables begin in
an un-initialized state and are populated over time by the arrival of messages, or
computations being executed which populate them.
104
For example, let us assume that the checkStock process returns a message containing
the number of the required items still in the warehouse. Code 1 shows how a BPEL
variable may be created based on this message type.
<wsdl:types>
<xsd:schema>
<xsd:simpleType name="stockCountType">
<xsd:restriction base="xsd:int"/>
</xsd:simpleType>
</xsd:schema>
</wsdl:types>
<variables>
<variable name="itemsAvailable"
type="props:stockCountType"/>
</variables>
Code 1. An example of a BPEL variable
As we mentioned earlier, BPEL denes some XPath extension functions. One of those
functions is shown in Code 2 and is used to get a variables value:
bpws:getVariableProperty(variableName, propertyName)
Code 2. The XPath function to return a property value from a message
The rst parameter is the name of a source variable (message type) and the second
is a qualied name (QName) of a property to select from within that variable. If the
property does not appear in the message, then it is implementation dependant as to
what is returned. Otherwise, a node set containing the single node representing the
property is returned.
For example, let us return to the checkStock process that returns a message indicating
the amount of stock left. Our workow process may prioritize clients and only fulll
a purchase request for a specic low-priority client if it does not drop the stock below
a specic level, in case a high-priority client request comes in before the stock can
be replenished. Code 3 illustrates how this could be modeled: the source variable
stockResult is the message returned by executing the checkStock process and part
of this message has an element level that contains the current amount of the request
items remaining in stock.
105
<case condition="bpws:getVariableProperty(stockResult,level)
> 100">
<flow>

</flow>
</case>
Code 3. An example of using getVariableProperty
We have looked at some of the basics of BPEL, such as variables and the ability to
dene expressions. In the next section we will build on these to show how relationships
between business processes can be dened and interactions between those processes
controlled.
2.3. Dening Business Relationships

Because BPEL is a ow denition language it has to provide a means of capturing
enterprise interdependencies with various roles known as partner link, partner
link types and business partners. The partner link type is used to characterize the
conversational relationship between two services, by dening the type of role each
service plays in a specic interaction. For example, the purchaseOrderProducer
and purchaseOrderConsumer roles could be represented as shown in Code 4:
<partnerLinkType name="OrderProducerConsumerLink"
xmlns="http://schemas.xmlsoap.org/ws/2003/05/partner-link/">
<role name="Consumer">
<portType name="consume:ConsumerPortType"/>
</role>
<role name="Producer">
<portType name="produce:ProducerPortType"/>
</role>
</partnerLinkType>
Code 4. The purchase order Producer and Consumer PartnerLinkType
As you can see, each role species exactly one portType. Importantly, the partner
link type can either be dened independently of the services WSDL, or may appear
within the WSDL document. Although it is more usual to nd a partner link type
with two roles, as above, there are situations where a single role may be appropriate
and this is allowed for within the BPEL syntax. For example, a service that wishes
to be able to interact with any other service without knowing the type of the service
106
a priori (a general logging service may fall into this category if it is implemented to
take log information from any service in a corporate network).
The partner links are used to model the actual service with which a business
process interacts. It is important to understand the distinction between partner
links and partner link types: more than one partner link can be characterized by
the same partnerLinkType. If we go back to the process order application, it is
possible that the checkStock task may use more than one supplier, but the same
partnerLinkType. The syntax for a partnerLink is shown in Code 5:
<partnerLinks>
<partnerLink name="ncname" partnerLinkType="qname"
myRole="ncname"? partnerRole="ncname"?>+
</partnerLink>
</partnerLinks>
Code 5. The partnerLink syntax
As can be seen, each partnerLink must be named and this name is used for all
interactions via that link. myRole is used to indicate the role of the business process,
whereas partnerRole shows the role of the partner.
The business partner role is used to represent relationships with a specic partner.
Such business partnerships often require more than one conversational relationship.
For example, as shown in Code 6, the AuthorizationCapture partner is required
to provide the roles of both payment authorization and payment capture. The
denition of these capabilities appears in the partner element:
<partner name="AuthorizationCapture"
xmlns="http://schemas.xmlsoap.org/ws/2003/05/partner-link/">
<partnerLink name="PaymentAuthorization"/>
<partnerLink name="PaymentCapture"/>
</partner>
Code 6. An example of a business partner role
Partner denitions are optional and need not cover the entire partner links dened in
the process. From the process perspective a partner denition introduces a constraint
on the functionality that a business partner is required to provide. It is important that
partner denitions do not overlap, i.e., a partner link cannot appear in more than
one partner denition.
107
2.4. Message Correlation

Once we have captured the relationships between our enterprise and its partners, we
can begin to exchange messages based on these relationships. However, WSDL is
essentially a stateless model based on two possible patterns:
Synchronous (remote procedure call) interactions, where the underlying
communication system ties together requests and their responses, or
Asynchronous interactions, where request messages and any possible responses
have to be tied together at the application level. Usually this happens in an
ad hoc manner; for example by encoding a unique sequence number in the
outgoing request that the sender of the response echoes back. (Actually that is
pretty much what happens in the synchronous case too, but the communication
system takes care of it for you.)
During a business activity, a given process will often interact with many different
partners in order to conduct work. Those interactions may be based on either
synchronous or asynchronous transport mechanisms. However, the typical interaction
pattern is based on asynchronous (one way) messages because this has the benet of
allowing loose coupling of application entities: the sender of a given message that
requires a response need not be the same as the ultimate receiver of that response.
This allows for great exibility in choosing service deployments, particularly in
environments that may be error prone or require dynamic changes to roles and
responsibilities.
As we mentioned above, in this case some kind of message correlation is required in
order to ensure that messages are delivered to the right BPEL instance: it is obviously
important that messages belonging to one business relationship do not get mixed with
another. Fortunately BPEL does not mandate a specic interaction pattern, but it
does provide language-level support for asynchronous interactions through correlation
sets. For example, let us return to the process order scenario and look at the possible
message interactions between the various processes in a BPEL implementation, as
shown via the UML interaction diagram in Figure 4. Here we have added an Order
Process Orchestrator, which is meant to represent the overall controlling aspect of
the BPEL ow. As you can see, in this example we have modelled all interactions
as asynchronous one-way operations, although we could just as easily produced a
diagram that relied on synchronous interactions.
However, because we are using one-way operations, it is necessary to tie up
responses with requests. For example, if you look at the interaction between
the Order Process Orchestrator and the Payment Authorization Process you can
see the paymentAuthorization request and the paymentAuthorized response.
108
Figure 4. Message Interactions for the Process Order Example

The order process workow will probably be dealing with many client order
requests concurrently, so the Orchestrator will be sending and receiving many
paymentAuthorization and paymentAuthorized messages. It is obviously
important that it can match up these responses with their requests. That is where
correlation sets come in.
A correlation set is a declarative mechanism for dening related groups of operations
within a service instance. It is a mechanism for dening a named group of properties
that serve as a way to uniquely identify an application-level conversation within a specic business protocol (e.g., the paymentAuthorization and paymentAuthorized
interaction we saw earlier). Application messages are associated with correlation sets
and the BPEL implementation is responsible for ensuring that requests and responses
are matched using this unique data.
In order to declare a correlation set for a message it is necessary to dene which
subset of the message properties declared for a message type can uniquely identify a
message instance. Let us look at the order process example in more detail and see how
correlation sets may be dened for it. We will begin by dening three message properties: customerID, orderNumber and dispatchNumber. All of these properties,
illustrated in Code 7, are dened as part of the processOrderCorrelation.wsdl:
109
<definitions name="properties"
targetNamespace="http://example.com/orderProcessCorrelation.wsdl"
xmlns:poc="http://example.com/processOrderCorrelation.wsdl"
xmlns:bpws="http://schemas.xmlsoap.org/ws/2003/03/businessprocess/"
xmlns="http://schemas.xmlsoap.org/wsdl/">

<bpws:property name="customerID" type="xsd:string"/>
<bpws:property name="orderNumber" type="xsd:int"/>
<bpws:property name="dispatchNumber" type="xsd:id"/>
</definitions>
Code 7. An example of message properties

Then we can dene the process order messages, shown in Code 8. For the sake of
simplicity we will ignore error conditions and associated messages:
<definitions name="correlatedMessages"
targetNamespace="http://example.com/processOrderMessages.wsdl"
xmlns:pom="http://example.com/processOrderMessages.wsdl"
xmlns:poc="http://example.com/processOrderCorrelation.wsdl"
xmlns:bpws="http://schemas.xmlsoap.org/ws/2003/03/businessprocess/"

<types>
<xsd:schema>
<xsd:complexType name="ProcessOrder">
<xsd:element name="CID" type="xsd:string"/>
<xsd:element name="order" type="xsd:int"/>
...
</xsd:complexType>
<xsd:complexType name="ProcessOrderResponse">
<xsd:element name="CID" type="xsd:string"/>
<xsd:element name="order" type="xsd:int"/>
...
</xsd:complexType>
<xsd:complexType name="Dispatch">
<xsd:element name="VID" type="xsd:string"/>
<xsd:element name="dispNum" type="xsd:int"/>
</xsd:complexType>
</xsd:schema>
</types>
110
<message name="POMessage">
<part name="PO" type="pom:ProcessOrder"/>
</message>
<message name="POResponse">
<part name="PO" type="pom:ProcessOrder"/>
</message>
<message name="DispatchMessage">
<part name="IVC" type="pom:Dispatch"/>
</message>
<bpws:propertyAlias propertyName="poc:customerID"
messageType="pom:POMessage" part="PO"
query="/PO/CID"/>
<bpws:propertyAlias propertyName="poc:orderNumber"
messageType="pom:POMessage" part="PO"
query="/PO/Order"/>
<bpws:propertyAlias propertyName="poc:dispatchNumber"
messageType="pom:DispMessage" part="IVC"
query="/IVC/InvNum"/>
...
</definitions>
Code 8. The process order messages
Finally, the portType used is dened, in a separate WSDL document, as shown
in Code 9. This example shows both the synchronous (request-response) and
asynchronous (one-way) operations.
<definitions name="orderingPortType"
targetNamespace="http://example.com/ordering.wsdl"
xmlns:pom="http://example.com/processOrderMessages.wsdl"
<portType name="OrderingPortType">
<operation name="SynchronousOrder">
<input message="pom:POMessage"/>
<output message="pom:POResponse"/>
</operation>
<operation name="AsynchronousOrder">
<input message="pom:POMessage"/>
</operation>
</portType>
<portType name="OrderSubmitterPT">
<operation name="AsynchronousOrderResponse">
<input message="pom:POResponse"/>
111
</operation>
</portType>
</definitions>
Code 9. Process order portType
Both the properties and their mapping to purchase order and invoice messages will
be used in the correlation set examples shown in Code 10:
<correlationSets
xmlns:poc="http://example.com/processOrderCorrelation.wsdl">

<correlationSet name="ProcessOrder"
properties="poc:customerID poc:orderNumber"/>

<correlationSet name="Dispatch"
properties="poc:invoiceNumber"/>
</correlationSets>
Code 10. The process order correlation sets
If we take the case of the ProcessOrder operation, then we can see how it is supposed
to work. Figure 5 shows the interaction messages between the client and the process
order service in a UML interaction diagram. For simplicity we have omitted the
interactions between the Orchestrator and the internal processes we saw in Figure 4.
In this case the customerID and orderNumber will be associated with the request
message. The combination of the values of these properties must be unique in the
scope of this workow. Any response message will also have these values associated
with it such that the response can be delivered to the right process instance. As we
have already mentioned, it is up to the BPEL implementation to ensure that these
messages are correctly delivered, so all you need to do is make sure that a correlation
set is appropriately dened and related to your message invocations.
Now that we have seen how to create business relationships and manage the
interactions between them through correlation sets, we need to discuss how an actual
workow (or business process interaction ow) can be dened within BPEL. We will
do that in the next section, where we will discuss in some detail the different types of
interaction patterns that BPEL supports.
112
Figure 5. Client Interactions with the Order Process System
2.5. Activities
As we saw when rst talking about workow systems, they are typically used when
controlling the ow of an application within a corporate intranet. However, there is
nothing intrinsically restrictive about workow models that prevent them from being
used across an enterprise boundary. In fact this is precisely the area where BPEL is
aimed: describing inter-enterprise business interactions, where the individual business
processes from each enterprise are represented as Web services.
BPEL models all stages of a workow as activities. Activities are composed into scopes
to form algorithmic workows. Each scope has a primary activity that denes its
normal behavior; the primary activity can be arbitrarily complex and be composed of
many sub-activities (nested activities). In this case, the scope is shared by all nested
activities.
In order to execute a process, we must have some means of describing its behavior.
The BPEL language provides a number of fundamental activities which form the
basic building blocks of the workow and provide support to manipulate data, iterate,
113
call external functions etc., and how to compose these primitives into meaningful
workows. The language also provides for the structuring of activities to manage
control ow dependencies in a workow. They are responsible for serializing and
parallelizing activities, choosing from alternative paths in a workow, etc.
BPEL classies these activities as either basic or structured types. Basic activities deal
with state management, communication and exception handling, while structured
activities deal with process control-ow issues. In the following sections we will
briey examine the different activity types. If you are familiar with most high-level
programming languages (e.g., C++, Java, C# etc.), then most of the principles behind
these primitives should be fairly intuitive.
2.5.1. <assign>
This construct can be used to update the values of variables with new data (e.g.,
messages). It can contain any number of elementary assignments and it is important
to understand that the type of the source from which the data comes must be the
same as the type of the destination to which it is being assigned.
A typical example of assignment is where the contents of one message are copied
into another. This is illustrated in Code 11, where the address details of a customer
placing a purchase order are copied to some dispatch data prior to shipping the order.
<assign>
<copy>
<from variable="processOrder" part="customerAddress"/>
<to variable="dispatchData" part="customerAddress"/>
</copy>
</assign>
Code 11. An assign example
The syntax for <assign> is actually quite powerful. The syntax for from is captured
in Code 12:
<from variable="ncname" part="ncname"?/>
<from partnerLink="ncname"
endpointReference="myRole|partnerRole"/>
<from variable="ncname" property="qname"/>
<from expression="general-expr"/>
<from> ... literal value ... </from>
Code 12. The assign-from syntax
114
The to eld must be one of the following, shown in Code 13:

<to variable="ncname" part="ncname"?/>
<to partnerLink="ncname"/>
<to variable="ncname" property="qname"/>
Code 13. The assign-to syntax
As we saw in the example, the rst from and to variants simply provide the name of
the variable. If the type of the variable is a WSDL message, then the optional part
attribute can be used to provide the name of a part within that variable (so that only
a subset of the message can be selected).
The second variation of from and to allows for dynamic manipulation of the endpoint
references associated with partner links. The value of the partnerLink attributed is
the name of a partnerLink declared in the process. You can see that in the case of
from, the role of the link must be specied: this is because a process may need to
communicate an endpoint reference corresponding to either its own role or that of
the partners role.
The third form of from and to allows the manipulation of message properties. They
provide a way to clearly dene how distinguished data elements in messages can be
used. The fourth form of from allows processed to perform basic computations on
properties and variables, while the fth allows a literal value to be given as the source
value.
Let us look at how some of the other features of <assign> could be used. Assume we
have three variables customerName, dispatchToName, and dispatchToAddress,
representing the name of the customer who places a purchase order, the name of
the person to whom the order should be dispatched and the address to use when
dispatching the order, respectively, as shown in Code 14:
<variable name="customerName" messageType="foo:person"/>
<variable name="dispatchToName" messageType="foo:person"/>
<variable name="dispatchToAddress" element="foo:address"/>
Code 14 Variable setups
The BPEL sample below illustrates copying one variable to another as well as copying
a part of a variable to a compatible element type:
<assign>
<copy>
<from variable="customerName"/>
115
<to variable="dispatchToName"/>
</copy>
<copy>
<from variable="customerName" part="address"/>
<to variable="dispatchToAddress"/>
</copy>
</assign>
Code 15. An extended assign example
2.5.2. <receive>
Web service operations are exposed to the outside world by a receive activity. This
construct allows the business process to do a blocking wait for a matching message
to arrive. The receive activity is the workow entity which a WSDL operation
maps onto, and so it species the partner it expects to invoke the corresponding
operation, and portType and operation that it expects the partner to invoke.
Code 16 shows the <receive> activity that the purchase order system might expose
as its ProcessOrder operation. The workow uses a variable called PO to hold
incoming messages from customers. When a POMessage is received from a customer
the <receive> activity is activated and a new instance of the workow is created to
run.
<variables>
<variable name="PO" messageType="POMessage">
</variables>
<receive partnerLink="ordering" portType="purchaseOrderPortType"
operation="ProcessOrder" variable="PO">
</receive>
Code 16. A receive example
<receive> is a blocking activity, which means that any workow process that
depends upon the receipt of an appropriate message cannot execute until that event
occurs. The creation of process instances in BPEL is always implicit. Activities that
receive messages (<receive> and the <pick> activity we will see later) can be
annotated to indicate that the triggering of that activity should cause a new instance
of the business process to be created. This is done by setting the createInstance
attribute on the activity to yes. To be instantiated, each business process must
contain at least one such start (initial) activity; obviously such an activity cannot have
a preceding activity in the ow.
116
In some cases it may be possible for a ow to be started by the arrival of one of a

number of different messages and it may not be possible to determine a priori which
message will arrive rst (it is non-deterministic). To cater for this situation, BPEL
allows multiple initial activities to be declared. It is up to the BPEL implementation
to determine which of the concurrent arriving messages will actually trigger the
instantiation of the workow, though this will typically be the rst message to arrive.
Subsequent messages will be routed to the newly created workow.
2.5.3. <invoke>
This construct allows the business process to invoke a one-way or request-response
operation on a portType offered by a partner. In order to execute a request-response
operation, two variables are required for input (request) and output (response)
messages.
<invoke partnerLink="PlacePurchaseOrder"
portType="PurchaseOrderSystemPortType"
operation="ProcessOrder"
inputVariable="CustomerDetails"
outputVariable="InvoiceDetails"/>
Code 17. An invoke example
In Code 17, the invoke activity calls the ProcessOrder operation (from the
PurchaseOrderSystemPortType which is exposed by the PlacePurchaseOrder
partner) when a message from the CustomerDetails variable is received. The result
of executing the operation is the placement of a message in the InvoiceDetails
variable. Obviously sending a message may result in an error (the recipient has
crashed, for example), so a request-response operation may also have to deal with
fault messages, but we will look at those later.
An asynchronous one-way operation is slightly simpler, since there is no
outputVariable. This would t the scenario where interactions between partners
may occur over a long period of time, for example, as illustrated in Figure 5. The
response to a given one-way operation would then be another one-way operation. If
you consider the example above, we could imagine separating the message request
for the purchase from the message response that is the invoice into two one-way
interactions.
117
2.5.4. <reply>
This construct allows the business process to send a synchronous message in reply to a
message that was received through a <receive>. The combination of a <receive>
and a <reply> forms a request/response operation on the WSDL portType for the
process.
<reply partnerLink="purchasing" portType="purchaseOrderPortType"
operation="sendPurchaseOrder" variable="Invoice"/>
Code 18 A reply example
In Code 18 you can see the purchaseOrder systems response to the initial
purchase request. The purchasing partner is sent a message from the Invoice
variable (which contains messages with invoice details), as a result of invoking the
sendPurchaseOrder operation.
2.5.5. <throw>
This activity generates a fault from inside the business process. Every fault is required
to have a globally unique name, and the <throw> must provide the name for the
fault; it can optionally provide additional data (via variables) that may give more
information about the fault. A fault handler can then use the data to inspect and
handle the fault and perhaps use the information to populate any fault messages that
may need to be sent to other services; we will examine fault handlers in the next
section.
<throw faultName="CustomerNotSubscribed"
faultVariable="CustomerOrderFailureVariable"/>
Code 19. A throw example
For example, in Code 19, we have slightly modied the orderProcess ow that we
have seen already. In this case, only customers who have previously subscribed with
the system are allowed to place orders. Therefore, in the case where a customer who
has not subscribed attempts to place an order, the <throw> activity will populate
the CustomerOrderFailureVariable with sufcient data so that the workow
process that deals with the initial placement of orders can create an appropriate failure
message and sent it to the prospective customer.
118
2.5.6. <catch>
Handling faults is also follows a similar pattern to common programming languages
like C++, Java, and C#. A <catch> declaration is used to handle specic faults
occurring within a scope, and a catchAll declaration within the activitys scope
faultHandlers declarations will handle any faults not handled by a more specic
<catch> at the same scope.
For example, in Code 20 we have dened two specic faults that clients of
our order process system may encounter when attempting to place an order and
which we can handle explicitly: the CustomerNotSubscribed fault we mentioned
earlier and a fault that indicates that the system may be unavailable (e.g., crashed),
OrderProcessSystemNotResponding. However, other faults may arise that cannot
be handled by these two fault handlers and in which case we rely on the catchall
activity.
<bpws:faultHandlers>
<bpws:catch faultName="CustomerNotSubscribed">

</bpws:catch>
<bpws:catch faultName="OrderProcessSystemNotResponding">
</bpws:catch>
<bpws:catchAll>

...
</bpws:catchAll>
</bpws:faultHandlers>
Code 20 An example of using the catch activity
2.5.7. <terminate>
When this activity is carried out, the process instance is immediately terminated
without the ongoing work being undone and compensated. The BPEL implementation must terminate all running activities in the process as soon as possible without
any attempt at fault handling.
If we look at Code 21, we can now use <terminate> in the catchall block to exit the
process if any unexpected error occurs. Obviously, this may not be the best course of
action for all processes, but this is just an example.
119
<bpws:catch faultName="CustomerNotSubscribed">
</bpws:catch>
<bpws:catch faultName="OrderProcessSystemNotResponding">
</bpws:catch>
<bpws:catchAll>

<terminate/>
</bpws:catchAll>
Code 21 Using terminate in the catchall handler
2.5.8. <sequence>
This construct allows you to dene a collection of activities to be performed
sequentially in lexical order (the order in which they occur in the syntax). In the
example below, the sub-activities are executed serially.
<sequence>
<invoke partnerLink="paymentAuthorization" .../>
<invoke partnerLink="dispatch" .../>
<invoke partnerLink="paymentCapture" .../>
</sequence>
Code 22 An example of the sequence activity
2.5.9. <ow>
This construct allows you to specify one or more activities to be performed concurrently. A <flow> completes when all of its constituent activities have completed.
If you look back at the process order scenario depicted in Figure 3, it is possible
to see that when an order is rst made, the Authorization stage (whether the client
can place orders) and the checking of whether the order can be fullled, occur in
parallel. How this may be accomplished in BPEL is shown in Code 23, where the
two <invoke> activities (PaymentAuthorization and CheckStock) are enabled
to start concurrently as soon as the <flow> is started. The <flow> completes when
both of the activities respond (we will assume they are invoked in a synchronous
request/response manner). The Dispatch is invoked only after the <flow> completes
(we will assume that there are no preceding errors).
120
<sequence>
<flow>
<invoke partnerLink="PaymentAuthorization" .../>
<invoke partnerLink="CheckStock" .../>
</flow>
<invoke partnerLink="Dispatch" .../>
</sequence>
Code 23 The process order example in BPEL
In the example above, the two concurrent activities have no dependencies between
them. This is not always going to be the case and synchronization dependencies will
exist between activities in a <flow>, i.e., some activities may have to be executed in
a specic order. As such, BPEL provides a link construct that can be used to express
these dependencies. A link is named and all links in a <flow> must be dened
separately. Two activities can be linked together using source and target elements.
Every link must have precisely one activity in the <flow> as its source and precisely
one <flow> activity as its target. The source may also specify a transition condition
through the transitionCondition attribute of the source element. All links that
do not have an explicit transitionCondition attribute have a default attribute
associated with them with a value of true.
BPEL provides another XPath extension to assist in using links, which is shown in
Code 24:
bpws:getLinkStatus (linkName)
Code 24 The XPath function for checking the status of a link
This function returns a Boolean value which indicates the status of the link in the parameter. The status of a link is determined by evaluating its transitionCondition
attribute. This evaluation occurs on the actual values of the variables referenced in
the expression, so if they are also modiable via a concurrent path, the result may be
non-deterministic. If the evaluated result is true then the status is positive, otherwise
it is negative.
getLinkStatus can only be used in a <join> condition. The parameter must refer
to the name of an incoming link for the activity associated with the <join>.
In the example shown in Code 25, we have taken the example of sourcing a book
from an online book shop, pricing shipping and handling and insurance costs, and
then charging the entire cost to a bank account. Locating the book, shipping and
handling and insurance costs can all occur in parallel and the example BPEL code
shows this because each service interaction is a separate <sequence> or <invoke>.
121
However, the nal operation, debiting the bank account, cannot occur until all three
previous activities have completed. This is ensured because it is related to the other
activities through the link synchronizations dened at the top of the <flow>.
<bpws:flow>
<bpws:links>
<bpws:link name="BookCost"/>
<bpws:link name="ShippingCost"/>
<bpws:link name="InsuranceCost"/>
</bpws:links>
<bpws:flow>

<bpws:sequence>
<bpws:invoke name="OnlineBookShopInvocation"
partner="BookShop">
...
</bpws:invoke>
<bpws:receive name="BookShopResponse" ...>
<bpws:source link="BookCost" .../>
</bpws:receive>
</bpws:sequence>

<bpws:sequence>
<bpws:invoke name="ShippingInvocation"
partner="ShippingAndHandlingService">
...
</bpws:invoke>
<bpws:receive name="ShippingResponse" ...>
<bpws:source link="ShippingCost" .../>
</bpws:receive>
</bpws:sequence>

<bpws:invoke name="InsuranceInvocation"
partner="InsuranceBroker">
<bpws:source link="InsuranceCost"/>
...
</bpws:invoke>

<bpws:invoke name="BankAccountInvocation"
partner="Bank" ...>
122
<bpws:target link="BookCost"/>
<bpws:target link="ShippingCost"/>
<bpws:target link="InsuranceCost"/>
...
</bpws:invoke>
</bpws:flow>
Code 25. An example of ow and link
2.5.10. <scope>
This construct allows you to dene a nested activity with its own associated variables,
fault handlers and compensation handler. It is a way of allowing activities to share
common error handling and compensation routines. A <scope> consists of the
primary activity that denes the behaviour of the <scope>, a set of optional fault
handlers and a single optional compensation handler. Each <scope> can be named,
which is important when you may have to refer to one <scope> from within another.
<bpws:scope>
<bpws:catch faultName="OutOfStockFault" .../>
<bpws:catch faultName="UnknownStockFault" .../>
<bpws:catch faultName="UnauthorizedClientFault" .../>
<bpws:catchAll>
<bpws:compensate/>
</bpws:catchAll>

</bpws:scope>
Code 26. An example of the scope activity
Code 26 illustrates how an order may be placed against the
processOrderApplication. At the bottom of the <scope> is the normal behaviour
of the activity, i.e., the placement of an order by a client. However, the <scope>
declares a number of fault handlers with catch activities; the aim of these handlers
it to provide a means to cope with the different types of failures that may occur
during the order placement process. For example, the quantity of the item that is
required by the client may not be available in the warehouse and as such an error
message (OutOfStockFault) will be returned and caught; the process then had the
123
opportunity to try something else, e.g., request less stock, or wait until the warehouse
has been restocked.
The catchall handler has been declared slightly differently in this example: if
we assume that the other fault handlers allow the application to continue to make
forward progress (albeit at a reduced capacity), what happens if an error occurs that
simply cannot be handled in this way or is unexpected? The answer is that in these
situations the entire <scope> may have to be compensated. In the example, the
compensate activity runs the compensationHandler for all of the inner scopes,
which will perform any work necessary to bring the application to the state it had
before the scope was executed. If nested scopes were named, it would be possible to
limit the compensation to a specic scope.
<scope>
<compensationHandler>
<invoke partnerLink="Buyer"
portType="SP:OrderPlacement"
operation="CancelOrder"
inputVariable="getResponse"
outputVariable="getConfirmation">
</invoke>
</compensationHandler>
</scope>
Code 27. A compensationHandler example
Compensation is an important requirement for long running business processes and
had existed in workow systems from the start. Automatic compensation of failures
is possible in only the simplest of situations for business transactions, which rely on
traditional ACID transaction systems to scope their work. In these cases, backward
compensation is provided for by the underlying transaction system, which does not
have (nor does it need to have) any semantic knowledge of the application or what it
is compensating. However, in the case of long running business transactions, ACID
transactions are unsuitable and automatic compensation is not possible: in order
to compensate for complex business-to-business interactions does require semantic
knowledge of the application. As such, compensation handlers have to be written by
the application programmer, since this where the semantic knowledge resides.
Where fault handlers provide alternative forward execution paths through a scope,
compensation handlers, when invoked, undo the work performed by a scope. Since
a compensationHandler for a specic scope reverses that scopes work, the handler
can potentially be as complex and intricate as the scopes normal original activity.
124
A compensationHandler can also be set to compensate an entire business process

after its normal completion (instead of individual scopes).
2.5.11. <wait>
This construct allows a process to unconditionally wait for a given (xed amount)
time period or until a certain time has passed. A typical use for this feature is shown
in Code 28, where the checkStock process waits for a specic time before it checks.
<bpws:while condition="true">

<bpws:wait until="2004-11-12T18:00+01:00"/>

</bpws:while>
Code 28. An example of the wait activity
2.5.12. <pick>
This activity type allows you to block and wait for a suitable message to arrive or
for a timeout to expire. When one of these triggering events occurs, the associated
activity is performed. A <pick> activity declares events that it will trigger on and
corresponding activities to execute once those events occur. Events can take the form
of either the receipt of a message or a time-based event including both duration (time
relative from now) and deadline (xed future time).
It is possible for a <pick> activity to be the rst activity executed by a workow by
being the rst recipient of a message in that workow, in which case it acts like a
receive activity and is annotated with a createInstance="yes" attribute.
Irrespective of whether or not a <pick> activity is used to instantiate a workow
instance, every such activity contains at least one onMessage event. For those <pick>
activities which do not instantiate workows, the onMessage events constrain the
scope in which the activity becomes active.
An example <pick> activity is shown in Code 29, where the activity waits on a
response message from the checkStock system via its onMessage declaration. In
this example, the checkStock system should respond within 1 day and 10 hours or
an error is assumed to have occurred and a timeout is triggered through the onAlarm;
we will assume some fault handling mechanism is executed here.
125
<bpws:pick createInstance="no">
<bpws:onMessage partner="CheckStock"
portType="CheckStockPortType"
operation="checkResponse" variable="WarehouseResponses">
<bpws:correlations>
<bpws:correlation set="BookingsCorrelationSet"/>
</bpws:correlations>

</bpws:onMessage>
<bpws:onAlarm for="P1DT10H">

</bpws:onAlarm>
</bpws:pick>
Code 29. An example of a pick activity
2.5.13. <switch>
This construct allows you to select exactly one branch of activity from a set of
possible choices. It is similar in name and design to switch statements you will
nd in popular languages such as C++ and Java. As you can see in Code 30
(which illustrates the checkStock process using the getVariableProperty XPath
extension we discussed earlier), the case activity consists of an ordered list of one
or more conditional branches (dened by case elements), followed by an optional
otherwise branch (the same as the default case branch in most programming
languages).
<bpws:switch>
<bpws:case condition="bpws:getVariableProperty(stockLevel,
level) < 10" >

</bpws:case>
<bpws:case condition="bpws:getVariableProperty(stockLevel,
level) <= 100" >

</bpws:case>
<bpws:otherwise>

</bpws:otherwise>
</bpws:switch>
Code 30. An example of the switch activity
126
2.5.14. <while>
This construct allows you to indicate that an activity is to be repeated until a
certain success criteria has been met. For example, as shown in Code 31, part of the
checkStock process checks the level of stock in the warehouse and while it remains
below a threshold level it places a request for more stock and waits until the request
has been fullled.
...
<variable name="stockLevel" type="xsd:integer"/>
...
<while condition= "bpws:getVariableData(stockLevel) <= 10">
<scope>

</scope>
</while>
Code 31. A while activity example
2.5.15. <empty>
This construct (shown in Code 32) allows for the insertion of a null-operation
instruction into a business process, which may be useful for synchronization of
concurrent activities.
<empty standard-attributes>
standard-elements
</empty>
Code 32. Syntax of the empty activity
So far we have discussed how to dene business relationships and manage their
interactions. However, we have not talked about transactions: the ability to scope
units of work such that either all of that work happens or none of it does, despite
failures of any of the processes involved in the work. As we saw earlier, when
BPEL4WS was originally released, it came with two other specications; the WSCoordination and WS-Transaction specications. Although we will talk much more
about transactions in a later chapter, it is worth discussing how they relate to BPEL
in the following section.
127
2.6. Transactions
At this stage you may be wondering about the relationship between BPEL and Web
Services transactions. For example, in some cases it may seem appropriate to map a
<scope> with its compensationHandlers to an atomic transaction: the <scope>
is the body of the transaction, while the compensation is the normal behavior of
a transaction that is forced to roll back (abort). However, you will have noticed
that none of the BPEL elements mention transactions and neither does the OASIS
specication. The reasons for a lack of transactions in BPEL are both technical
and political. As we will see in Chapter 7, the traditional transaction model is
not always appropriate in a long running business process example. As such, other
transaction models (such as those based on forward compensation techniques) may be
more appropriate. Chapter 7 will show that these extended transaction models (and
specications based around them) exist in the area of Web Services. Unfortunately,
none of them have yet become the standard for Web Services transactions.
The original BPEL4WS specication did refer to the IBM, Microsoft and BEA
Web Services Transactions specication. However, since BPEL has entered OASIS,
it has gained a wider audience who are not as ready to jump on the proprietary
IBM, Microsoft and BEA bandwagon of specications in other areas. One proposed
solution in the BPEL standards process was to abstract away from the specics of
a give transaction specication: dene what is required without dening how it is
implemented. The other solution was to remove transactions entirely. Due to time
constraints, at this moment the BPEL committee has settled on the latter solution.
What this means is that BPEL implementations do not have to provide any
transaction support in order to be compliant with the specication. However,
in our experience, when deploying workow systems inter- and intra-enterprise,
some form of transaction support (whether it be traditional ACID transactions or
forward compensation-based transactions) is required. Therefore, it is likely that the
more advanced and enterprise ready BPEL implementations will provide transaction
support over and above what the OASIS specication requires. Obviously this may
affect the portability of any workows implemented.
To help place all of what we have just discussed into context and make it much more
concrete, in the rest of this chapter we will look at a worked example. We will slowly
build up this example, using various aspects of BPEL that we have mentioned, to
illustrate how it can be applied to fairly complex business relationships.
128
3. A Worked Example of Web

Services Orchestration
To illustrate the utility of BPEL we will now take our processOrder scenario and
show how it could be implemented using the BPEL 1.1 language. To do this we
obviously need a BPEL implementation. For an area that is relatively new to Web
Services, there are already a large number of implementations. In the J2EE space, for
example, there are:
Collaxa BPEL Server (http://www.collaxa.com), now owned by Oracle,
IBM BPWS4J (http://www.alphaworks.ibm.com/tech/bpws4j) for IBM WebSphere Application Server,
OpenStorm Service Orchestrator (http://www.openstorm.com), and
Vergil VCAB Server (http://www.vergiltech.com/products_VCAB.php),
while in the .NET space:
Microsoft BizTalk 2004 (http://www.microsoft.com/biztalk/), and
OpenStorm Service Orchestrator (http://www.openstorm.com).
It is worth noting that many vendors are splitting their offerings into the graphical
design-time tool, which should produce vendor-neutral BPEL ows, and the run-time
execution engine, which is used to drive these ows. Although you could write BPEL
scripts in any editor (after all, it is only XML schema), BPEL is extremely powerful
and complex, with the result that your application structures may also become
complex. Using a graphical tool to layout the ows is much better as they allow you
to see immediately this structure and assist in reasoning about it, especially in the
case of failures.
For this chapter we have chosen to use the offerings from Collaxa, who were an
independent company until 2004 when they were acquired by Oracle, who is now
offering the product embedded in their Web Services toolkit. However, it is still
possible to use the software stand-alone or with other application servers.
The Collaxa BPEL implementation is split into two components: the design-time
graphical tool (Collaxa BPEL Designer 1.0.3), and the run-time BPEL engine
(Collaxa BPEL Server v2.0 rc6).
Note that it is not going to be possible to demonstrate all of the functionality
available from the BPEL language in the scenario we have been using. Neither will it
129
be possible to show the full power and exibility of the tools we will use. However, we
hope to give you a avor of what is possible with BPEL and encourage you to explore
this area more. As we have already said, tying together Web Services into process
ows offers a powerful opportunity to leverage existing infrastructural investments
without being tied into one specic vendor.
The example scenario we are going to look at in more depth is the order process
example we have seen throughout this chapter and is illustrated in Figure 3.
Unfortunately, to show how to implement this entire scenario in BPEL, covering all
aspects including failure cases, will take more space than we have available. Therefore,
we will concentrate on one specic aspect of the scenario, the payment authorization
segment, and give an overview of how the rest could be tackled.
4. Design-Time Demonstration
In this section we will look at what it means to design a BPEL application using the
tools available. We will examine the design tool and process execution engine and
explain the fundamental steps in implementing this scenario.
4.1. Task Denitions

If you look at the process order scenario again, you will see that there is the
processOrderApplication, which will be a single Web Service, and four sub-tasks:
1. paymentAuthorization: this task determines whether or not the client of processOrderApplication is allowed to place orders. What criteria this uses will depend
upon the implementation, but for now we will assume it is based on the name of
the client.
2. checkStock: this task determines whether or not there is sufcient stock available
to fulll the clients order. Again, what criteria it uses will be implementationdependent, and we have already discussed some alternatives in this chapter and
how they might be implemented in BPEL.
3. dispatch: this task is responsible for fullling and sending the clients order.
This could involve another BPEL process ow, for example between the
processOrderApplication vendor and the delivery company.
4. paymentCapture: this task bills the client. Again it could involve another ow, for
example with an online bank facility.
130
As we have already mentioned, to show how all of this scenario could be implemented in BPEL will consume too much space. So, we will concentrate on the
paymentAuthorization task and how it is related to the overall processOrderApplication
service.
Note that in the following examples we are going to use basic Web Service
implementations that do very little in the way of real business process. This is
because we are going to focus on BPEL and how the scripting language can be used to
glue together Web Services. As we mentioned at the start of this chapter, the advantage
of the Service-Oriented Architecture approach to application development is that
the underlying implementations of specic services can change without affecting the
application. Therefore, replacing the basic Web Services with real-world equivalents
is straightforward, but beyond the scope of this chapter because of space limitations.
4.2. The ProcessOrderApplication Flow

We begin by dening the processOrderApplication Web Service (and by denition, the
global coordinating BPEL ow). In the Collaxa Designer, this is straightforward: you
simply create a new project and the design tool will populate it with templates for
the services WSDL and the ow.
If you look at Figure 6, you can see how this appears in the graphical designer, with
the input (<receive>) and response (<reply>) activities labeled. Part of the XML
corresponding to the ProcessOrderApplication denition of the basic types and service
WSDL is shown in Code 33 (we have removed some of the elements for brevity).
<types>
<element name="ProcessOrderApplicationRequest">
<complexType>
<sequence>
<element name="clientName" type="string"/>
<element name="item" type="string"/>
<element name="delivery-address"
type="string"/>
</sequence>
</complexType>
</element>
<element name="ProcessOrderApplicationResponse">
<complexType>
<sequence>
<element name="result" type="string"/>
Figure 6. Initial ProcessOrderApplication BPEL Flow

</sequence>
</complexType>
</element>
</types>
<message name="ProcessOrderApplicationRequestMessage">
<part name="payload"
element="tns:ProcessOrderApplicationRequest"/>
</message>
<message name="ProcessOrderApplicationResponseMessage">
<part name="payload"
element="tns:ProcessOrderApplicationResponse"/>
</message>
<portType name="ProcessOrderApplication">
<operation name="process">
<input message=
"tns:ProcessOrderApplicationRequestMessage"/>
<output message=
"tns:ProcessOrderApplicationResponseMessage"/>
</operation>
Code 33. The BPEL process denition for ProcessOrderApplication
131
132
As you can see from the portType, the Web Service that clients interact with has an
input and an output (for simplicity we will model this as a synchronous invocation).
The invocation from the client (input) is expected to carry the clients name, the
name of the item being requested and the delivery address for the order (assuming
that it is allowed and can be fullled). The response to the client (output) will
eventually be a string indicating whether or not the order has been allowed. We will
add error handling later.
4.3. The PaymentAuthorization Sub-Task

At the moment our ow does not do much, so let us move on to the paymentAuthorization sub-task. As with the previous service, we will dene a PaymentAuthorization
Web Service and associated BPEL ow from scratch. Obviously if you already have
a Web Service dened and implemented, then you would be able to import that
denition to this ow instead. By default, the Collaxa designer will give us the same
basic message denitions and WSDL templates. So we obviously need to change
these to more accurately reect what this sub-task will do. For simplicity, we will
assume that the authorization criterion is that each client must be pre-registered with
the order process system before they can place an order. Upon registering, each client
is given a unique login name that must be used when making orders; we will assume
that the names all begin with 1234.
So the denition of input and output message types for our PaymentAuthorization
service can be dened as shown in Code 34:
<element name="PaymentAuthorizationRequest">
<complexType>
<sequence>
<element name="clientName" type="string"/>
</sequence>
</complexType>
</element>
<element name="PaymentAuthorizationResponse">
<complexType>
<sequence>
<element name="result" type="boolean"/>
</sequence>
</complexType>
</element>
Code 34. The input and output message type denitions for PaymentAuthorization
133
This will take as input the clients name and then return a Boolean indicating whether
or not the client is authorized to place the order. How the actual Web Service performs
this validation is an implementation choice and could involve diving down into a
specic programming language such as Java or C#. However, the Collaxa design tool
gives us the exibility of staying entirely within the BPEL domain and we will use
that because it further illustrates the power of BPEL and some of the activities we
previously mentioned.
So, the rst thing we want to do is dene an <assign> activity that will look
at the input message for our PaymentAuthorization service and based on the
contents, determine what the output message will be. We can do this by dragging
and dropping an <assign> activity to after the <receive> activity, but before the
<reply> activity, as shown in Figure 7.
Figure 7. PaymentAuthorization and the <assign> Activity

At this stage we have a blank <assign> activity, so we need to make it do something
useful. Working within the design tool, we can associate a copy-rule with our activity,
as shown in Figure 8.
Here, the from-eld is dened as coming from an XPath expression, which uses the
input message (the clients name) and checks whether or not it starts with 1234.
The entire expression is shown in Code 35:
134
Figure 8. The PaymentAuthorization Copy Rule

starts-with (bpws:getVariableData( input,
payload,/PaymentAuthorizationRequest/clientName ), 1234)
Code 35. A sample XPath expression
You should recognize the getVariableData XPath extension that we mentioned
earlier: it was created specically by the BPEL authors for this kind of use. It works
by scanning the payload portion of the input message for the element clientName
within the PaymentAuthorizationRequest.
4.3.1. Testing the Sub-Task within the Design Tool

Using the design tool and the execution engine we can test this ow and associated
service. This is shown in Figure 9, where our client requesting an order will be called
badname, which obviously should not pass the validation criterion. As you can see,
the Collaxa run-time engine provides us with a form that allows us to construct a
suitable input message for the ow. It also allows us to send that message to the ow
and await a response. This is an important and very useful feature because it allows
Figure 9. An Unauthorized Client
135
136
us to stay entirely in the BPEL world in order to test our ow. Even if we eventually
replace the sample Web Service with one that does authorization in a more realistic
manner, the fact that we can test and validate the required message interaction with
a dummy service means that we can isolate any future problems to the service
implementation or a lack in our requirement specication.
If you are interested in seeing what the equivalent BPEL source is that the design tool
has auto-generated for us, then this is shown in Code 36.

<process name="PaymentAuthorization"
targetNamespace="http://acm.org/samples"
suppressJoinFailure="yes"
xmlns:tns="http://acm.org/samples"
xmlns="http://schemas.xmlsoap.org/ws/2003/03/business-process/"
xmlns:bpelx="http://schemas.collaxa.com/bpel/extension">
<!-
PARTNERLINKS:List of services participating in this BPEL
process
-->
<partnerLinks>

<partnerLink name="client"
partnerLinkType="tns:PaymentAuthorization"
myRole="PaymentAuthorizationProvider"/>
</partnerLinks>
<!-
VARIABLES: List of messages and XML documents used in this
process
-->
<variables>

<variable name="input"
messageType="tns:PaymentAuthorizationRequestMessage"/>

<variable name="output"
messageType="tns:PaymentAuthorizationResponseMessage"/>
</variables>
137
<!-
ORCHESTRATION LOGIC: Set of activities coordinating the
flow of messages across the services integrated within
this business process
-->
<sequence name="main">

<receive name="receiveInput" partnerLink="client"
portType="tns:PaymentAuthorization" operation="process"
variable="input" createInstance="yes"/>

<assign name="authorisedResponse">
<copy>
<from expression="starts-with
(bpws:getVariableData( input, payload,
/PaymentAuthorizationRequest ), 1234)">
</from>
<to variable="output" part="payload"
query="/PaymentAuthorizationResponse/result"/>
</copy>
</assign>
<reply name="replyOutput" partnerLink="client"
portType="tns:PaymentAuthorization" operation="process"
variable="output"/>
</sequence>
</process>
Code 36. BPEL language for the PaymentAuthorization service
If we then rerun the example with a valid name (1234goodclient in this case), the
results are as expected, and shown in Figure 10.
Hopefully, at this stage you can see how relatively straightforward it would be to
replace the default PaymentAuthorization Web Service with one that was based
on something more meaningful.
Finally, we will publish our PaymentAuthorization service so that it can be
integrated within the overall business scenario. How the service is made available to
the rest of the world is a deployment choice that will typically be based on a number
of factors, including trust relationships between partners. For example, the WSDL
138
Figure 10. An Authorized Client

could be exported as a plain text le that is available by virtue of a shared disk. Another
alternative would be to use a UDDI registry. There may be other options that are
specic to the BPEL implementation that you use. The Collaxa design tool allows,
for example, allows a deployment into a jar or zip le. In this example, however, we
will use the simplest approach and export our service to the BPEL run-time engine.
4.4. Gluing Them Together

If we now return to the global BPEL ow, we can start to glue together the
processOrderApplication into a form similar to the one in Figure 3. If you remember,
so far we have a ow shown in Figure 6 for the ProcessOrderApplication ow.
Now we can add in the PaymentAuthorization service we have just dened. For
the sake of this example, in what follows we will ignore the checkStock sub-task
because we have not dened it yet.
First we will begin by adding an authorization <scope> activity, because we want
to be able to catch any failures that may occur and we also need to be able to
create variables for our ow to use. As with most things in the design-tool, we can
drag-and-drop the desired activity.
139
Figure 11. The Blank Scope Activity
As one can see from Figure 11, initially the <scope> has no fault handlers and does
nothing. However, we can quickly rectify this by putting an <invoke> activity within
the scope. Remember, what we are modeling is the ProcessOrderApplication
process requesting of the PaymentAuthorization task to validate the clients
credentials.
In order to do this with the <invoke> activity, we must dene the partner link
relationship. In the design tool that we are using, once the <invoke> activity has
been installed within the <scope>, a special partner link creation template is made
available for that activity, which is shown in Figure 12. In this template, we rst
dene the name of the partner link (PaymentAuthorizationPartner) and then
give the location of the WSDL for that partner. As you can see, this tool allows us to
specify the WSDL location either explicitly (which we do here), or we could browse
a UDDI registry.
Once the WSDL has been dened, the tool goes and fetches it so that it can
offer us the partnerLinkType and partnerRole denitions. The resultant BPEL
partnerLink denition is shown in Code 37. Once again, this is all automatically
generated by the tool.
140
Figure 12. Establishing a PartnerLink

ProcessOrderApplication.wsdl -->
portType="tns:ProcessOrderApplication" operation="process"
<scope>
<invoke partnerLink="PaymentAuthorizationPartner"
portType="tns:PaymentAuthorization"/>
</scope>
variable="output"/>
</sequence>
Code 37. Sample partnerLink denition
Next we have to be able to take some of the message that was input to the
ProcessOrderApplication (the client name) and pass them through to the
141
PaymentAuthorization sub-task. Likewise, we want to take the result of the

PaymentAuthorization sub-task and be able to use it within the rest of our
application. As you may have guessed by now, we do this through the use of <assign>
activities. However, we need to create some global variables to store intermediate data.
We do this by establishing that the operation we want ProcessOrderApplication
to invoke is process, which has both input and output messages; the design tool
knows from the XML denition, what then types of these messages are as well.
We will need a global variable for storing the outgoing client data in (called
clientCredentials), and one for storing the PaymentAuthorization response
(called clientValidation). Using the tool we establish these as the input and
output variables for the <invoke> activity. Now we need to use <assign> to set
up these two variables. For brevity we will only show how this is accomplished
for clientCredentials. The <assign> template is shown in Figure 13, with
the relevant XPath expression for extracting the clientName element from the
ProcessOrderApplication input message.
Figure 13. Initializing the ClientCredentials Variable

The resulting ow is shown in Code 38, and illustrated graphically in Figure 14.
142
Figure 14. The PaymentAuthorization Scope Activity
ProcessOrderApplication.wsdl -->
<scope name="authorization">
<sequence>
<assign name="InitializeClientCredentials">
<copy>
<from variable="input"
part="payload" query="/ProcessOrderApplication/clientName">
</from>
<to variable="clientCredentials"
part="payload"/>
</copy>
</assign>
<invoke
143
partnerLink="PaymentAuthorizationPartner"
portType="tns:PaymentAuthorisation" name="paymentAuthorization"
inputVariable="clientCredentials"
outputVariable="clientValidation" operation="process"/>
<assign name="ObtainAuthorizationResponse">
<copy>
<from variable="clientValidation"
part="payload">
</from>
<to variable="output"
part="payload"/>
</copy>
</assign>
</sequence>
</scope>
variable="output"/>
</sequence>
Code 38. The BPEL source for the PaymentAuthorization scope activity
4.5. Fault Handling

So far we have ignored any faults that may occur during the ow of our application.
Unfortunately faults of one kind or another are common place. For example, they
may occur because of a machine or process failure, or simply because of invalid
input data. As we have already seen, BPEL provides a fairly comprehensive suite
of facilities to both model faults (e.g., the <throw> activity) and deal with them
(e.g., fault handlers).If we consider the PaymentAuthorization section of our
ProcessOrderApplication, then there are several types of fault that we may want
to model and deal with. For example, if the format of the clientName element is
invalid, then we would want to report that back to the client; this could be through a
normal response message, or (probably more useful) through an error response, i.e.,
a fault message.
Likewise, the client identied by clientName may actually be in debt to the order
system, and in which case the next time that client attempts to place an order an
entirely separate process ow may be required, e.g., a DebtCollectionService.
There are obviously other failure scenarios that we could model here, but the aim
is not to include them all but to give a avor of how such fault types can be
144
catered for within BPEL. In order to model these two types of fault we need
to return to the PaymentAuthorization service and allow it to throw these
two fault types: ClientNameFormatFault and UnpaidBillFault respectively.
Then we go back to the authorization <scope> activity we declared earlier in the
ProcessOrderApplication ow. If we remember from Figure 11, the design tool
shows that the activity has scope for fault handlers as well as a compensation handler
and event handlers (e.g., onMessage). It is relatively straightforward to add two fault
handlers to this scope, one for each of the faults that the PaymentAuthorization
service can now throw. Figure 15 shows the handler for the UnpaidBillFault.
Figure 15. Handling the UnpaidBillFault
4.6. The Entire Flow

With a little more effort it is possible to dene the entire processOrder application in
BPEL. This is shown in Figure 16 and looks a lot like what we saw in Figure 3.
145
Figure 16. All of the Order Process Application in BPEL
5. Run-Time Demonstration
Now that we have designed our ProcessOrderApplication, we can execute it and
examine what actually happens from one step to another. As we did when designing
it, we will only concentrate on the PaymentAuthorization sub-task for simplicity.
5.1. Tracking the Flow

If one recalls, the ProcessOrderApplication is invoked with three pieces of
information:
1. The client name.
2. The type of item required.
3. The clients shipping address.
It then invokes the PaymentAuthorization sub-task, passing it the client name,
to determine if the client is allowed to place an order. We previously saw how the
146
ProcessOrderApplication and PaymentAuthorization processes have been

glued together, so now let us see how they execute. Once again the Collaxa run-time
engine facilitates this by providing us with a suitable form, as shown in Figure 17.
Here we have three elds, one for each of the data variables that a client must supply.
Figure 17. Placing an Order

Through its Flow option, the Collaxa Server interface allows us to see all of the
message exchanges that occur in this ow. The result of submitting this form is
a message being posted to the ProcessOrderApplication. The format of that
message is shown in Code 39.
<input>
<part name="payload">
<ProcessOrderApplicationRequest
xmlns="http://acm.org/samples">
<clientName>1234goodclient</clientName>
<item>widget</item>
<delivery-address>Widgets R Us, NY, NY
</delivery-address>
</ProcessOrderApplicationRequest>
</part>
</input>
Code 39. The client message to the ProcessOrderApplication ow
147
Upon receiving the clients message, the InitializeClientCredentials

<assign> activity we saw earlier scans the message for the clientName eld,
as shown in Code 40, in order to update the clientCredentials global variable.
<clientCredentials>
<PaymentAuthorisationRequest
xmlns="http://acm.org/samples">1234goodclient
</PaymentAuthorisationRequest>
</part>
</clientCredentials>
Code 40. Updating the clientCredentials variable
Once the clientCredentials variable has been successfully initialized,
the PaymentAuthorization ow can be invoked. The message that the
ProcessOrderApplication sends to this sub-task is shown in Code 41, along with
the response that eventually comes back.
<messages>
<clientCredentials>
<PaymentAuthorisationRequest
mlns="http://acm.org/samples">1234goodclient
</PaymentAuthorisationRequest>
</part>
</clientCredentials>
<clientValidation>
<PaymentAuthorisationResponse
xmlns="http://acm.org/samples">
<result>true</result>
</PaymentAuthorisationResponse>
</part>
</clientValidation>
</messages>
Code 41. The PaymentAuthorization request and response
Likewise, the response is placed in the clientValidation variable and eventually
used in the rest of the ProcessOrderApplication ow.
148
5.2. The Audit Trail

Through the interface to the Collaxa Server run-time, it is possible to obtain an
audit trail of this segment of the ProcessOrderApplication. This gives similar
information to the Flow we saw previously, but in a slightly different format, as
shown in Figure 18.
Figure 18. An Example of the Audit Option
6. Summary
In this chapter we have looked at the principles behind workow systems and why
they have been used successfully for a variety of enterprise integration strategies.
However, as we saw, the lack of a rigorous standard for workow that concentrated
on interoperability of implementations often resulted in vendor lock-in. Fortunately,
Web services are specically about fostering systems interoperability and the SOA
is deliberately not prescriptive about what happens behind service endpoints: Web
149
services are ultimately only concerned with the transfer of structured data between
parties, plus any meta-level information to safeguard such transfers.
As a result, combining Web Services and workow offers the best of both worlds: an
interoperable integration strategy that allows you to leverage existing investments. We
then discussed the Business Process Execution Language (BPEL) now in OASIS, but
originally from BEA, IBM and Microsoft, which is a workow scripting language for
specifying business process behavior, based on Web services and is rapidly becoming
the standard for Web Services workow. The language (entirely XML-based) can be
used to formally dene the behavior of business processes and their interactions.
More importantly, BPEL has the potential to commoditize the capabilities provided
by the old workow and proprietary EAI and BPM solutions. This is extremely
important because it should allow integration tasks to leverage existing (legacy)
investments in EAI and provide an integration path between different vendor
implementations.
After discussing the principles behind BPEL we further illustrated the advantages that
it offers by using a worked example (an order process scenario). As we saw, with the
aid of a suitable graphical tool, designing fairly complex process ows using BPEL is
relatively straightforward. With only a few mouse-clicks and a basic understanding
of the BPEL language, we were able to dene the paymentAuthorisation task of our
original processOrder application. What BPEL (and by implication Web Services), its
supporters and implementers have done is to take the once elitist area of workow
systems and bring them to the masses. It is no longer a requirement that specialist
EAI vendors have to be the rst and only port of call for corporations large and small
who wish to automate their internal or external business processes.
4
WORKING WITH
REGISTRY AND UDDI
Look at a day when you are supremely satised at the end. Its not a day when you
lounge around doing nothing, its when youve had everything to do and youve done it!
Margaret Thatcher
The demeanor of conducting business has changed over the past decade. However,
there are certain entities, the real players in a transaction that have remained the same
buyers, sellers and marketplaces. Whether you buy a book over the Internet, shop at the
nearest mall or take advantage of the spa service at your local club, the above entities
hold true. The buyers and sellers meet at a market place where the seller showcases its
products and services. The buyer browses through it, evaluates it and may decide to
perform the transaction. In the context of SOA, a registry resembles a market place.
It provides applications and businesses a central point to store information about
their services. It is expected to provide the same level of information and the same
breadth of services to its clients as that of a conventional market place. However,
the vision of a registry solution does not end here. The dream of the enterprise
architects is to have a solution that facilitates the automated discovery and execution
of e-commerce transactions and enabling a liquid and frictionless environment for
business transactions. Therefore, a registry is more than an e-business directory.
It is an inherent component of the SOA infrastructure that should be mature to
propose pragmatic standards yet exible enough to manage the growing complexity
and dynamism in business relationships and interoperability.
There are more than one industry initiatives to handle the issue of registry. In this
chapter, we will take a look at Universal Description, Discovery and Integration
(UDDI). Before we examine UDDI, it is imperative to understand the role and basic
elements of a registry solution in business transactions.
151
152
1. Introducing the Registry

A Service-Oriented Architecture is based on the interactions between three primary
functionaries: service provider, service registry and service requestor. The service
provider creates the service and publishes the service description in a registry. The
service requestor nds the service description in the registry and uses the information
to bind and execute the service. The following diagram illustrates this concept.
Figure 1. Publish, Find and Execute
1.1. Why Do I Need It?

On a small scale, it is not very difcult to discover, manage and interface with business
partners. One option is to call each partner on the phone, and then try to nd the
right person to talk with. It?s cumbersome but still possible to maintain a list of
partner companies that adhere to your requirements and their service access points
and interfaces. However, this model breaks down, as the number of companies that
you need to interact with grows, along with the number and types of interfaces they
export. To manage the list of growing partners and their changing interfaces becomes
an exponential task. It not only makes your system frail but it also restricts your
business ability to discover and collaborate with new partners. A registry solution
Chapter 4: Working with Registry and UDDI
153
based on agreed upon standards provides a common way to publish and discover
services. It offers a central place where you query whether a partner has a service that
is compatible with in-house technologies or to nd a list of companies that supports
shipping services on the other side of the globe.
1.2. How Do I Use It?

The repository of services has several different uses, based on the perspective of who is
using it. From a business analysts perspective, it is similar to an Internet search engine
for business processes. Similar to a web user that uses a search engine like Google to
nd websites of interest, a business analyst can browse one ore more registries to nd
new businesses. A series of marketplaces and business search portals are coming up to
assist business analysts with the right tools in their quest to discover new partners.
The other set of users are the software developers who use the registrys programming
APIs to publish services and query the registry to discover services matching various
criteria. Figures 2 describes the relationship between registries and its users.
Figure 2. Registry Usage
154
1.3. Registry vs Repository

It is important to note that registry is an enabler of services and that it is not a
part of the service. It allows for the registration of services, discovery of metadata
and classication of entities into predened categories. Unlike a repository, it does
not have the ability to store business process denitions or WSDL or any other
documents that are required for trading agreements. A registry stores information
about the items and not the items themselves. A repository, on the other hand, is
used to physically store the items. Thus a repository serves as data storage and makes
use of a registry service as an interface to outside parties.
A registry service can be thought of as an advanced address book, where every entry
contains information in a pre-dened format and the entries are organized for easy
access. Whenever a consumer or client is looking for a service, it can browse or search
the address book by specifying its criterion. Any distributed programming framework
needs a module to serve this need. J2EE has the notion of JNDI, where the client
Java programs can look up for distributed objects like EJBs. CORBA provides a
similar COS Naming Service that allows the clients to look up CORBA objects
by name. Web Service Inspection Language (WS-Inspection) and WS-ServiceGroup
from the WS-Resource Framework specications also attempt to address the issue of
service discovery. Electronic business XML (ebXML) that makes use of XML and
provides a standard, non-proprietary way of handing Electronic Data Interchange
(EDI) between business partners also has a discovery solution. However, due to the
simplicity, exibility and a strong focus on the business and service listings, Universal
Description, Discover and Integration (UDDI), the topic of this chapter has gained
the maximum traction in the ESOA context.
2. Universal Description, Discovery

and Integration (UDDI)
A good registry solution should provide a platform-independent way of describing
services, discovering businesses, and integrating business services using the Internet.
It should facilitate consumption of these services by providing the clients the ability
to query and retrieve details about the business, the services offered and details about
the services, such as how and where to invoke them. The Universal Description,
Discovery and Integration projects provides an open, standardized and exible way
to achieve the above.
155
Microsoft, IBM and Ariba rst proposed the UDDI specications in the year 2000.
By the time version 3.0 of the specications was released in the year 2002, the
UDDI consortium consisted of more than 200 members. Keeping with the spirit of
open standards, the consortium handed over the future UDDI development efforts
to the open-standards organization OASIS. The specications have gone under
signicant evolution since its rst inception. While the initial UDDI specication
focused on Web service description and discovery on an Internet scale, subsequent
UDDI renements helped it become a more effective and pragmatic solution. The
introduction of the subscription API and the means to facilitate the gradual migration
of a service from an internal development setting to the enterprise level registry and
nally to the web is evidence that the specications have matured enough to handle
real world issues.
In simplistic terms, a registry solution can be compared to a phone book. It provides
the ability for service providers to register their services and for service consumers
to easily locate the business entities and the services they provide. Conceptually, a
business can register three types of information into a UDDI registry.
1. White pages
Basic contact information and identies about a company, including business
name, address, contact information and unique identiers such has DUNS or tax
ids. This information allows consumers to locate your serivce based upon your
business identication. This is similar to looking up either the phone number or
address of a business when you know the businesss name.
2. Yellow pages
Yellow pages describe a service using classication information. For instance,
the phone directory can provide information to nd an Italian restaurant in
the San Francisco area. It allows consumers to discover a service based upon its
categorization (taxonomy). We will discuss more about taxonomies later in this
chapter.
3. Green pages
Green pages allow to describe the service that the business offers. They contain
technical information about the behavior, support functions and access point for
the service.
2.1. Technical Overview

Figure 3 illustrates how UDDI ts into an overall stack of services. Rather than trying
to reinvent the technologies, it builds on a network transport layer and a SOAP-based
XML messaging layer and uses WSDL for the denitions of services.
156
Figure 3. Stack of Elements Required for an Established Service
UDDI presents a number of APIs such as inquiry API, publisher API and subscriber
API for the consumers to access the registry. The specications also present an
information model composed of instances of persistent data structures called entities.
The core entities are businessEntity, businessService, bindingTemplate
and tModel. Each of the entities are expressed in XML and are persistently stored
by UDDI nodes. A set of services that implements at least one UDDI API set is
termed as a UDDI node. One or more UDDI nodes may form a UDDI registry, with
the restriction that a node may belong to one, and only one, registry. The nodes
of a registry collectively manage a well-dened set of UDDI data. Typically, this is
supported by the use of UDDI replication between the nodes in the registry, which
reside on different systems. The nodes are hosted by operators who are responsible
for durable recording and backup of all data, providing security and maintaining the
integrity of data.
Within a registry, each instance of the core data structures is uniquely identied by a
UDDI key. By choosing appropriate policies, multiple registries may form a group,
known as an afliation, whose purpose is to permit controlled copying of core data
structures among them. In the next few sections, we will examine each of above
concepts in more detail.
157
2.2. Informational Structural Model

UDDI registry consists of four core data structure types, the businessEntity,
the businessService, the bindingTemplate and the tModel. Additionally,
publisherAssertion data structure is used to dene relationships between the
entities and operationalInfo can be used to maintain the tracking information.
The data structures and their relationships are displayed in Figure 4.
Figure 4. Structural Model

To summarize the relationship described in the above diagram, the businessEntity
element represents a business and contains references to one or more
businessService elements. businessService and bindingTemplate elements
denes the technical and business descriptions of the service and contains a reference
to one or more tModels. The tModel element is used to dene the technical
specication for the service. Additionally, publisherAssertion element can be
used to dene a relationship between two or more businessEntity elements. Let
us examine each of these elements in more detail.
158
2.2.1. Business Information: The BusinessEntity

Element
Partners should be able to discover your business with a small set of facts about
your business such as your name, category or some other key identier. The
<businessEntity> structure serves as a top-level information manager. Each
<businessEntity> entity contains descriptive information about a business
or organization. This information includes contact information, categorization,
identiers, descriptions, and relationships to other businesses. From an XML
standpoint, the <businessEntity> is the top-level data structure that holds
descriptive information about the business or organization it describes. Figure 5
lists the elements that are contained in the structure. Each <businessEntity> is
identied by its businessKey. If a businessKey is not specied at publication
time, the registry automatically generates one. categoryBag and identifierBag
are used for classication of entities and are discussed later in the section. The
Signature element species the digital signature of the publisher and can be used
to verify the integrity of the data. Each contained <businessService> describes a
logical service offered by the business or organization.
Figure 5. BusinessEntity
159
2.2.2. Service Information: The BusinessService

element
A <businessEntity> contains one or more <businessService> structures. A
<businessService> structure represents a logical service and contains descriptive
information in business. A <businessService> document is re-usable, i.e. a
<businessService> element can be used by several <businessEntity> elements.
For instance, a large enterprise may decide to publish multiple <businessEntity>
structures, one each of its several subdivisions. But it has a shipping service that it
would like to be published within each of the entities. It can do so by using service
projection where the <businessService> will be included by reference as opposed
to containment.
Technical information about the <businessService> is found in the contained <bindingTemplate> entities. Figure 6 lists the various elements of
<businessService> structure.
Figure 6. BusinessService
2.2.3. Specication Information: The

BindingTemplate Element
A <bindingTemplate> contains pointers to technical descriptions and the access
point URL, but does not contain the details of the services specications. It describes
the type of service being offered using references to <tModels>, application-specic
160
parameters, and settings. The key to a <bindingTemplate> is that it allows a

service to expose what bindings it supports. A service may choose to support multiple
binding protocols, including HTTP, HTTPS, SMTP and so forth. Figure 7 describes
the <bindingTemplate> structure.
Figure 7. BindingTemplate
2.2.4. Technical Fingerprint: The TModel Element

Software that communicates with other software invariably adheres to some preagreed specications. In order to invoke a purchase order service with a business
partner, you need more than a URL. You need to know the format of the purchase
order, the required protocols, security handshake and the expected result type. This is
achieved by publishing information about the specication in a <tModel> (Technical
Model). A <tModel> acts as an abstract description of a particular specication or
behavior to which the web service adheres. It is a type of digital ngerprint for
determining the specics of how to interact with a particular service. It is important
to note that the actual specication or set of documents that describes the concept
of a <tModel> is not a part of the registry and is remotely referenced using the
overviewDoc structure. Once a <tModel> is published, other parties can express
the availability of Web services that are compliant with a specication the <tModel>
represents. They can do so by simply including a reference to the <tModel>
i.e., its tModelKey in their technical service descriptions <bindingTemplate>
data. Companies can use the information pointed to by a <tModel> to determine
whether a service is compatible with their business requirements. For example, let us
161
consider the situation where the company DummyMortgages has a relationship with
several lenders for processing home loan applications, each lender having its own
implementation. Since the service denition requirement for DummyMortgages is
the same for all partners, it could publish the service denition tModel as a standard
for all of its partners. The several lenders can reference the DummyMortgages service
denition and publish their own service implementation details. The concept of
tModel allows various entities like DummyMortgages or standard bodies to publish
abstract service specications that can then be used by other partners that implement
services. By referencing to the tModel, the partners agree to be compliant with the
service specication dictated by the tModel.
Figure 8 lists the elements in the <tModel> structure.
Figure 8. TModel
2.2.5. Relationships: The PublisherAssertion Element

Many businesses and organizations are not effectively represented by a single
<businessEntity>. Examples include corporations with a variety of subsidiaries,
private exchanges with sets of suppliers and their customers and industry consortiums
with their members. An obvious solution is to publish several <businessEntity>
structures. However, these entities are still under the same umbrella company
and might want to be coupled with each other and make some of their relationships visible in their UDDI registrations. This is accomplished by using the
162
<publisherAssertion> structure. To eliminate the possibility that one publisher

claims a relationship to another that is not reciprocated, both publishers must publish
identical assertions for the relationship to become visible. Thus a company can
claim a business relationship only if its partner asserts the same relationship. One
companys assertion about a business relationship is not visible to the public until its
partner creates a similar, but separate, <publisherAssertion> document for its
own <businessEntity> structure. Thus, if Company A asserts a relationship with
Company B (fromKey=A,toKey=B), then the relationship will become public when
Company B asserts a relationship with Company A (fromKey=B,toKey=A).
Figure 9. PublisherAssertion
2.2.6. Operations Information: The OperationalInfo

Element
Information about a publishing operation is captured whenever a UDDI core data
structure is published. This data includes the date and time that the data structure was
created and modied, the identier of the UDDI node at which the publish operation
took place, and the identity of the publisher. The <operationalInfo> structure
is used to convey the operational information for the UDDI core data structures,
that is, the <businessEntity>, <businessService>, <bindingTemplate> and
<tModel> structures. Figure 10 lists the elements of an <operationalInfo>
structure.
2.3. UDDI Keys

When the above components are published in the registry, they are assigned a globally
unique identier or key. It is via that key that you refer to a registry entry. With the
163
Figure 10. OperationalInfo
earlier versions of UDDI, the registry generated the keys. Version 3 of specications
allows businesses to suggest a key at the time of publishing the entry. With Version
3, the key can be generated either by the registry (UUID/uddiKey), or specied by
the client (domainKey) or a combination of the above two (derivedKey). Let us
examine the various key types.
2.3.1. UUID
Universally unique identier, known as UUID is assigned when the data structure
is rst inserted into a UDDI registry and the user does not specify a key. They
are 16 digit hexadecimal strings and the registry guarantees the generation of a
unique identier by concatenating the current time, hardware address, IP address,
and random number in a specic fashion. An example of a uuid-based registry key is:
uddi: 4CD7E4BC-648B-426D-9936-443EAAC8AE23
2.3.2. DomainKey
domainKey allows the publisher to the register its domain name as part of the key
and in a format that is more palatable for humans to read. Because domain keys
are based on an organizations internet domain name, they also facilitate the aim of
global uniqueness, and allow a business to dene its own policy for generating their
service keys. The following is an example of a valid domain key.
uddi: springer.com
164
2.3.3. DerivedKey
The derived key can be generated by appending a key appending string (KSS) to
a UUID or a domainKey. A KSS may consist of any non-null alphanumeric ASCII
character, as well as a few other special characters. The following listing illustrates a
valid derived key based on UUID.
uddi: 4CD7E4BC-648B-426D-9936-443EAAC8AE23:Enterprise SOA Book
In a similar manner, a derived key can be generated using a domainKey
uddi: springer.com:Enterprise SOA Book
The introduction of domain keys and derived keys makes managing service keys
much more organized. But even with user-specied keys, a UDDI registry still has
to enforce global key uniqueness. When publishing a new registry entry, the UDDI
registry checks with its internal key verication policy if the suggested registry key
is acceptable. If it is, the entry will accept the provided registry key; otherwise, the
UDDI registry will generate a UUID.
Data is only considered information if it can be discovered and used. It is worthless
if it is lost within a mass of other data. If a client application cannot effectively nd
information within a registry, then the purpose of the registry service is considerably
compromised.
2.4. Classication Where Is My Data?

Depending on the number of entries in the registry and the search criteria, the result
could be a large and unmanageable set. Providing the structure and modeling tools to
address this problem is at the heart of UDDIs design. As illustrated in the previous
section, the core components of UDDI, businessEntity, businessService,
bindingTemplate and tModels have categoryBag and identifierBag
elements to help discover these entities in the registry. In this section, we will look at
categorization and identication.
Taxonomy The term is derived from the Greek taxis (arrangement) and nomos
(law). It is dened as the science, laws, or principles of classication.
165
2.4.1. Categorization
Categorization allows data in a UDDI registry to be associated with an industry,
product or geographic code set. UDDI facilitates a classication system to be used on
every entity contained in the registry. It is generally recommended that all services be
classied before being made publicly available. You can either use any of categorization
schemes supported by the base specications or create your own.
As part of the base specications, UDDI supports three canonical taxonomies that
be used for categorization of businesses and the services they offer. Each taxonomy
categorization is registered as a <tModel> structure within UDDI. This registration
means that each categorization has a <tModel> name and UUID that can be used to
reference it. Table 1 summarizes the list.
Table 1. Taxonomies
Taxonomy Name
TModel name
Description
NAICS
ntis:gov:naics:1997
UNSPSC
unspsc-org:unspsc:3-1
ISO 3166
iso-ch:3166:1999
The North American Industry Classication system jointly developed by US, Canada and Mexico. Provides classication for services and manufacturing including categories for Dog and Cat
Food, Knit Outerwear and Computer Storage Devices. More information can be found at
http://www.census.gov/epcd/www/naics.html
The Universal Standard Products and Services
Classication provides classication for products
and services for worldwide use. More information can be found at http://www.unspsc.org
International standard geographical regions.
This taxonomy includes codes for countries
and their administrative support staffs.
More information can be found at
http://www.din.de/gremien/nas/nabd/iso3166ma
UDDI registries also allow using your own classication schemes. Sometimes it
is necessary to dened additional classication schemes particularly when you are
dening businesses for internal registries. You could create a new taxonomy based on
price, service or other factors. However, identifying categories for easy searching is
not an easy proposition. Anyone who has ever searched the web for an item should
be familiar with the problem. If the category is too broad, like manufacturing, it can
return countless businesses and services thus overwhelming the client. If the category
166
is too specic, like manufacturing yellow trucks in San Diego, it might be too
restrictive to return any results. UDDI allows multiple classication schemes to be
applied to a single entity. The categories are assigned with the use of <categoryBag>
element that is part of the core UDDI entities.
Figure 11. CategoryBag

A <categoryBag> structure contains zero or more <keyedReference> structures and may also contain a list of keyedReferenceGroup structures. Each
<keyedReference> structure contains the name and value of a category to which the
data element belongs. A keyedReference contains the three attributes tModelKey,
keyName and keyValue.
For example, you would use the following code snippet in order to categorize a
businessEntity as offering goods and services in California, USA, using the
corresponding ISO 3166 tModelKey within the UDDI Business Registry.
<keyedReference
tModelKey="uddi:uddi.org:ubr:categorization:iso3166"
keyName="California, USA"
keyValue="US-CA"/>
The tModelKey refers to the tModel that represents the categorization system, and
the keyValue contains the actual categorization within this system. The keyName is
optional and is used to provide a descriptive name of the categorization.
A keyedReferenceGroup is simply a list of keyedReference structures that
logically belong together. It also contains a tModelKey attribute that species the
structure and meaning of the keyedReferences contained in the group. For
instance, the following listing illustrates a keyedReferenceGroup to categorize a
businessEntity as being located at the geodetic point that is specied by the
167
latitude/longitude pair 49.6827/8.2952 using the corresponding World Geodetic

System 1984 (WGS 84) tModelKey within the UDDI Business Registry.
<keyedReferenceGroup tModelKey="uddi:uddi.org:ubr:
categorizationGroup:wgs84">
<keyedReference
tModelKey="uddi:uddi.org:ubr:
categorization:wgs84:latitude"
keyName="WGS 84 Latitude"
keyValue="+49.682700"/>
<keyedReference
tModelKey="uddi:uddi.org:ubr:categorization:
wgs84:longitude"
keyName="WGS 84 Longitude"
keyValue="+008.295200"/>
</keyedReferenceGroup>
2.4.2. Identiers
Another way to search an entity in the registry is by using a unique key or identier.
UDDI registry has the ability to mark entities with identier. An identier is a
type of property or keyword that can be used to uniquely identify a business or
specication. The purpose is to allow others to nd the published information using
more formal identier systems. For instance, businesses may want to use their DU-N-S number, Global Location Number (GLN), or tax identier in their UDDI
registration data, since these identiers are shared in a public or private community in
order to unambiguously identify businesses. In UDDI registries that are only used in
private communities, businesses may also want to use privately known identiers. For
example, in a UDDI registry that is used as a service registry in a private exchange,
supplier identiers that are only known in this community might be used to identify
the businesses. Identiers can be applied to <businessEntity> and <tModel>
structures.
Identiers and categorizations are implemented similarly. Identiers are attached
to <businessEntity> and <tModel> documents through an <identifierBag>
structure. As illustrated in Figure 12, the <identifierBag> structure can have one
or more <keyedReference> structures that provide the name, value and <tModel>
UUID reference for locating more information.
For example, the following listing can be used to identify Springer publications using
their Dun & Bradstreet D-U-N-S Number and the corresponding tModelKey
within the UDDI Business Registry:
168
Figure 12. IdentierBag
<identifierBag>
<keyedReference
tModelKey="uddi:uddi.org:ubr:identifier:dnb.com:d-u-n-s"
keyName="SAP AG"
keyValue="31-626-8655"/>
</identifierBag>
However, as Figure 13 illustrates, more than one identier can be attached to the
entity.
Figure 13. IdentierBag-businessEntity
The businessEntity with the name Shipping R Us specied three identiers

in its identifierBag, D-U-N-S number, a Global Location Number, and a US
Tax Code identier. Any of the identiers can be used in a find_business call
in order to locate the businessEntity in the UDDI registry. The description for
find_business call is provided in the next section.
169
3. Programming UDDI
Three of the main API sets supported by UDDI specications are for inquiry,
publishing and subcription. The inquiry API locates information about a business,
the services a business offers, the specications of those services, and information
about what to do in a failuare situation. Any read operation from a UDDI registry uses
of one of the inquiry APIs message. The inquiry API does not require authenticated
access and is subsequently accessed using HTTP.
The publishing API is used to create, store or update information located in a
UDDI registry. All functions in this API require authenticated access to a UDDI
registry; the UDDI registry must have a logon identity, and the security credientials
for this identity must be passed as a parameter of the XML document for each
UDDI invocation. Because publishing requires authenticated access, it is accessed
over HTTPS, with a different URL than the one used with the inquiry access point.
The subscription API, which has been introduced with UDDI v3 as an optional
API allows consumers to register interest in the UDDI registry as a subscription.
Whenever there is change in the registry that matches the subscription criteria, the
register sends a notication to the client.
3.1. Searching with UDDI

UDDI denes an inquiry API to support the programmatic discovery of registry
data. The API primarily consists of nder and retriever methods. The nder methods
return result sets based upon general criteria, while the retriever methods return
detailed information about a specic registry entry. Table 2 lists the available nder
methods.
The nder methods (find_xx) are designed to return basic information about the
structures. Given the key to one of the major data structures, you can drill down into
the registry to get a full listing of the details in that structure.
The inquiry API provides a series of methods (get_xx) for retrieving information
from the registry. Table 3 provides a listing.
The inquiry API provides three patterns that can be used to query the registry: the
browse, drill-down and the invocation pattern.
170

Table 2. Finder Methods
Method Name
Return type
Description
Locates bindings within or across

one ore more registered business
services. The API call returns a
<bindingDetail> that contains
zero or more <bindingTemplate>
structures.
find_business
<businessList>
Locates information about one or
more businesses. The API call returns a <businessList> that contains zero or more <businessInfo>
structures.
find_relatedBusiness <relatedBusinessesList> Locates
information
about
businessEntity
registrations
that are related to a specic business
entity whose key is passed in the
inquiry.
find_service
<serviceList>
Locates one or more business services within registered business entities. The API call returns a
<serviceList> that contains zero
or more <serviceInfo> structures.
find_tModel
<tModelList>
Locates one or more tModel information structures. The API call
returns a <tModelList> that contains zero or more <tModelInfo>
structures.
find_binding
<bindingDetail>
3.1.1. Browse Pattern

The browse pattern characteristically involves starting with some broad information,
performing a search, nding general result sets and then selecting more specic
information for drill-down. The find_xx API methods listed in the above table are
used to support the browse pattern. For instance, you might want to nd out if
a particular business has any information registered (find_business). Once you
know the business name, you might want to know about the services offered by the
business (find_services). The find_xx API let callers start with a broad notion
of the kind of information they wish to retrieve from a registry, retrieve summary
information, and then drill down to get details.
171
Table 3. Retriever Methods

Method Name
Retrun type
Description
Returns the run-time binding-template information.

get_businessDetail <businessDetail>
Returns one or more complete businessentity objects. The API call returns a
<businessDetail> that contains a list of
<businessEntity> structure corresponding to
each of the businessKey values specied.
get_operationInfo <operationalInfos> Returns operational information pertaining to
one or more entities. The API call returns
an operationalInfo structure that contains
an operationalInfo element for each entity
requested by the inquirer.
get_serviceDetail <serviceDetail>
Returns a complete service object. The API call
returns a <serviceDetail> that contains a
list of <businessService> structures corresponding to each of the serviceKey values
specied.
get_tModelDetail
<tModelDetail>
Returns a complete tModel object. The API call
returns a <tModelDetail> that contains a list
of <tModel> structure corresponding to each of
the tModelKey values specied.
get_bindingDetail
<bindingDetail>
3.1.2. Drill-Down Pattern

Drill-down patterns involves getting details information using the results returned
by the find_xx operations. This is done using the retriever methods (get_xx) that
are listed in Table 3. Each instance of the core data structures businessEntity,
businessService, bindingTemplate and tModel has a key which is one of the
items in the summary information retrieved by find_xx APIs. Given such a key, it is
easy to retrieve the full registered details for the corresponding instance by passing the
key to the relevant get_xx API. For instance, if you are looking to nd information
about a business, you could call the find_business method which will return
a <businessList> structure. You can then use the businessKey (embedded
in businessList) to retrieve more information about the business using the
get_businessDetail method. Upon success, this API returns a businessDetail
containing the full registered information, including the businessEntity structure
for the entity whose key value was passed.
172
Most likely, you would use a tool with a graphical interface to achieve the above two
patterns. It should allow you to nd information on the basis of business, service or
tModel and then let you drill down. It should further let you drill deeper, locating
entities in the current context as your navigate through the registry.
3.1.3. Invocation Pattern

One of objectives with SOA applications is to have loose coupling between the
participating businesses. Prior to SOA, a typical development approach for developing
cross-application access was to dene the interfaces in two locations. CORBA, J2EE
among other distributed technologies also required the stub code to be available at
the client location. In the SOA paradigm, services can be discovered dynamically
and do not require stub code to be copied on the client location. Traditionally,
discovering and identifying details about the inter-business call has been a task that is
undertaken at the development time. The existence of UDDI registry entries makes
it signicantly easier to do dynamic binding using the invocation pattern. Here are
the steps that you could execute to invoke a remote service.
Using the above browse-and-drill-down patterns, you can browse the registry
and retrieve bindingTemplate about the service that is of interest. The bindingTemplate contains the specic details about an instance of a given interface
type, including the location at which a program starts interacting with the
service.
The calling application can then cache this information and uses it to contact
the service at the registered address whenever it needs to communicate.
If a call fails using cached, the client application could take the bindingKey,
call the get_bindingDetails API and get a new bindingTemplate.
If the new bindingTemplate returned is different from the cached information,
the application should retry the invocation using the fresh information.
If the result of this retry is successful, the new information can then update the
cached information.
By using the invocation pattern, the client application can locate new service
information about the partners without undue communication and coordination
costs. Caching also helps in reducing the unnecessary round trips to the UDDI
registry to locate services and binding information thus optimizing the service
environment.
173
3.2. Publishing with UDDI

UDDI denes a publishing API to support the programmatic publication of registry
data to a UDDI registry. All the publishing APIs are implemented as synchronous and
atomic from the point of view of the caller. As discussed previously in Section 2.3
(UDDI Keys), the publisher can specify keys at the time of publishing. If no keys
are specied, the registry generates a key at the time of registration. Unlike inquiry,
publishing a message requires authenticated access to the registry. Publishing message
requests use a different access points (HTTPS) than do the inquiry messages (HTTP).
When information is inserted into an operator node, that site becomes the owner
of that datas master copy. Any subsequent updates or changes to the data must be
performed at the same operator node.
This API consists of four sets of methods that are used for the following purposes:
1. Adding or updating registry entries;
2. Removing registry entries;
3. Managing publisherAssertions; and
4. Retrieving the status of published entries.
Table 4 lists the methods used to add or update registry entries, whereas Table 5 lists
the methods used to remove registry entries. Upon completion, all these methods
return a dispositionReport structure with a single success indicator.
Table 6 lists the methods used to modify publisher assertions and Table 7 lists the
methods used to retrieve status information about registry data.
The publication API supports a comprehensive set of methods to create, update
and delete entries from the UDDI registry. There are also a good number of tools
(e.g. Systinet) in the market that provides a web interface to achieve most of these
functions.
3.3. Subscribing with UDDI

Subscription provides clients, known as subscribers, with the ability to register
their interest in receiving information concerning changes made in a UDDI
registry. The subscription API allows the monitoring of businessEntity,
businessService, bindingTemplate, tModel, related businessEntity and
174

Table 4. Adding or Updating Registry Entries
Method Name
Return type
Description
Inserts or updates a UDDI registry with the

<bindingTemplate> documents passed as input.
The method returns a <bindingDetail> structure
containing the results of the call.
save_business <businessDetail> Inserts or updates a UDDI registry with the
<businessEntity> documents passed as input. This
API has the broadest scope of all of the save_xx
API calls and can be used to control the full set of
information about the entire business, including its
businessService and bindingTemplate structures.
It returns a <businessDetail> structure containing the
nal results of the call.
save_service <serviceDetail> Inserts of updates a UDDI registry with the
<businessService> documents passed as input. This
API can modify <businessService> and any references to <bindingTemplate> structures. It returns a
<serviceDetail> structure containing the nal results
of the call.
save_tModel
<tModelDetail>
Inserts or updates a UDDI registry with the <tModel>
documents passed as input. It returns a <tModelDetail>
structure containing the nal results of the call.
save_binding
<bindingDetail>
Table 5. Removing Registry Entries

Method Name
Return Type
Description
<dispositionReport> Removes existing bindingTemplate entry from

the UDDI registry.
delete_business <dispositionReport> Removes existing businessEntity entry from
the UDDI registry. Deleting these documents
also causes the deletion of any contained
<businessService> or <bindingTemplate>
data.
delete_service <dispositionReport> Removes existing businessService entry from
the UDDI registry.
delete_tModel
<dispositionReport> Logically removes existing tModel entry from the
UDDI registry by marking them as hidden. The
documents are not actually destroyed. Hidden
<tModel> documents are still accessible, via the
get_registeredInfo and get_tModelDetail
APIs, but are omitted from any results returned by
calls to find_tModel.
delete_binding
175
Table 6. Modifying Publisher Assertions

Method Name
Return Type
Description
Adds a new relationship assertion to the current set of assertions. A publisher assertion creates
an association between two businesses. The relationship becomes
publicly visible only when both
the businesses have added matching <publisherAssertion> documents to their collection.
delete_publisherAssertions <dispositionReport>
Removes
a
specic
publisherAssertion. Deleting
an assertion causes any relationships
based on that assertion to become
incomplete.
set_publisherAssertions
<publisherAssertions> Saves a new complete set of
assertions for a publisher, completely replacing any previous
assertions. The call returns a
<publisherAssertions> document that contains the current
collection as they are stored in the
registry.
add_publisherAssertions
<dispositionReport>
Table 7. Retrieving Status Data

Method Name
Return Type
Description
get_assertionStatusReport <assertionStatusReport> Retrieves a report identifying all

registered assertions and their current status for the requesting
publisher.
get_publisherAssertions
<publisherAssertions>
Retrieves a list of all assertions for
a particular publisher.
get_registeredInfo
<registeredInfo>
Retrieves an abbreviated list
of <businessEntity> and
<tModel> documents currently
managed by a given publisher.
publisherAssertion. The subscribers register their interest and are notied if

there is a new, updated or deleted entry for any of the above entities.
176
Let us illustrate the use of the above API by imagining a company, DummyMortgages,
which provides a brokerage solution for home mortgages. Consumers or Mortgage
buyers provide basic information about the type of loan required, and DummyMortgages supplies the best interest rate as available from its partners. DummyMortgages
deals with many banks and nancial institutions on a routine basis as increasing
its partners network and providing the best interest rate for its customers is critical
to its business growth. In order to integrate with its system, DummyMortgages
asks its partners to implement a mortgage query service. The service would take
the amount, and type of loan as the arguments and return the available interest
rate. DummyMortgages publishes the interface denition of the expected service
at its partners website so that a potential partner can implement the service and
register it with the companys UDDI service. On a periodic basis, DummyMortgages
business team queries the UDDI registry for a list of services that have implemented
the mortgage query interface. If the list returns a new partner, the business team
adds the name to its existing directory of partners. And from this point on, any
consumer request for a loan will also be forwarded to this newly founded partner. The
subscription API allows DummyMortgages to register its interest with the UDDI
registry. Figure 14 illustrates the scenario.
Figure 14. Subscribing with UDDI

Whenever a new partner publishes its service with the UDDI registry or an
existing partner make changes to its registration, the registry will send notications
to DummyMortgages. This not only eliminates the need for DummyMortgages
business team to keep polling the UDDI registry for new partners but it also speeds
up the process of integrating new partners services with the Dummy Mortgages
interest query service.
The business team would normally invoke the find_service method to retrieve
the services that are compliant with the interest query tModel. The subscription
API allows the query to be saved in the UDDI registry along with other parameters
177
on how and when to notify the DummyMortgages business team. It also allows
specifying the length of time that the company is interested in monitoring the registry
changes. The query and preferences are stored in the subscription structure that
is shown in Figure 15. The subscription API supports two monitoring patterns.
Figure 15. Subscription Schema
3.3.1. Asynchronous Notication

This pattern is also referred to as subscriber listener since it enables subscribers to
inform the UDDI node that they wish to be directly notied when subscribed registry
data changes. The subscription data structure allows registering a service or an email
in the binding key element. Any time this is a change, the register either invokes
the service or send an email notication about the change. Thus to be automatically
notied of the new partners that have implemented the companys interest query
service, the DummyMortgages business team will save a subscription that uses the
find_service call with the interest query services tModel key as a parameter.
3.3.2. Synchronous Notication

This pattern is also referred as change tracking. It enables subscribing clients to
issue a synchronous request to retrieve registry changes that match their subscription
preferences. To support that synchronous subscription mode, you leave the subscriptions binding key empty and invoke get_subscriptionResults method in a
synchronous manner and it returns the same information what the registry would
send in a notication.
178
Figure 16. Synchronous vs Asynchronous
Note that the subscription API is optional and the UDDI registry is not required to
support it. Table 8 lists the methods dened by the specication.
Table 8. Subscription API

Method Name
Return Type
Description
Deletes one or more subscriptions.

get_subscriptionResults
<subscriptionResultsList> Synchronously returns registry data related to a particular subscription.
get_subscriptions
<subscriptions>
Returns current list of subscriptions associated with the
subscriber.
notify_subscriptionListener <dispostionReport>
Receives asynchronous notications of changes to
data that the subscriber is
monitoring. The method is
optional for client.
save_subscription
<subscriptions>
Registers a new subscription,
or modies or renews an
existing one.
delete_subscription
<DispositionReport>
179
4. Internationalization
The U in UDDI stands for Universal and the one of the goals is to provide a registry
solution for Universal description, discovery and integration of business entities and
their services. The UDDI registry design includes support for internationalization
features. Most of these internationalization features are directly exposed to end users
through the API sets. Others are built into the design in order to enable the use of
the UDDI registry as an international services discovery and description mechanism
with multilingual descriptions of business entities worldwide. Version 3 of the UDDI
specications supports the ability to classify entities using multiple languages and
multiple scripts of the same language. It also allows for additional language-specic
sort orders and provides for consistent search results in a language-independent
manner.
This section discusses the internationalization features supported by the UDDI
specications:
multilingual descriptions, names and addresses;
multiple names in the same language;
internationalized address format; and
language-dependent collation.
4.1. Multilingual Descriptions, Names

and Addresses
UDDI registry allows names, descriptions and address elements to be specied in
other languages that may have characters from language scripts other than the Latic
script found in ASCII. For instance, if you have a shipping company that also provides
services in China, you would want to register your business in that language too. The
xml:lang attribute provides the solution.
<businessEntity ...>
........
<name xml:lang="en">ShippingRUs</name>
<name xml:lang="zh">?????</name>
.....
</businessEntity>
180
Similarly, variants of names, due to transliteration, e.g. romanization, to different

languages can also be indicated through the use of the xml:lang attribute. The
following shows an example of romanization where the primary name of the business
(a Chinese ower shop) is in Chinese, and its alternative name is a romanization
........
<name xml:lang="zh">?????</name>
<name xml:lang="en">Rus Yun Shu Gong Si</name>
.....
</businessEntity>
The following sample XML fragment shows an address written in two languages,
English and Chinese, as indicated by the xml:lang attribute:
<address useType="Sales office" xml:lang="en"
tModelKey="uddi:...">
<addressLine>7 F</addressLine>
<addressLine>No. 245</addressLine>
<addressLine>Sec. 1</addressLine>
<addressLine>Tunhua South Road</addressLine>
<addressLine>Taipei</addressLine>
</address>
</address><address useType="Sales office" xml:lang="zh"
tModelKey="uddi:...">
<addressLine>???</addressLine>
<addressLine>????</addressLine>
<addressLine>??</addressLine>
<addressLine>245</addressLine>
<addressLine>7</addressLine>
...
</address>
4.2. Multiple Names in the Same

Language
The registry also supports to publish muliple names in the same language. This is
useful for multi-script languages. An example is Kanji and Katakana for Japanese
where the registry allows publishing the name in both scripts for the same entity. It is
also useful if you want to publish an acronym for your business. The following shows
an example of use of multiple name elements to support a multi-script language
181
and also the use of acronym. In the example, the rst <name> element is the
primary name of the business (a Japanese ower shop) in Japanese Kanji. The second
<name> element is the business name transliterated into Japanese Katakana. The
third <name> element gives the business full English name, and the fourth <name>
element gives its English acronym:
........
<name xml:lang="ja">???):</name>
<name xml:lang="ja">......</name>
<name xml:lang="en">ShippingRUs</name>
<name xml:lang="en">SRU</name>
.....
</businessEntity>
Where multiple name elements are published, the rst name element is treated as the
primary name, which is the name by which a business would be searched and sorted
in case the business has multiple names.
4.3. Internationalized Address Format

Different parts of the world have their own ways of specifying their postal addresses
using different elements such as lot numbers, oor numbers, city, street and so
forth. In UDDI, the address is supported by the <address> element that is
part of the businessEntity data structure. The address element contains a list
of <addressLine> elements. UDDI Business Registry has a canonical tModel,
ubr-uddi-org:postalAddress, that identies a canonical postal address structure
with common address sub-elements (e.g. states, cities). This canonical address
structure describes address data via name/code pairs, enabling each common address
sub-element to be identied by name or code. The following XML fragment illustrates
a way to specify an address
<address useType="Sales office"
tModelKey="uddi:uddi.org:ubr:postaladdress">
<addressLine keyName="Street"
keyValue="40">SpringWell Road</addressLine>
<addressLine keyName="House number"
keyValue="70">181</addressLine>
...
<addressLine keyName="Country"
keyValue="30">United Kingdom</addressLine>
</address>
182
With the use of the keyName/KeyValue pair together with the codes assigned in the
ubr-uddi-org:postalAddress tModel, you can programmatically determine
the address semantics and evaluate them even if the sub-elements are specied in
different sequence or language.
Since there is large variation in address sub-elements of different countries, the address
structure also supports free form address lines.
4.4. Language-Dependent Collation

Each supported language within the registry is based on the Unicode 3.0 specication
and ISO 10646, which support the majority of languages. Each language has its own
unique behavior when it comes to sort-order collation. For instance, the languages
that share the same alphabetic script, such as English, Spanish and French, the letters
have different collation weights depending on the other languages with which they
are used. For languages that have both upper and lowercase letters, sorting depends
on whether sorting is specied as case-sensitive. The UDDI specications allow the
collation sequence of results returned by the Inquiry APIs to be specied via nd
qualiers. For instance, the following snippet can be used to sort the entities by name
in ascending order using the JIS X 4061 Japanese collation sequence.
<find\_business xmlns="urn:uddi-org:api\_v3">
...
<findQualifiers>
<findQualifier>
uddi:uddi.org:sortorder:jis-4061
</findQualifier>
<findQualifier>
uddi:uddi.org:findqualifier:sortbynameasc
</findQualifier>
</findQualifiers>
...
</find\_business>
4.5. Federation of Registries

The UDDI Version 3.0 specication addresses the complexity of registry topology,
and denes how registries may form a federation. A key aspect of such federations
is the mechanism by which a registry entity may be promoted from one registry to
183
another and how global registry key uniqueness is maintained. In the earlier versions
of UDDI specications, only the UDDI node could generate keys and a publisher
was not allowed to pre-assign the keys. In order to maintain the uniqueness of keys, a
publisher could not import or export a UDDI registry entity in its entirety from one
registry to another.
As we discussed earlier in this chapter, Version 3 of UDDI approaches the issue of
key generation in a signicantly different fashion and it allows the publisher to use its
own keys while registering an entity. It therefore allows copying an entity from one
registry to another while preserving the key. This behavior is called entity promotion
and it permits data sharing between UDDI registries and it establishes UDDI as
a more distributed registry solution. The publisher could potentially publish the
entirety of a registrys contents into another registry, effectively mirroring the data.
Or, the publisher might be interested in only a subset of data from another registry
and only copy a portion of that data. However, while copying the entities across the
registries, it is still critical to avoid key collisions. The recommended way to prevent
such collisions is to establish a root registry, a registry that acts as the authority for
key spaces. A root registry serves to delegate key partitions such that other registries
can rely upon the root registry for verication and validation of a given key partition.
All other registries that interact with the root registry are called afliate registries. By
relying on a common root registry as an arbitrator of key spaces, afliate registries can
share data with both the root registry and among one another with the knowledge
that a given partition is unique.
An important example of a root registry is the UDDI Business Registry, which
has a set of policies in place to generate unique uuidKeys as well as to validate
domainKeys through signatures that correlate with DNS records. By acknowledging
the UDDI Business Registry as a root, an afliate registry can establish inter-registry
communication policies and procedures with both the UDDI Business Registry and
any other registry, which is an afliate of the UDDI Business Registry. Following this
approach, registries could form hierarchical relationships where they can share data
with their parents and afliate registries and form a federation.
The two scenarios below illustrate the use of inter-registry communication in real
business situations.
4.6. Private Test Registry

Applications based on SOA follows the same laws of software development life cycle.
You develop the application in a development environment, unit test it and promotes
it to a test or staging environment. Once the system test and UAT is done, the
184
Figure 17. Federation

application is promoted to a production environment. Your registry should follow
the same route as your service while progressing its way from a test registry to a
production registry. In order to be close to the real-world production environment, it
is desirable that the modications required while switching from test to production
mode are kept minimal. It is also critical that the development/test versions of
software must not interfere with actual production systems. Because services can be
highly distributed and are loosely coupled maintaining this distinction is paramount
to ensure that dependencies are managed systematically. You would not want two
hundred thousands copies of the same book delivered to your door step while load
testing your book ordering application in test environment which by mistake was
pointing to the production service.
The entity promotion facilitates the setup of a private test registry and the upgrade
from test to production environment. When development versions of software have
been fully tested and certied in the test environment, the registry entry in its entirety
can be copied to the production sphere.
A number of companies are following the practice of setting up a private test registry
for internal development and testing and then using standard operations procedures
to migrate it to the production environment. Figure 18 illustrates the scenario.
4.7. Shared Registry

The enhanced security and federation of the UDDI registries also allows setting up
shared registries that support collaboration among partners. The shared registries
185
Figure 18. Private Registry

might reside outside a corporate rewall but with restricted access only to clients of
trusted business partners. For instance, DummyMortgages from our earlier example
can set up a shared UDDI registry and invite its partners to register relevant Web
service implementations in it. Let us consider another scenario where a manufacturing
business purchases raw material from its partners, fabricate it and sell it to its
consumers. Being in the middle of the supply chain, it prots relies on its integration
with its suppliers and consumers and the efcient management of its inventory
system. Due to the disparate systems of its partners, SOA appears as the natural
choice. In order to avoid separate integration and coordination efforts with each of
its partners, it sets up shared partner registries where its partners can implement their
services. In addition, by establishing subscription-based relationships with partners
registries, the company can ensure that the information is up-to-date and is processed
in the most efcient manner. The registries can be deployed inside the DMZ trusted
environment to limit the exposure of the registry. Figure 19 illustrates the scenario.
Such shared UDDI registries can either be hosted by companies or by industryspecic organizations or consortia. These shared registries will restrict access to its
members and facilitate an electronic market place. For instance, all the automakers
can setup a registry to unify their purchasing power on commodity parts. There
are already some efforts underway to this effect. For instance, the Star Standard
(http://www.starstandard.org/) is an auto-industry-wide initiative to dene a standard
way for automotive dealerships to communicate with car manufacturers; RosettaNet
(http://www.rosettanet.org) is a consortium of major Information technology and
186
Figure 19. Shared Registry

manufacturing companies to dene open e-business process standards for supply
chain partners on a global basis. OpenTravel (http://www.opentravel.org) is a similar
effort by major airlines, hoteliers, car rental companies, and travel agencies.
4.8. Security
The security model for a UDDI registry can be characterized by the collection of
registry and node policies and the implementation of these policies by a UDDI
node. The principal areas of security policies and mechanisms in the UDDI
specication are related to data management, user identication, user authentication,
user authorization, condentiality of messages and integrity of data.
UDDI v3 also supports XML Digital Signatures on UDDI data. XML digital
signature is discussed in depth in the next chapter on Security. In the context of
UDDI, it enables consumers to verify the integrity of the data with respect to the
publisher. Publishers of entities can now ensure that some malicious party who claims
187
to own the entity does not misrepresent them. Furthermore, once a publisher signs
the data, altering it in any way breaks the signature, providing condence in the datas
integrity. By verifying the signature, consumers using the registry can also be assured
that a signed entity is valid and that the publisher represented by the signature created
it. Consumers can also issue queries that only return results pertaining to signed
data. The reliability of data is even more critical in the multi-registry environment.
When signed data is copied between registries, you can guarantee its integrity by
simply validating the signature. The use of digital signatures improves both the
quality of data in UDDI and provides users with the trust and protection needed for
service-oriented applications.
5. Summary
A registry solution is all about sharing business information, making it easier to publish
your preferred means of doing business, nding trading partners and have them nd
you, and interoperate with these trading partners over the internet. It removes the
barriers to rapid participation in the global economy and allows businesses to fully
participate in the new digital economy. The automated application-to-application
discovery and integration over the Internet helps eliminate many of the conguration
and compatibility problems that are preventing businesses from more widely adopting
B2B, despite B2Bs potential for cost savings and improved efciency. However, just
like the early days of web, the registries have a Catch 22 situation. Users and businesses
will not search the registries if there are not many services listed in the registry. With
the scarcity of potential consumers, the providers will be reluctant to put the extra
effort required to publish a service in the registry. The initial idea of the architects
was to have a global registry where everybody can publish their services, but this has
changed and the industry is now seeing many private and shared registries being used
within the partner networks.
However, there are still a lot of places, where in spite of the service enablement of
the business functions, the requestor and the provider binds directly to each other.
True ESOA is only achieved when your organization accomplishes the dynamic
interoperability between services. When the provider can announce a service and
does not have to communicate individually with all the users. And a business analyst
can select a service based on the functionality without worrying about the underlying
infrastructure. There are a few key points that strongly support the value of a registry
solution in an ESOA environment:
Visibility: Without a registry solution, the business user or client cannot nd
what is available. You cannot use what you cannot nd. The architects and
188

the development teams can build a comprehensive set of services but without
a standard mechanism to communicate these to the potential consumers, it is
worthless. A registry works as the central point of reference for the distributed
SOA assets built and deployed by an enterprise.
Re-usability: It is one of the keys to the success of ESOA. The extra investment
in enabling your business functions as services pays off, when it helps in avoiding
the duplicate effort within the enterprise. However, without a registry solution,
participants involved in creating, managing and consuming services will not
know what others are planning and building and this will result in redundant,
conicting and overlapping services within the enterprise.
Congurability and Adaptability: Registry provides a layer of abstraction
between the consumer and the service provider. It allows for the modications
in the business requirements, policies or needs of the consumer as well as the
implementation changes of the provider without affecting each other. It allows
enterprises to assert new policies that dene specic security, process, semantic,
and governance constraints by which consumers can bind to specic business
services, and for clients to adapt without having to redeploy applications.
The APIs provided by the registry solutions allows random searching for businesses.
It is conceivable that software will eventually discover a service dynamically and use
it without requiring human interaction. In the near future, the business analysts
with specic knowledge of the problem at hand will use UDDI portals to discover
potentially interesting services and partners, and technologists will write programs to
use the services from companies that have already been discovered. UDDI portals
and SOA management solutions are coming up fast in the market that provides
user friendly interfaces and not only allows customized searches for business but also
provides support to update information in a registry.
5
UNDERSTANDING
ENTERPRISE SECURITY
The difculties you meet will resolve themselves as you advance. Proceed,
and light will dawn and shine with increasing clearness on your path.
Jim Rohn, Author of The Art of Exceptional Living
A key benet of the service-oriented architectures is the ability to deliver integrated,

interoperable solutions. Ensuring the integrity, condentiality and security of these
services through the application of a comprehensive security model is critical, both
for organizations and their customers. Security is considered one of the most critical
aspect of an enterprise system. According to Zapthink, Security is the immediate
roadblock facing widespread implementation of service oriented integrations across
the enterprise. The advent of N-tier systems, enterprise application integrations
and service-oriented applications has resulted in a more complex and fragmented
application layer. Due to the distributed nature of the current enterprise systems, we
have difculty in administering security policies and bridging diverse security models.
This leads to increased opportunities to make mistakes and leave security holes, hence
the chance of accidental disclosure and the vulnerability to attack goes up.
The security infrastructure in the traditional distributed computing world resembles
islands of security. Each component or network acts as an island, with its own
perimeter security, and only users within the network are considered to be trusted.
The arrival of Enterprise SOA breaks this trusted vs untrusted dichotomy. The
promise of ESOA is to be able to locate and request a service on a different network
or enterprise in a similar manner as if it is co-located.
The easy and historical approach to handle the security issue has been to put the
responsibility on application developers. They have to discover the security policies
189
190
of upstream providers and downstream requesters. Then they have to determine

appropriate policies for the application layer and implement security infrastructure
along side business code. This tactic of course results in intermingled code with
business logic and security tightly coupled and a maintenance nightmare on hand. The
other approach is to put the responsibility on security administrators. They have to
understand the detailed security logistics of the different components, intermediaries
and middleware infrastructure. However, without the proper knowledge and control
of the security needs of other systems and to avoid the conicts and interoperability
concerns, this results in conservatively congured components.
The fundamental sources of these problems are twofold. First, middleware paradigms
assume that the application is the center of the world. They often do not fully
acknowledge the rest of the enterprise security ecology, which includes rewalls,
directory servers, web servers, authentication providers, and databases. Second,
the built-in security models of most middleware paradigms do not fully support
the dynamic policies necessary to meet the realities of modern business processes.
Evaluating who owns an account or the strength of encryption used on a connection
is cumbersome if not impossible. The advent of SOA systems has extrapolated these
security issues.
The solution lies in a change of perspective middleware paradigms own the
business logic but co-operate on security. Therefore, they should gather as much
information from the rest of the security ecology as possible and enable the security
administrators to evaluate this information before it ever reaches the business logic.
However, in order to be interoperable and for administrators to understand this
information, there is a need of mutually agreeable standards. For systems based on
service-oriented architectures to be successful, there is a need of a security solution
that is standards-based but powerful, exible and extensible. It should be part of the
application infrastructure and still separate from the application.
We will start this chapter explaining some of the prevalent security concepts. These
concepts are valid whether it is a J2EE world or a .NET application, an online sale
or a trip to your local mall. In the sections that follow, we will examine each of these
concepts and how could be handled in an enterprise system. We will also look at the
common specications that are prevalent in the ESOA space. But before that, let us
rst discuss the need for a security solution where the security is built within the
message.
Chapter 5: Understanding Enterprise Security
191
1. Need for a Message Level

Security Solution
Enterprise systems already have transport layer security mechanisms such as SSL/TLS
and IPSec. Most of the applications, whether they are built on J2EE or .NET, also
implement their own security framewok to block against outside attacks. Why this is
not enough? The reason lies in the distributed and heterogeneous nature of the SOA
systems. SOA enables application topologies that include a broad combination of
mobile devices, gateways, proxies, middlewares, outsourced data centers, and globally
distributed, dynamically congured systems. Many of the bigger problems involve
sending the message along a path more complicated than request/response or over a
transport that does not involve HTTP. In an SOA eco system, there are more than
one endpoints that span the physical and application boundaries. Your application is
only as secure as the weakest link (endpoint or intermediary) in the message ow. In
the next few sections, we will discuss the issues that surfaces due to the distributed,
heterogeneous nature of the SOA-based applications.
1.1. Point-to-Point vs End-to-End Security

Secure Socket Layer (SSL) along with the de facto Transport Layer Security (TLS)
provides sufcient security for point-to-point systems. SSL/TLS offers several security
features including authentication, data integrity and data condentiality.
Figure 1. Point-to-Point Security

However, in service-oriented architecture a message travels from the originator to the
ultimate destination, potentially by passing through a set of intermediaries along the
message path. Intermediary is an application that is capable of both receiving and
forwarding messages. When an intermediary receives a message, it processes header
entries intended for it and must remove them before forwarding the message. It may
also insert a similar header entry intended for another intermediary. The actor
192
attribute of a header entry is used to indicate the URI of an intermediary who will
process the header entry.
Figure 2. End-to-End Security

Transport level security ensures the integrity of data between the two end points.
But it does not ensure data security within the intermediarys system. It forces any
upstream message processors to rely on the security evaluations made by previous
intermediaries and to completely trust their handling of the content of messages.
Secure transport protocols can assure the security of messages during transmission.
However, since messages are received and processed by intermediaries, secure endto-end communication is not possible if these intermediaries are not completely
trusted even though the communication links between them are trusted. End-to-end
security is also compromised if any one of the communication links is not secured.
A comprehensive end-to-end security architecture requires the security context to be
propagated along with the data to all the listed actors.
The identity, integrity, and security of the message and the caller need to be preserved
over multiple hops. More than one encryption key may be used along the route.
Trust domains will be crossed. HTTP and its security mechanisms only address
point-to-point security. More complex solutions need end-to-end security baked in.
The solution needs to maintain a secure context over a multi-point message path.
1.2. Application Independence

End-to-end security ultimately needs to be achieved at the application level. This
is because if there is any point between the communicating parties where messages
are in plain text, it can be a potential point of attack. However, it is not an easy
task to integrate cryptographic functionality into an application without introducing
additional security vulnerabilities. Most of the available commercial cryptographic
libraries are extremely exible to meet many different levels of security requirements,
and using them properly may require good understanding of cryptographic technologies. A standardized, application-independent security layer will provide good
193
protection without worrying about cryptographic details. You should be able to

change the security policies without really modifying the application. Consider the
analogy to a secure physical ling system. When the security policy changes, you do
not rewrite the document. You put it in a different ling cabinet. Or you change the
key of the cabinet and a security ofcer controls the distribution of keys. Similarly,
application developers should not have to change business logic when security policy
changes. A security administrator should simply alter the protection given to related
components. The other reason to have a separate security layer is to be able to
handle security breaches more efciently. It is a lot harder to test the security of every
application component individually than a security system as a whole. For instance,
to validate the adherence to the dened security policy for the ling system, you do
not read all the documents in all the cabinets. You check the integrity of the locks. It is
thus desirable, in most cases, to have security functionality as close to the application
level as possible but not built into the application itself.
1.3. Technology Independence

Beyond doubt, every company in the ESOA space has a dream of standardizing
each and every application, endpoint, middlewares to their platform. It will
denitely make things simple but it is not going to happen. As the scope of
your application increases, so does the count of technological variations. There
is too much to administer, too many applications, too many variations, and too
rapid a pace of technology change to design a single infrastructure to meet all
requirements effectively. Extensible standards are required that can adapt to changing
requirements, that can incorporate new technologies while continuing to work with
legacy technologies, and that can be deployed modularly as needed without requiring
use of unnecessary portions. The security solution should be able deal with different
languages, cryptographic mechanisms, hashing algorithms among other things. In a
computers world abstraction has an answer for everything. The security solution
should be exible enough to abstract the security techniques used by a given system
and have a mechanism to exchange it with other partners that are using a different
standard.
2. Security Concepts
Technologies designed to meet the security requirements have evolved and have come
a long way. However, the requirements have remained relatively constant. The risks
of a system without a security infrastructure have not changed:
194
Tampering An unauthorized person changing the information in the transit.

For instance, someone could change the shipping address of an order placed
online.
Eavesdropping The data remains intact, but privacy of the data is compromised.
For instance, someone could learn the credit card number used during an online
order placement.
Impersonation Also referred to as identity theft where a person can pretend
be someone else. Privacy laws of corporations both help and hurt this cause.
For instance, with the number of free email providers available, it does not take
much effort to get an email id with the words bill gates in it and start posting
messages on the newsgroups.
In order to handle the risks above, there are ve basic elements that are required for a
secure application.
2.1. Authentication Who Is It?

Determine the identity or role of a party attempting to perform some action such as
accessing a resource or participating in a transaction. The recipient of the message
must be able to conrm the identity of the sender of the message. Just like any online
shopping site asks the user to login before allowing him to place an order, most
service providers requires that consumers are authenticated before processing the
service request. Security tokens either in the form of username/password or binary
tokes (e.g. X.509 certicates) can be used to establish the users identity. WS-Security
allows the security tokens to be included in the SOAP header.
2.2. Authorization What Can They Do?

Determine whether some party is allowed to perform a requested action, such as
accessing a resource, making a transaction, viewing a web page or changing a password.
Authorization is a means of ensuring that only authorized users are allowed to access
the resources within a system. All authorization inquiries have the same general form:
Can {Principal X} perform {Action Y} on {Resource Z}. The authorization queries
can be answered by dening the authorization policies or rules. There are several
ways to capture the above rules, for instance access control list (ACL) or role-based
access. Later in this chapter, we will look at Extensible Access Control Markup
Language (XACML) which is getting the maximum traction. XACML provides the
XML Schema to capture the rules that specify the who, what, when, and how of
information access.
195
2.3. Integrity Ensure That Information

Is Intact
Ensure that information is not changed, either due to malicious intent or by accident.
The recipient of a message must be able to guarantee that a message has not
been tampered with in transit. Remember, in a SOA system there may be multiple
endpoints and intermediaries before the intended recipient gets the message. Digitally
signing the message ensures the data integrity of the message as any data modications
during the message transit are detected in signature verication at the receiving end.
XML Signature specications, discussed later in the chapter, cover the syntax and
processing of digitally signing the message.
2.4. Condentiality You Cant Read

Ensure that content may only be viewed by legitimate parties and the intruder of
transmitted message can not read the data. For instance, consider if the message
contains the credit card number of a customer. You not only need to verify that
the data is not altered, but also ensure that the credit card information can only be
read by the intended party. This basically involves encryption of the message using
a mechanism such as SSL to ensure condentiality. The sender authenticates the
receiver and negotiates on the scheme of encryption with the receiver. The receiver has
the option to authenticate the sender. The sender and receiver establish a connection
exchanging the encrypted messages. XML Encryption specications dene the syntax
and processing rules for encrypting and decrypting selected elements in an XML
document. Since encryption/decryption algorithms are computationally expensive, it
is required to be able to encrypt only the selected portions of the message.
2.5. Non-Repudiation You Sent It,

I Got Proof
The sender and the recipient of the message must be able to guarantee that the sender
sent and the recipient received the message. It provides protection against false denial
of involvement in a communication. For instance, in an online stock trading system,
the broker must be able to verify that the given transaction was requested. XML
Signature requires the sender to sign the document using its key and thus verify the
senders identity. Logging and auditing is another way to address non-repudiation
claims.
196
Not exactly an absolute requirement but other useful features in the security
infrastructure are given below.
2.6. Single Signon How Many Times Do I

Have to Tell You?
SSO allows users to gain access to resources in multiple domains without having to
re-authenticate after initially logging in the rst domain.
2.7. Key Management Give Me a Key

Chain
With the increased amount of data and the increasing number of parties involved,
each signing and encrypting different elements using different keys, there is a need of
key management solution so that the creation and location of the keys is outside the
main application realm.
3. Security Technologies
With the growing acceptance of XML as de facto language for documents and
protocols, it is only logical that security should be integrated with XML solutions.
There are a number of XML Security standards to handle the security needs of
service-oriented applications. These standards use legacy cryptographic and security
technologies, as well as emerging XML technologies, to provide a exible, extensible
and practical solution toward meeting security requirements.
Figure 3 displays the building blocks required for the security framework.
The following items will be discussed in this chapter.
Security Tokens.
XML Digital Signature for integrity, Nonrepudiation and signing solutions.
XML Encryption for condentiality.
XML Key Management (XKMS) for public key registration, location and
validation.
197
Figure 3. Blocks for Security Framework

Security Assertion Markup Language (SAML) for conveying authentication,
authorization and attribute assertions
XML Access Control Markup Language (XACML) for dening access control
rules.
WS-Security and others Relies on the above specications to provide a
comprehensive solution.
Policy and Federation Layer.
3.1. Authenticaton and Security Tokens

Before a service provider allows the requestor any access to its resource, it needs to
verify the credentials of the requesting party and verify its identity. For instance, if you
go to a bank you need to show your drivers license or other proof of identity before
you can withdraw money from your account. Or if you are in a different country,
198
you need a passport to prove your identity. Both the drivers license and passport
are credentials to vouch for your identity. However these documents do not perform
authentication. Authentication is performed by a person based on these credentials.
The bank teller or the immigration ofcer will verify the picture, signature or other
attributes of these documents and validate that you are the rightful owner. In the
software world, security tokens serves the purpose of drivers license or passport. Like
the license bureau or passport authority, there are security token service providers that
can issue security tokens that can be used by applications. The service provider will
accept these tokens as part of the request and perform authentication before serving
the request.
The requestor can send a security token along with the data as a proof if its identity.
Security token is dened as a representation of security-related information (e.g.
X.509 certicate, Kerberos tickets and authenticators, username, etc.). A security
token is considered signed that contains a set of related claims cryptographically
endorsed by an issuer. Examples of signed security tokens include X.509 certicates
and Kerberos tickets. Username/password is considered unsigned security tokens.
The requestor and provider can either have their own implementations to support
the security token or they can rely on an external security token service. Figure 4
illustrates a common message ow.
Figure 4. Security Token Service
3.1.1. Username/Password
The most common ways to pass around credentials is to use a username and password.
The password can be sent along with the message in plain text or as a digest hash.
The other option is to communicate the password in advance to avoid sending it
with the message.
199
3.1.2. PKI through X.509 Certicates

Another option to use when authenticating uses is to simply send around an X.509
certicate. Using PKI, you can map the certicate to an existing user in your
application. When a message sends along an X.509 certifcate, it will pass the public
version of the certicate along the message. Anyone with the public version of the
certicate can validate the identity of the sender.
3.1.3. Kerberos
To use Kerberos, a user presents a set of credentials such as username/password or
an X.509 certcate. The security system them grants the user a ticket granting ticket
(TGT). The TGT is an opaque piece of data that the user cannot read but must
present in order to access other resources. The user will typically present the TGT in
order to get a service ticket (ST). The way the system works is as follows:
1. A client authenticates to a Key Distribution Center (KDC) and is granted a
TGT.
2. The client takes the TGT and uses it to access a Ticket Granting Service (TGS)
3. The client requests an ST for a particular network resource. The TGS then issues
the ST to the client.
4. The client presents the ST to the network resource and begins accessing the
resource with the permissions the ST indicated.
3.2. Integrity and Signing

Integrity means that the message has not been altered in the transit. If the message is
signed, the receiver of the message knows that the signed elements have not changed
en route. A signature is based on a digest created according to the content of the
document that needs to be digitally signed. It veries to the receiver:
The user identied by the X.509 certicate, UsernameToken, or Kerberos ticket
signed the message.
The message has not been tempered with since it was signed.
Before we go in the details of signing in the SOA perspective and XML Signature, let
us quickly review the basics of digital signature. The rst step in the process is to use
200
a one-way hash mechanism to generate a message digest for the actual content. Once
created, it is not possible to change a message digest back into the original data from
which it was created. The second step is to encrypt the message digest and create a
signature. While sending the data, the sender appends the signature with the message
data.
Figure 5. Digital Signature

The receiver decrypts the signature changing it back to a message digest. The next
step is to hash the message data to generate a message digest in a similar fashion as
the sender. If the message digest created by the above two steps are same, it ascertains
that the data has not been altered in the transit.
All three of the authentication mechanisms mentioned in the previous section on
security tokens provides a way to sign the message. X.509 allows the sender to sign
the message using the private key. Kerberos provides a session key that the sender
creates and transmits in the ticket. Only the intended receiver of that message can
201
Figure 6. Verication of Digital Signature

read the ticket, discover the session key, and verify the authenticity of the signature.
XML Signature, a W3C standard allows a mechanism for digitally signing XML
documents using any of the above methods. Rather than reinvent what already exists,
SOA implementations simply points to this existing specication and lls in a few
details on how to use it with SOAP messages.
3.3. XML Signature

The project aims to develop XML syntax for representing digital signatures over any
data type. The specication also denes procedures for computing and verifying such
signatures. Like most of the other specications in this space, XML Signature takes
a very exible approach. It allows you to sign any type of data and not just XML
data and provides means to support diverse set of internet transaction models. For
example, you can sign individual items or multiple items of an XML document. The
document you sign can be local or even a remote object, as long as those objects can
be referenced through a URI (Uniform Resource Identier). Consider for instance,
an order entry system where a user orders an item for the companys website. Behind
the scene, the order ows through the inventory system, a credit card company,
provisioning and billing systems, followed by shipping. XML Signature allows each
system to sign and verify the sections related to them. Billing system for instance does
not care for the shipping address or the item ordered and only wishes to verify the
integrity of billing data.
XML digital signature also allows multiple signing levels for the same content, thus
allowing exible signing semantics. For example, the same content can be semantically
signed, cosigned, witnessed, and notarized by different people.
You can associate a digital signature of an XML Document in different ways:
202
Enveloped The signature is a child of the data being signed.

Enveloping The signature encloses the data being signed.
Detached The signature is a sibling of the element being signed and is
referenced by a local link, or it can be located elsewhere on the network.
Let us examine the structure of XML Signature. The main thing dened by the XML
Signature specication is a <Signature> element whose contents include both the
digital signature itself and the information about how this signature was produced.
Later in the section, we will illustrate the process of signing a le (order.xml) to
create a signed document (signature.xml) and to how to validate its integrity.
Figure 7 is the schema generated from the signed document. Before we dive into the
sample, let us examine the various elements of <Signature>.
Figure 7. Schema for Signature Element

A typical instance of <Signature> as used with SOAP contains the <SignedInfo>
and <KeyInfo> elements.
<SignedInfo> element describes the information that is being signed. This
element itself contains several subelements, each of which species something
about the signed information. The most important of these include the following
elements.
<CanonicalizationMethod> identies the algorithm that was used to convert
this XML document into a standard form before the signature was generated.
203
This process is necessary because for digital signatures to work, each partys
view of the signed document must be identical bit-for-bit. Canonicalization
transforms an XML document into a standard form, ensuring that both parties
view a document in the exact same way before signing it. We will look more
into Canonicalization in a little bit.
<SignatureMethod> identies the algorithm used to create the digital signature. Both the XML Signature standard and WS-Security require support for
the Digital Signature Standard, which uses Secure Hash Algorithm (SHA)-1 to
create a message digest together with the DSA algorithm, and also recommend
support for using SHA-1 together with the RSA algorithm.
<Reference> identies the information being signed, along with any transformations that were applied to it. When used with SOAP, this part of the
signature typically references the signed parts of this SOAP message, such
as the body and parts of the header. The Reference element also contains
the two subelements: <DigestMethod> and <DigestValue>. The rst of
these identies the algorithm used to create the message digest for this digital
signature, while the second contains the message digest value itself.
<SignatureValue>: This element contains the actual signature. As signatures
are always binary data, XML DSIG species that the signature value is always a
simple element with Base64-encoded content
<KeyInfo> indicates what key should be used to validate this signature. The
key can be identied in a variety of ways, such as referencing a certicate that
contains the appropriate public key. This element can even be omitted entirely,
since the recipient may know what key to use by some other means.
<dsig:Object> element contains the actual data that is being sent and used in
creating the digest.
The following example illustrates how XML Signatures work. We are using IBMs
XML Security Suite to generate and verify the signature. The following listing displays
the document order.xml that we will using in the next few examples:
<OrderData id="123">
<CustomerData>
<FirstName>John</FirstName>
<LastName>Doe</LastName>
<BillingData>
<Street>123 MainStreet</Street>
<City>Denver</City>
<State>Colorado</State>
204
<Country>US</Country>
<ZipCode>80021</ZipCode>
</BillingData>
<ShippingData>
<Street>123 MainStreet</Street>
<City>Denver</City>
<State>Colorado</State>
<Country>US</Country>
<ZipCode>80021</ZipCode>
</ShippingData>
<Payment Data>
<CreditCardData>
<Name>John Doe</Name>
<Number>122324211234343</Number>
<Issuer>AmericanExpress</Issuer>
<Expiry>20040405</Expiry>
</CreditCardData>
</PaymentData>
</CustomerData>
<ItemData>
<Name>Enterprise SOA</Name>
<Type>Book</Type>
<Reference>83761761</Reference>
<Quantity>1</Quantity>
<Price>44.95</Price>
</ItemData>
</Order>
3.3.1. Generate Certicate

In real world, to be able to create a digital signature you need a certicate, which is
issued by certicate authorities such as Verisign. However, for our example here we
will use the Java keytool command.
keytool -genkey -dname "CN=John Doe, OU=Enterprise SOA Book,
O=Manning, L=Denver, S=Colorado, C=US" -alias john
-storepass password -keypass security
// dname = distinguished name
// CN = Common Name
// OU = Organizational Unit
// O = Organization

//
//
//
//
//
//
205
L = Location
S = State
C = Country
alias = Alias for this certificate
storepass = Password for the key store
keypass = Pasword for the private key
3.3.2. Signing
To create digital signatures, we use the SampleSign2 application that is shipped with
the IBMs XML Security Suite. The sample application can be used to create both
detached and enveloping signatures. The application can be used as
java dsig.SampleSign2 <your-alias> <your-storepassword>
<your-keypassword>
<resource><resource> ... > signature.xml
where resource is
-ext URL for an external resource;
-embxml URL for an embedded XML resource (content of specied URL is
embedded in the signature).
Here is how order.xml is signed to generate signature.xml:
java dsig.SampleSign2 john password security -embxml
file:///d:/tools/xss4j/samples/order.xml > signature.xml
The alias, store password and key password are the ones that were generated in the
previous step. The output is directed to a signature.xml, which will have the
signature and the embedded data.
The following listing shows the content of the le signature.xml.
<Signature xmlns="http://www.w3.org/2000/09/xmldsig#">
<SignedInfo>
<CanonicalizationMethod
Algorithm="http://www.w3.org/TR/2001/
REC-xml-c14n-20010315">
</CanonicalizationMethod>
<SignatureMethod
Algorithm="http://www.w3.org/2000/09/
xmldsig#dsa-sha1">
206
</SignatureMethod>
<Reference URI="#Res0">
<Transforms>
<Transform
Algorithm="http://www.w3.org/TR/2001/
REC-xml-c14n-20010315">
</Transform>
</Transforms>
<DigestMethod
Algorithm="http://www.w3.org/2000/09/
xmldsig#sha1">
</DigestMethod>
<DigestValue>aJH0LRYgJaAtuqC/PJ/wfFiEI2A=
</DigestValue>
</Reference>
</SignedInfo>
<SignatureValue>
XJjZ7OwwGn9Zhyga/t4ipuFKOnhREqX9UIDUaA2sPJVmhw+y3BksMA==
</SignatureValue>
<KeyInfo>
<KeyValue>
<DSAKeyValue>
<P>/X9TgR11EilS30qcLuzk5?</P>
<Q>l2BQjxUjC8yykrmCouuEC/BYHPU=</Q>
<G>9+GghdabPd7LvKtcNrhXuXmUr7?</G>
<Y>NTijOltictv5/SCbDZhUwKNlGlIOH?</Y>
</DSAKeyValue>
</KeyValue>
<X509Data>
<X509IssuerSerial>
<X509IssuerName>
CN=John Doe,OU=Enterprise SOA Book,
O=Manning,L=Denver,ST=Colorado,C=US
</X509IssuerName>
<X509SerialNumber>1079744797</X509SerialNumber>
</X509IssuerSerial>
<X509SubjectName>
CN=John Doe,OU=Enterprise SOA Book,
O=Manning,L=Denver,ST=Colorado,C=US
</X509SubjectName>
<X509Certificate>
MIIDHDCCAtoCBEBbmR0wCwYHKoZIzjg?
207
</X509Certificate>
</X509Data>
</KeyInfo>
<dsig:Object xmlns=""
xmlns:dsig="http://www.w3.org/2000/09/xmldsig#" Id="Res0">
<OrderData id="123">
<CustomerData>
...
</CustomerData>
</OrderData>
</dsig:Object>
</Signature>
The actual signed data is underneath the dsig:Object. As we will see in the next
section, any alteration with the data will result in a verication failure.
3.3.3. Verication
The XML Security Suite provides a utility, VerifyGUI that reports validity of each
resource and validity of the signature. You can check a given signature to ensure that
the signed resource has not changed, and you can check that the signature matches
the information in the senders certicate.
java dsig.VerifyGUI < signature.xml
If the signature and all of signed resources were not modied, VerifyGUI reports the
result of verication as Core Validity: OK.
Figure 8. VerifyGUI Success
208
If the signed le is changed, the signature will no longer be valid. To illustrate this,
change the order id in the signature.xml to 234 and again run the utility. This
time, VerifyGUI reports as Core Validity: NG Digest Value mismatch.
Figure 9. VerifyGUI Failure
3.4. Canonicalization
XML documents can have the same logical meaning but different physical implementations based on several characteristics such as character encoding, attribute ordering
or white spaces. Digital signature checks whether a message has changed along the
way, but an XML documents physical representation can change during processing,
even though its information content remains the same.
Let us examine the snippets below:
<OrderData id="123" type="web">
<BillingData/>
</OrderData>
and
<order type="web" id="123">
<BillingData>
</BillingData>
</order>
Though both snippets have identical information content, because of their permissible
differences, their digests will denitely differ. XML Canonicalization solves this
209
problem. The Canonical XML specication establishes the concept of equivalence

between XML documents and provides the ability to test at the syntactic level.
It enables the generation of the identical message digest and thus identical digital
signatures for XML documents that are syntactically equivalent but different in
appearance. By calculating the digest over a documents canonical form, we can
ensure that, as long as the documents information content stays the same, its
signature will still verify, even if the physical representation has somehow changed
during processing.
XML Signature and message signing proves the senders identity and ensures that the
message has not been tempered. However, message signing does not prevent external
parties to view the contents of the message. There are cases when data integrity is
not enough and the receiver expects the data condentiality, i.e. that the data has
not been read in transit. For instance, if you are sending a credit card number as
part of the message, validating the data is not sufcient enough. You would like the
data encrypted in such a way that only the intended message recipient can read the
message. Anyone watching the wire exchange should remain oblivious to the contents
of the message. This is where encryption comes in.
3.5. Condentiality and Encryption

Encryption involves the securing of messages exchanged between a sender and receiver.
Figure 10 depicts the process involving encryption and decryption of messages.
Figure 10. Encryption

In Figure 10, a sender wants to send a message over the internet to a receiver. The
message is usually referred to as cleartext or plaintext. To preserve condentiality, the
message rst undergoes a process called encryption. The encrypted message (also called
as ciphertext) renders the actual message text invisible. When the message reaches the
receiver, a process called decryption takes place, in which the encrypted message is
converted to plaintext. Encryptions are done using complex mathematical techniques
called as Cryptographic algorithms or ciphers. Most cryptographic algorithms can be
divided into two main categories: one-way encryption and two-way encryption. In
a one-way encryption scenario (e.g., RSA, MD4, MD5 and SHA), encryption and
210
decryption involve a different key where in a two-way encryption, the same key is
used for both operations.
Figure 11. One Way/Two Way Encryption

When you encrypt data, you can choose either symmetric or asymmetric encryption.
3.5.1. Symmetric Encryption

Symmetric Encryption also known as secret-key cryptography requires a shared secret.
This is a two-way encryption model because the same key is used for both encryption
and decryption. Symmetric encryption is good if you control both endpoints and
can trust the people and applications on both ends. The inherent problem with this
encryption scheme is the distribution of the key. Prior to the invocation of the service,
the key needs to be sent to the receiver using some other trusted mechanism. If a third
party knows the key in can hack into message exchanges between the two parties.
Figure 12. Symmetric Encyrption

1. A secret key is shared between sender and receiver.
2. Sender encrypts the plaintext using the secret key.
3. The ciphertext is sent to the receiver.
211
4. Receiver decrypts the ciphertext using the secret key.

5. Receiver retrieves the plaintext message.
3.5.2. Asymmetric Encryption

Asymmetric encryption, also known as Public-key cryptography is a one-way encryption model where two keys (public key and private key) are used for encrypting and
decrypting data, respectively. In the asymmetric encryption, the endpoint receiving
the data can publicly post its certicate and allow anyone and everyone to encrypt
information using the public key. Only the receiver knows the private key and thus
only the receiver can decrypt the data into something meaningful. This solves the
key dispersal problem of the symmetric encryption model. Figure 13 explains the
procedure.
Figure 13. Asymmetric Encryption

1. Receiver holds the private key.
2. Sender encrypts the plaintext using the public key.
3. The message is sent to receiver in encrypted form.
4. Receiver decrypts the message using the private key.
5. Receiver retrieves the plaintext message.
Let us again look at the example that we mentioned in the earlier section where a
user orders an item from a company website. The message (order.xml) travels back
and forth between several parties. The message has information about the customer
(e.g. credit card data) as well as the item that he ordered. The credit card company
does not need to know about the item ordered and the shipping vendor does not
need to know about the credit information. We can resolve this issue by selectively
212
encrypting the data contents of the message, and the various parties will only be able
to read the data that is meant for them.
Like everything else in the ESOA space, there is an XML solution to address the issue
of encryption. The W3C and IETF have published a specication for encryption:
XML Encryption.
3.6. XML Encryption

XML encryption extends the power of the XML digital signature system by enabling
the encryption of the message that has been digitally signed. The specication outlines
a standard way to encrypt any form of digital content and permits encryption of
an entire XML message, a partial XML message, or an XML message that contains
sections that were previously encrypted.
The following is the section from order.xml that pertains to credit card information.
<PaymentData>
<Name>John Smith</Name>
<CreditCardData>
<CreditCard Limit=5,000 Currency=USD>
<Number>4019 2445 0277 5567</Number>
<Issuer>American Express</Issuer>
<Expiration>04/02</Expiration>
</CreditCardData>
</PaymentData>
The application can encrypt just the credit card element.
<PaymentData>
<Name>John Smith</Name>
<EncryptedData
Type=http://www.w3.org/2001/04/xmlenc#Element
xmlns=http://www.w3.org/2001/04/xmlenc#>
<CipherData>
<CipherValue>A23B45C56</CipherValue>
</CipherData>
</EncryptedData>
</PaymentData>
By encrypting the entire <CreditCardData> element from its start to end tags, the
identity of the element itself is hidden. An eavesdropper does not know whether the
transaction used a credit card, money transfer or a check. The CipherData element
213
contains the encrypted serialization of the CreditCardData element. Figure 14

explains the structure of the encrypted data.
Figure 14. XML Encryption
<EncryptedData>: This is the core element of XML encryption and replaces

the data being encrypted. It contains (via one of its children content) or identies
(via a URI reference) the cipher data.
<CipherData> element is created to contain the encrypted serialization of the
data. It either envelopes or references the raw encrypted data. If enveloping, the
raw encryptied data is the <CipherValue> elements content; if referencing,
the <CipherReference> elements URI attribute points to the location of the
raw encrypted data.
<CipherValue> element holds the encrypted result of the data.
<EncryptionMethod> describes the method that used used for encryption
such as RSA.
<KeyInfo> borrowed from XML Signature. Here, it is assumed that both sides
already know all of the keys available for use in encryption and decryption, so
the only thing required is to inform the receiver of which one was chosen.
It is also possible to convey an encrypted key in the same message that carries
data that was encrypted using that key. For instance when the data is encrypted
using a symmetric key, then that symmetric key is encrypted using the public
key of the messages recipient and sent along with the data. When the message
arrives, the recipient can use its private key to decrypt the embedded symmetric
key, then use this symmetric key to decrypt the actual data. The encrypted key can
be sent using the element <EncryptedKey>. The format of <EncryptedKey> is
much like <EncryptedData> has three main subelements: <EncryptionMethod>,
<KeyInfo>, and <CipherData>. Rather than having the message data, these
214
elements contain information about the key, how it has been encrypted and the
encrypted value.
The following example demonstrates the use of XML Encryption. As in the previous
section, we use IBMs XML Security Suite for the task. We will also reuse the
certicate keys generated in the earlier section on XML Signature.
3.6.1. Encryption
We use the DomCipher application that is shipped with the IBMs XML Security
Suite. The sample application can be used to create both detached and enveloping
signatures. The application can be used as
java enc.DomCipher -e <provider> <keyinfo> <target>
[xpath template...]
//Encryption
and
java enc.DomCipher -d <provider> <keyinfo> <target>
[xpath template...]
//Decryption
Here is how to encrypt order.xml
java enc.DOMCipher -e "*" john security password
file:///d:/tools/xss4j/samples/order.xml > encrypted.xml
The alias, store password and key password are the ones that were generated in the
previous step. The output is directed to an encrypted.xml.
The application also allows us to encrypt parts of data. For instance, the following
command will only encrypts the credit info in the order.xml
java enc.DOMCipher -e "*" john security password
file:///d:/tools/xss4j/samples/order.xml
"//*[name()=CreditCardData]" >
encrypted.xml
3.6.2. Decryption
The XML Security Suite provides a utility, VerifyGUI that reports validity of each
resource and validity of the signature. You can check a given signature to ensure that
215
the signed resource has not changed, and you can check that the signature matches
the information in the senders certicate.
Here is how to decrypt encrypted.xml:
java enc.DOMCipher -d "*" john security password
file:///d:/tools/xss4j/samples/encrypted.xml > order.xml
3.7. Authorization
Security tokens provide a way of authentication and establish your identity. However,
authentication does not allow you to perform a specic action or to access a resource.
For instance, you can login into a database system using your login id and password.
It will let you navigate around the tables and read the data but you will not be
allowed to delete a table unless you are a DBA. Similarly, your credit card will prove
your identity and who you are but you may not be allowed to make a million dollar
transaction through it due to the policy of your credit card company. In a software
language, a policy refers to the set of conditions and constraints that must be met in
order for an entity to perform some type of operation. The requester can be a person
or another application but it has to abide by the policies of the provider to be able to
access its services. Extensible Access Control Markup Language (XACML) provides
XML documents with support for dening access control and policies.
3.8. Extensible Access Control Markup

Language (XACML)
XACML is an OASIS standard that describes both a policy language and an access
control decision request/response language (both written in XML). The policy
language is used to describe general access control requirements, and has standard
extension points for dening new functions, data types, combining logic, etc. The
request/response language lets you form a query to ask whether or not a given action
should be allowed, and interpret the result.
3.8.1. Key Concepts

XACML includes the concepts of Policy Enforcement Point (PEP) and Policy
Decision Point (PDP). PEP is the entity that is charged for access control. When a
client makes a resource request, the PEP will form a request based on the requesters
216
attributes, the resource in question, the action, and other information pertaining to
the request. The PEP will then delegate the authorization decision to the PDP. Policy
Decision Point will look at the request and the policy that applies to the request, and
come up with an answer about whether access should be granted.
Figure 15. XACML
3.9. Top-Level Constructs: Policy and

PolicySet
At the root of all XACML policies is a Policy or a PolicySet. A PolicySet is a
container that can hold other Policies or PolicySets. The Policy element is
composed principally of Target, Rule and Obligation elements and is evaluated
at the Policy Decision Point to grant or deny access to the resource.
Since multiple policies may be found applicable to an access decision (and since a
single policy can contain multiple Rules), XACML provides support for Combining
Algorithms that are used to reconcile multiple outcomes into a single decision.
3.10. Key Management

Most of the concepts like authentication, digital signature, encryption discussed
earlier, rely on public and private keys, digital certicates among others. There is a
need to simplify the management of these security components and to keep them
organized and secure. There are many PKI solutions available like X.509, Pretty Good
Privacy (PGP) and Simple Public Key Infrastructure (SPKI). In an ESOA, to be able
to talk to the other systems, each application needs to be aware of the PKI solutions
used by other systems. For instance, if service A uses an X.509 PKI solution and
217
sends encrypted documents to service B, which uses an SPKI solution, then service
B will not be able to decrypt and use the document sent by A. For A and B to work
together, one of them has to understand the others PKI solution. If you extrapolate
this scenario to a situation where multiple partners are involved, it becomes clear that
all of the partners will have to be aware of each others PKI solution, thus increasing
each applications complexity many times.
The XML Key Management Specication (XKMS), another W3C standard aims
to provide a standard XML-based solution for the management of keys for
authentication, encryption and digital signature services.
3.11. XML Key Management Specication

(XKMS)
XKMS allows application developers to outsource the processing of key management
(registration, verication etc.) to trust services accessed over the internet. This trusted
third paty will act as an intermediary that frees the SOA programmer from having
to track the availability of keys or certicates and ensures their validity. XKMS
specications are made up of two other specications.
3.11.1. XML Key Information Service Specication

(XKISS)
XKISS allows the client applications to authenticate encrypted/signed data. The
XKISS service specication denes the following two operations:
1. Locate: Locate resolves a <ds:KeyInfo> element that may be associated with
XML encryption or XML signature, but it does not prove the validity of data
binding in the <ds:KeyInfo> element.
2. Validate: This operation not only searches the public key corresponding to the
<ds:KeyInfo> element, but also assures that the key binding information that
it returns is trustworthy.
3.11.2. XML Key Registration Service Specication

(XKRSS)
XRSS allows for registration and subsequent management of public key information.
A client of a conforming service may request that the registration service bind
218
information to a public key. The information bound may include a name, an

identier or extended attributes dened by the implementation.
An XKRSS service specication denes four operations:
1. Register: Information is bound to a key pair through key binding. During
registration, either the client provides the public key, along with some proof of
possession of the corresponding private key, or the service provider generates the
key pair for the client. The service provider may request more information from
the client before it registers the public key (and optionally the private key as
well).
2. Re-issue: A previously registered key binding is re-issued. New credentials in the
underlying PKI are generated using this operation. While there is no lifespan
for the key binding information used by XKMS, the credentials issued by
the underlying PKI occasionally do have a time span that must be renewed
periodically.
3. Revoke: This operation allows clients to destroy the data objects to which a key
is bound. For example, an X.509 certicate that is bound to an XKMS key is
destroyed when this operation is called.
4. Recover: This operation allows clients to recover the private key. For this operation
to be meaningful, the private key must have been registered with the service
provider. One of the ways in which the service provider may have the private key
is when the key pair is generated at the server rather than the client.
3.12. Single Sign-On

Single sign-on authentication is the ability for an end user or application to access
other applications within a secure environment without needing to be validated
by each application. For instance, you should be able to log into your corporate
application infrastructure with a single username and password. Even if there are
different subsystems underneath for providing payroll information, vacation details
or 401K data, it should be oblivious to the user and he should be able to navigate
across the subsystems seamlessly without entering the username/password for each
sub-system.
The best approach to allow access to all applications without additional intervention
after the initial sign-on is by using a prole that denes what the user is allowed
to do. Many companies such as Ping Identity, RSA, Netegrity, support products
for web-based, single sign-on authentication and authorization. In general, these
219
products use an intermediary process that controls and manages the passing of user
credentials from one application to another. The responsibility of authentication
and authorization is shifted to a third party, leaving the application free to focus on
implementation of business logic.
The implementation of SSO varies a bit according to the business scenario. In an
ETrade like banking environment, the user can log into one module (say mortgage),
and the system uses SSO to enable the user in other modules as banking, securities
and options. If the user signs off from any of the module, the system signs off the
user from all the other modules. The situation is a bit different in a scenario where
there are more disparate applications participating in the SSO solution. Corporate
intranet, that provides links to other internal sites such as HR and Payroll is a good
example. When the user log in the intranet, he might not realize that he is also login
into other secure sites like Payroll that are SSO enabled. This can pose a security risk
if the user leaves his machine unlocked for an extended period of time. One solution
is to implement a short SSO timeout that mitigates the risk of unauthorized access
to sensitive applications.. The decision to enable an application for SSO lies with the
business. But there are considerations like mentioned above that should be taken into
account before enabling an application. One approach to handle the above issue is by
implementing security levels. The enterprise security group can divide applications
into security levels based on the applications sensitivity. The SSO is enabled within
the security level. Therefore, if the user login into the intranet (with Security level
0), he has to login again to access the Payroll information which is set at a higher
security level.
In ESOA, single sign-on is a subset of a bigger issue referred as identity management.
Single sign-on only deals with username/password. However, each user has an identity
and username/password can be considered as two of the many attributes that are
associated with the identity. One of the approaches seen in SSO implementation is
the use of attributes to dynamically determine the users access or authorization. Thus
when the user logs in, the SSO application (e.g. Netegrity) authenticates the user
against the company directory server and populates the users principal with selected
attributes. The various business applications can then implement a message handler
that can intercept the request and assign dynamic roles based on the attributes in the
request. For instance, if the manager attribute is set to true, the user is allowed
to view the payroll information of other employees. With this implementation, the
SSO application is only used for authentication, and the authorization decisions are
made within the applications domain.
220
3.13. Identity Management

Identity represents the core of any business relationship. It encompasses attributes
and characteristics critical to developing and deploying valuable services. Identity
Management is a set of processes for the creation, maintenance, and use of identities
and their attributes, as well as credentials and entitlements, plus a supporting
infrastructure.
Users today have multiple login ids and passwords for several applications. One
user-id and password for Amazon; another one for the mortgage lender; and one
each for every bank account and credit card company. This leads to what is known as
the sticky note problem. Since each individual has many usernames and password,
they end up writing them on sticky notes all over their desk posing a major
security risk. Similarly in business-to-business interactions, each service requires
has its own way of authenticating the request. The problem becomes even bigger
in enterprise SOA where the service spawns the boundaries of many applications
and operating systems, endpoints and intermediaries. Each component has its own
mechanism for managing identity, credentials and policies, leading to a fragmented
identity infrastructure. Such infrastructure contains overlapping identity repositories,
inconsistent policy frameworks, and process discontinuities. Needless to say that the
result is cumbersome and such infrastructures are expensive to manage, error prone,
and can have weaknesses that can lead to security breaches. One of the challenges for
a successful adoption of Enterprise SOA is to provide a simple and secure identity
management solution.
There are several industry initiatives to address this issue. One of them is Liberty
Alliance project, which we will explore in the next section. Single sign-on and
federated network identity (a system for binding multiple accounts for a given user)
are key to solving this. The basic philosophy is to have a separate identity management
system that contains a centralized repository of identity and policy information that
every component can access. When the user logs in, the system authenticates the user
via the identity management system, and creates a token that identies that user and
the privileges that the user has. That token then remains with the user request as it
traverses the service composition layer, the atomic Web Services, and nally to the
back-end systems.
3.14. Liberty Alliance Project

Liberty Alliance project is a business alliance that was formed in 2001, with the
goal of establishing an open standard for federated identity management. The goal
221
is to make it easier for businesses and consumers to conduct commerce while

providing protection mechanisms for privacy and identity information. The project
has established a specication for an open standard of federated network identity that
integrates with a variety of products and services.
The goal of the project is to enable the concept of network identity. Network identity
is the fusion of network security and authentication, user provisioning and customer
management, single sign-on technologies, and web services delivery. Internet users
today have multiple login Ids, passwords and other aspects that make up an identity.
The information is spread throughout the internet, buried in multiple sites. This is
where Identity Providers can t it. An identity provider can create, maintain and
manage the identity information for users (or subject in security language). The
users here represent an individual person, a group of individuals, a corporation or
any entity that is requesting the service. The identity provider develops a relationship
with the service providers(s) and provides the authentication services. With federation,
the Identity Provider and Service Provider together establish an opaque identier(s)
to be used to refer to a particular user. (Subsequent communications use this agreedupon pseudonym for the user). The Service Provider bases its trust in the Identity
Providers assertion through signatures, certicate chains, validity intervals and any
other technical mechanism. When the users and service providers rely on identity
providers as trusted sources for authenticated user information, it leads to a circle of
trust.
A circle of trust is a federation of service providers and identity providers that
have business relationships based on Liberty architecture and operational agreements
and with whom users can transact business in a secure and apparently seamless
environment. For instance, as illustrated in the following diagram, the users local
bank can act as an identity provider and collaborate with a bill payment company to
provide its services. The bill payment company in turn can then interact with external
federated partners as Utilities Company, Phone Company, Credit card company and
so on.
A federated identity architecture delivers the benet of simplied sign-on to users
by granting rapid access to resources to which they have permission, but it does not
require the users personal information to be stored centrally. Microsofts Passport
and WS-Federation are other technologies attempting to solve the similar problem.
WS-Federation specications are still in early stages but it will be interesting to
follow their progress and how it converges with the Liberty Alliance Project. Since
both are based on open standards and have industry support behind them but have
overlapping goals.
222
Figure 16. Circle of Trust
Multiple systems (e.g. Credit Card company, Phone Company), coordinating user
authentication decisions presents another challenge. The information about the
authenticated user needs to be exchanged between trusting service providers. First,
a user or a subject is likely to maintain different local identities with different
service providers. For instance John Doe may be referred as jdoe with the phone
company and johnd with the cable company. Similarly each system has its own way of
expressing security assertions. The successful adoption of Single sign-on and identity
management solutions required a common way to express assertion so that they can
be shared between trusting parties. That need led to the development of an XML
language to express security assertions, the Security Assertions Markup Language, or
SAML. SAML itself is a merger of two previously competing XML-based languages
aimed to specify security-related information, S2ML and AuthML. Currently, SAML
is careering through the OASIS (Organization for the Advancement of Structured
Information) open standards consortium and is poised to become the dominant
security-specic XML language. The next section explains the language in more
detail.
223
3.15. Security Assertion Markup Language

(SAML)
Most security solutions that exist today are based on the assumption that the
consumers and providers are located on the same physical (e.g. local LAN) and/or
logical (e.g. VPN) network and therefore are under the realm of a single trust domain.
A trust domain can be considered as a logical construct with a single set of access
control policies. It is easy to establish and manage security within a single domain.
For instance, obtaining access to an application within the single domain requires
creation of a new user account in the domain and granting the appropriate access.
However, the collaboration of services in the ESOA world requires a standard way
to communicate security attributes across the domains. If an individual identity has
been established and veried in one trust domain, he should be allowed to assert his
identity in another trust domain. Each domain could still control the local security
policies to evaluate whether to grant access to its services. When the service request
spans more than one domain, the security attributes and assertions must travel with
the message. This form of portable security could be achieved using SAML.
The Security Assertion Markup Language (SAML) standard denes a framework for
exchanging security information between online business partners.
SAML architecture is based on two key concepts: Asserting Party and Relying
Party. Asserting parties that are also known as SAML authorities asserts information
about a subject. For instance, the asserting party asserts that the user has been
authenticated and has given associated attributes. The relying party relies on the
information supplied by the asserting party. It still uses its local access policies to
decide whether the subject may access the resources. The information passed around
between asserting parties (SAML authorities) and relying parties is mostly in the form
of XML and the format of these XML messages and assertions is dened in a pair
of SAML XML schemas. SAML addresses the issue of SSO that allows the users to
gain access to resources in multiple domains without having to re-authenticate after
initially logging in to the rst domain. To achieve SSO, the domains need to form a
trust relationship before they can share an understanding of the users identity that
allows the necessary access.
The following listing illustrates how SAML can be used for authentication assertion.
<saml:Request MajorVersion="1" MinorVersion="0"
RequestID="123">
<saml:AuthenticationQuery>
224
<saml:Subject>
<saml:NameIdentifier
SecurityDomain="Enterprise SOA"
Name="JohnDoe"/>
</saml:Subject>
</saml:AuthenticatationQuery>
</saml:Request>
In the above listing, the requesting party creates a assertion query for the subject
JohnDoe in the Enterprise SOA Domain. The next listing shows the SAML
response.
<saml:Response MajorVersion="1" MinorVersion="0"
InResponseTo="123" StatusCode="Success">
<saml:Assertion MajorVersion="1" MinorVersion="0"
AssertionID="123"
Issuer="Enterprise SOA"
IssueInstant="2004-03-22T05:45:00Z">
<saml:Conditions
<NotBefore="2004-03-22T05:45:00Z"
<NotAfter="2004-03-22T09:45:00Z">
</saml:Conditions>
<saml:AuthenticationStatement
AuthenticationMethod="Password"
AuthenticationInstant="2004-03-22T05:45:00Z">
<saml:Subject>
<saml:NameIdentifier
SecurityDomain="Enterprise SOA"
Name="JohnDoe"/>
<saml:Subject>
</saml:AuthenticationStatement>
</saml:Assertion>
</saml:Response>
<Conditions> element species the conditions that must be considered when
evaluating the assertion. For instance, in the above listing assertion is only valid

for a specied time period. The statement types can be AuthenticationStatment,
AuthorizationDecisionStatement or AttributeStatement. In the above
listing, it is AuthenticationStatement and it states that the subject JohnDoe was
authenticated at a certain time and it was authenticated using password authentication.
Many of the known industry players in the world of SOA Security (IBM, Microsoft,
BEA, RSA) collectively proposed a number of new web services specications related
225
to security. We have reasons to believe that the intention is not just to cure the
insomniacs. The goal for these specications is to make it easier to apply business
polices and to implement security for a wider range of applications. None of these
specications are an attempt to invent new security solutions but rather use the
existing ones to work along with web services. In the next section, we will look at the
some of most commons one.
4. Web Services Security (WSS)

Web Service Security specication was specically created for using the security
technologies that we discussed earlier in the chapter in the context of a SOAP
message. Web Services Security (WSS) describes enhancements to SOAP messaging
in order to provide quality of protection through message integrity, and single message
authentication. The specications provide three main mechanisms: ability to send
security tokens as part of a message, message integrity, and message condentiality.
These mechanisms can be used to accommodate a wide variety of security models
and encryption technologies.
WS-Security addresses security by leveraging existing standards and specications.
This avoids the necessity to dene a complete security solution within WS-Security.
The industry has solved many of these problems. Kerberos and X.509 address
authentication. XML Encryption and XML Signature describe ways of encrypting
and signing the contents of XML messages. XML Canonicalization describes ways
of making the XML ready to be signed and encrypted. What WS-Security adds
to existing specications is a framework to embed these mechanisms into a SOAP
message. This is done in a transport-neutral fashion. WS-Security denes a SOAP
header element to carry security-related data.
The WS-Security specication denes a new SOAP header:
<xs:element name="Security">
<xs:complexType>
<xs:sequence>
<xs:any processContents="lax"
minOccurs="0" maxOccurs="unbounded">
</xs:any>
</xs:sequence>
<xs:anyAttribute processContents="lax"/>
</xs:complexType>
</xs:element>
226
The security header element allows any XML element or attribute to live within it.
This allows the header to adapt to whatever security mechanisms your application
needs.
WS-Security needs this type of structure because of what the header must do. It must
be able to carry multiple security tokens to identify the callers rights and identity. If
the message is signed, the header must contain information about how it was signed
and where the information regarding the key is stored. The key may be in the message
or stored elsewhere and merely referenced. Finally, information about encryption
must also be able to be carried in this header. In an ESOA application, the same
message ows through a number of intermediaries and endpoints. For each of these
components, the SOAP message can have multiple security headers identied by a
unique actor. The actor attribute in any SOAP header is meant to say this header
is meant for any endpoint acting in the capacity indicated by the actor URI. This
means that the intermediary may act in varying capacities and may consume zero,
one or more headers. Let us look at some examples of how WS-Security can be used
to exchange security information.
4.1. Security Tokens

The security tokens can be passed either as a username/password or X.509 Certicates
or Kerberos. To pass user credentials as username/password manner, WS-Security has
dened the UsernameToken element. When passing a UsernameToken in a SOAP
message, the XML may come across as following:
<wsse:UsernameToken>
<wsse:Username>scott</wsse:Username>
<wsse:Password Type="wsse:PasswordText">tiger</wsse:Password>
</wsse:UsernameToken>
The above example shows the password being sent as plain text. To be a little more
secure, the password can be sent in as a digest hash.
<wsse:Username>scott</wsse:Username>
<wsse:Password Type="wsse:PasswordDigest">
X456DXTQALOY!ADT&ADYT
</wsse:Password>
Credential using .509 Certicate or Kerberos can be passed using the
BinaryTokenElement. The following listing shows an example:
227
<wsse:BinarySecurityToken
ValueType="wsse:X509v3"
EncodingType="wsse:Base64Binary"
Id="SecurityToken-f32456fasdfft62......"
</wsse:BinarySecurityToken >
The valueType may be any of the following values, dened by the ValueTypeEnum
in the WS-Security schema document:
wsse:X509v3: An X.509, version 3 certicate.
wsse:Kerberossv5TGT: A ticket granting ticket as dened by Section 5.3.1 of
the Kerberos specication.
wsse:kerberossv5ST: A service ticket as dened by Section 5.3.1 of the Kerberos
specication.
The EncodingType attribute indicates the encoding method and can be set to either
wsse:Base64Binary or wsse:HexBinary.
4.2. Signature
Message integrity is provided by leveraging XML Signature in conjunction with
security tokens (which may contain or imply key data) to ensure that messages
are transmitted without modications. The integrity mechanisms are designed to
support multiple signatures, potentially by multiple actors, and to be extensible to
support additional signature formats. The following listing displays a sample signed
SOAP message.
<?xml version="1.0" encoding="utf-8"?>
<s:Envelope
xmlns:s="http://schemas.xmlsoap.org/soap/envelope/"
xmlns:wsse="http://schemas.xmlsoap.org/ws/2002/12/secext"
xmlns:wsu="http://schemas.xmlsoap.org/ws/2002/07/utility">
<s:Header>
<wsse:Security>
<ds:Signature xmlns:ds="http://www.w3.org/2000/09/xmldsig#">
<ds:SignedInfo>
<ds:CanonicalizationMethod
Algorithm="http://www.w3.org/2001/10/xml-exc-c14N"/>
<ds:SignatureMethod
Algorithm="http://www.w3.org/2000/09/xmldsig#rsa-sha1"/>
<ds:Reference URI="#MessageBody">
228
<ds:DigestMethod
Algorithm="http://www.w3.org/2000/09/xmldsig#sha1"/>
<ds:DigestValue>
aObd89l4kjfdsi...
</ds:DigestValue>
</ds:Reference>
</ds:SignedInfo>
<ds:SignatureValue>
jlkfds90dfl...
</ds:SignatureValue>
<ds:KeyInfo>
<wsse:SecurityTokenReference>
<wsse:Reference URI="#X509Cert"/>
</wsse:SecurityTokenReference>
</ds:KeyInfo>
</ds:Signature>
</wsse:Security>
</s:Header>
<s:Body wsu:Id="MessageBody">
...
</s:Body>
</s:Envelope>
4.3. Encryption
In the same way they addressed the question of integrity, the authors of WS-Security
chose to build on existing standards (XML Encryption) for condentiality rather
than create something new. The following listing shows an example of encrypted
message:
<s:Envelope
xmlns:s="http://schemas.xmlsoap.org/soap/envelope/"
xmlns:ds="http://www.w3.org/2000/09/xmldsig#"
xmlns:xenc="http://www.w3.org/2001/04/xmlenc#">
<s:Body>
<xenc:EncryptedData>
<EncryptionMethod
Algorithm=http://www.w3.org/2001/04/xmlenc#tripledes-cbc/>
<ds:KeyInfo>
<ds:KeyName>
CN=ESOA, C=US
229
</ds:KeyName>
</ds:KeyInfo>
<xenc:CipherData>
<xenc:CipherValue>
r5Kipsslkr490dDV ...
</xenc:CipherValue>
</xenc:CipherData>
</xenc:EncryptedData>
</s:Body>
</s:Envelope>
As we mentioned earlier in this section, the purpose of WS-Security standard is not
to invent new standards for security but to use the existing ones. It achieves that be
dening a SOAP security header and thus providing a standard place to list security
artifacts. The following listing represents a consolidated basic WS-Security SOAP
Header:
<s:Envelope>
<s:Header>
</wsse:Security>

...

<ds:Signature>
...
</ds:Signature>

<xenc:ReferenceList>
<xenc:DataReference URI="#body"/>
<xenc:ReferenceList>
</wsse:Security>
</s:Header>
<s:Body>
<xenc:EncryptedData>
<EncryptionMethod
Algorithm=http://www.w3.org/2001/04/xmlenc#tripledes-cbc/>
...
230
</xenc:EncryptedData>
</s:Body>
</s:Envelope>
WS-Security standard form a foundation upon which several other security specications rely on. The vision is to have several security components, each addressing a
unique security puzzle to work in unison with each other. The dream is still far from
reality but WS-Security provides a framework to support this unifying approach.
Figure 17 depicts the other security standards that are being built on the WS-Security
foundation.
Figure 17. WS-Security Foundation
5. WS-Policy
The Web Services Policy Framework (WS-Policy) is a specication jointly developed
by IBM, Microsoft, BEA and SAP. It species how senders and receivers can state
their requirements and capabilities regarding policies they require to conduct business
in an electronic format. The specication is highly exible and supports an extensible
grammar for expressing the capabilities, requirements and general characteristics
of entities in an XML Web Services-based system. It denes a framework and a
model for the expression of these properties as policies. A policy is expressed as
policy assertions. A policy assertion represents a capability or a requirement. Policy
assertions are dened in a companion specication (Web Services Policy Assertions
231
Language or WS-PolicyAssertions). WS-Policy expressions are associated with various

Web services components using the Web Services Policy Attachment specication
(WS-PolicyAttachment).
The following example illustrates a policy:
1 <wsp:Policy xmlns:wsse="..." xnlns:wsp="...">
2
<wsp:ExactlyOne>
3
<wsse:SecurityToken wsp:Usage="wsp:Required"
wsp:Preference="100">
4
<wsse:TokenType>wsse:Kerberosv5TGT</wsse:TokenType>
5
</wsse:SecurityToken>
6
<wsse:SecurityToken wsp:Usage="wsp:Required"
wsp:Preference="1">
7
<wsse:TokenType>wsse:X509v3</wsse:TokenType>
8
</wsse:SecurityToken>
9
</wsp:ExactlyOne>
10 <wsp:Policy>
Illustrate the expression of a security policy using assertions dened in
WS-SecurityPolicy. Lines to 10 represent a set of policy assertions for authentication.
Lines 2 to 9 represent the <wsp:ExactlyOne> policy operator that is used to group
policy assertions into policy sets. That is, a valid policy can contain any one of the
contained assertions (lines 3 to 8).
Lines 3 to 5 and 6 to 8 represent two specic security policy assertions that indicate
that two types of authentication are supported and that of the two types, Kerberos
authentication is preferred over X509 authentication.
6. WS-Trust
The Web Services Trust Language (WS-Trust) is a specication jointly developed by
IBM, Microsoft, Verisign, and RSA. In order for a secure communication between
two parties, they must exchange security credentials (either directly or indirectly).
However, each party needs to determine if they can trust the asserted credentials
of the other party. WS-Trust specications describe the model for establishing both
direct and brokered trust relationships (including third parties and intermediaries).
It provides a framework to support:
methods for issuing and exchanging security tokens, and
232
ways to establish and access the presence of trust relationships.

Typically, WS-Trust allows a client to send a request (using a X.509 certicate or any
other security token supported by WS-Security) to a Security Token Service (STS)
via a gateway. The STS maps the X.509 certicate to the security token expected
by the receiving party, for example a SAML assertion, and then returns the trusted
(i.e. signed) SAML assertion to the client. The client can now use the security token
expected by the receiving party.
7. WS-Privacy
Personal information becomes public in many ways. For instance, sharing of customer
data among partner companies, unsecured personal information stored on the
internet. WS-Privacy allows organizations to state their privacy policies and require
that incoming requests make claims about the initiators adherence to these policies.
The specication describes a model for how a privacy language may be embedded
into WS-Policy descriptions and how WS-Security may be used to associate privacy
claims with a message. WS-Privacy also describes how WS-Trust mechanisms can be
used to evaluate privacy claims for organizational practice as well as user preferences.
As many governments pass laws to protect sharing of personal information, this
specication will grow in importance.
8. WS-SecureConversation
WS-SecureConversation jointly proposed by IBM, Microsoft, RSA and Verisign,
describe how to manage and authenticate message exchanges between parties
including security context exchange and establishing and deriving session keys.
WS-SecureConveration is the SOAP layer equivalent of SSL at the transport layer.
The mechanisms dened in WS-Security provide the basic mechanisms on top
of which secure messaging can be dened. WS-Security is well suited for a single
message authentication model in which the message contains all the security attributes
necessary to authenticate itself. However, for multi-message conversation, this model
becomes inefcient since each message has to go through the same authentication /
verication process. Parties that wish to exchange multiple messages typically establish
a secure security context in which to exchange multiple messages. A security context
is shared among the communicating parties for the lifetime of a communications
association. WS-SecureConversation specication denes extensions to allow security
233
context establishment and sharing, session key derivation. The primary goals of this
specication are:
Dene how security contexts are established, and
Specify how derived keys are computed and passed.
9. WS-Federation
Web Services Federation Language (WS-Federation) is a specication jointly being
developed by IBM, Microsoft, BEA, Verisign and RSA. It will provide support for
secure propagation of identity, attribute, authentication, and authorization information. The specications describe how to manage and broker the trust relationships
in a heterogeneous federated environment including support for federated identities.
WS-Federation addresses the issue where the requestor in one trust domain interacts
with a service in another trust domain using different security models. For instance,
how does a consumer service using Kerberos invoke a producer service based on
X.509 in a trusted fashion. Liberty Alliance project and Microsoft Passport are an
attempt to solve the same issue and WS-Federation is working towards providing a
standard, generic approach to handle identity federation.
10. WS-Authorization
WS-Authorization specications describe how to manage authorization data and
authorization policies. It covers the description of how assertions can be made
within security tokens as well as how they will be interpreted by each endpoint.
This specication is designed to be exible and extensible with respect to both
authorization format and authorization language. This is important because each
security provider may have a different authorization format and language.
11. Summary
The loosely coupled nature of ESOA systems makes it easy for applications to
interoperate but it also opens the door for unauthorized access. The fragmented
234
nature of ESOA systems only adds to the problem. With a lot of industry momentum
behind ESOA, practical implementations have started to emerge in the market place
and companies have begun to realize the need and complexity of security around
these systems. The security standards mentioned in this chapter are essential for the
successful adoption of ESOA. Many of the specications are still in early stages and
it will take time for the vendors to catch up and provide robust implementations
around these specications. On the plus side, the proposed architecture is exible
enough and does not require a change to the existing security implementations.
6
SOA MANAGEMENT
Management means, in the last analysis, the substitution of thought
for brawn and muscle, of knowledge for folkways and superstition, and of
cooperation for force. It means the substitution of responsibility for
obedience to rank, and of authority of performance for the authority of rank
Peter Drucker
Service-Oriented Architectures can be deployed across corporate rewalls where

services can be accessed both internal and external to the enterprise. What makes
SOA benecial is the very thing that also makes them challenging. Once services
are accessed outside of the administrative control of the creators of the service, IT
development teams and infrastructure groups are now placed on the critical path for
all business processes that leverage that service.
Enterprise systems complexity increases over time and becomes increasingly important
that these systems and the services they offer support manageability. It is highly
desirable to dynamically view and control the state associated with all services
used within an application. Since services can be developed using a multitude
of languages, strategies surrounding management need to be provided using a
platform-independent methodology.
In this chapter, we will cover:
Systems Management,
Lifecycle Management,
Business Processes,
Architecture Management.
235
236
1. Problem Space
In previous chapters, you learned that a fundamental element of an SOA is the
contract which is a rst class citizen and which serves to dene the syntax of a service.
The contract also describes semantics of the service using comments contained within
an embedded description element or grouping of methods and/or operations. Many
discussions of SOA specify the sole contract as dening the business interface. In
reality, there are two contracts of which management is less dened within the
industry.
In web services, the contract is described within a WSDL document and describes
an interface in two ways. First, it provides an abstract description of messages and
types. Second, it provides one or more concrete descriptions that bind the abstract
denition of the service to the selected messaging transport. WSDL can be extended
to support other transport bindings, marshalling mechanisms and even proprietary
binding protocols but how this is accomplished is not well documented.
WSDL documents should be segmented to separate the business contract from the
services technical binding.
We have also learned that while the vast majority of current SOA implementations
are point-to-point, we may need for our services to interoperate in a loosely coupled
manner where producer and consumer of services have no prior knowledge of each
other. WSDL is typically created using wizard type tools that tie the producer and
consumer together since the binding information and endpoint information may
be actually contained within the WSDL document. WSDL can be used to describe
service and operation names, message parts, data types and a return value but can
also be inappropriately used for more than it was originally intended to support.
In order to further understand the problem, let us pretend you have developed a
service that provides quoting for business partners and is secured using SSL. A simple
modication of the WSDL will allow you to change the transport from HTTP to
HTTPS in the address element of the location attribute. Over time, this service
may become used by a multitude of parties and becomes successful. After the service
reaches critical mass, the head of the corporate security department creates a new
policy where all electronic communications with third parties must be encrypted
using either 168-bit Triple DES or RC4 algorithms. The problem now becomes
harder than simply changing a single element in WSDL.
Chapter 6: SOA Management
237
Figure 1. Find Bind and Execute

Solving this problem may require opening your favorite SOAP development kit and
creating and/or substituting a new encryption layer. This task becomes difcult to
realize for all but the most talented developers in our industry. Even if you have
gured out a clever way to modify your SOAP stack to support the new enterprise
security requirements, what about all the other service consumers who use this service
but will have their own release schedules of when they can support messages using the
newly chosen algorithm. In many enterprises, the business customer will demand that
security be downgraded to support protocols that are easier to implement for external
parties and suggest using username and password using HTTP digest mechanisms.
Now you have to support two different approaches to security and the endless loop
has begun.
Administrators within many enterprises are responsible for keeping the systems
running and ensuring the overall health of the system at run-time. This may include
determining the optimal number of instances of a service that should be running at any
one time, whether the service has stopped accepting requests and so on. IT operations
are usually the rst to respond with violations of service level agreements and/or
complaints from customers related to response times. Traditional tools provided by
network management vendors currently are limited to understanding network and
operating system metrics alone.
Conventional applications periodically encounter run-time errors that will show up
either in a log le and/or operators console that will allow operations to diagnose how
to solve the problem. Services however, generally do not have the same benet as they
may be silently implemented within the bowels of its hosting application. Visibility
238
into the internals of an application is sometimes warranted. The ideal situation is

the ability for the enterprise to have a console that can allow for management of all
service-oriented applications.
SOA also encourages reuse of services for internal consumption. Laws tied to legislative
acts such as Sarbanes Oxley, Graham Leach, and Bliley may force authentication of
access to consumers internal to the enterprise. The enterprise may require internal
service consumers to authenticate to the service using client-side certicates using PKI
technologies. A pattern emerges; an SOA starts out loosely coupled but mysteriously
and quickly becomes tightly coupled.
The example previously described hints towards issues that can arise for service
providers but can be extended to service consumers as well. Hopefully, you are now
asking yourself in the above scenarios how would you implement these options into
a service as well as support future management requirements without jeopardizing
existing service dependencies.
The problem space is not limited to just security concerns but can become pervasive
problems for other aspects of the infrastructure. Pressure surrounding business
reporting and controls surrounding legislative concerns and compliance to trade
agreements within a service-oriented architecture are equally challenging. Sarbanes
Oxley requires that both the Chief Executive Ofcer (CEO) and Chief Financial
Ofcer (CFO) must certify the appropriateness of the nancial statements and
disclosures contained in the (annual) report, and that those nancial statements
and disclosures fairly present, in all material respects, the operations and nancial
condition of the issuer and will be held personally liable for willful violations of the
section. This clearly places the responsibility of accurate controls of services on the
CEO. Developers simply adding in code to each service may no longer be acceptable.
If one were to analyze Sarbanes Oxley and its impact on a service-oriented architecture,
one would include that services cannot be simply left to grow organically and must
be supported by a management strategy. In Table 1 we look at elements of Sarbanes
Oxley and the requirements it brings to managing an SOA.
So far, we have outlined problems with rolling out service-oriented architectures Now
that the problem space of services is starting to emerge, lets start exploring scenarios
in which one can start managing enterprise SOAs.
239
Table 1. Sarbanes Oxley Effect on Service-Oriented Architectures

Compliance Requirements
Management Problem
Section 302 Corporate Responsibility for Financial Audits
The CEO and CFO need to attest that
the nancial statement and disclosures are
accurate.
Signing ofcers are responsible for establishing appropriate internal controls.
Ability to dene enterprise-wide audit trails

and comprehensive logging of all service
transactions.
Ability to create processes to ensure that
all requirements related to compliance and
performance are automatically handled.
Ability to enforce segregation of service interactions (where appropriate) and establish
that all corporate information policies are
fully adhered to.
Section 401 Disclosures in Periodic Reports

Adequate and comprehensive disclosures
are required where relationships with customers and/or suppliers are disclosed where
there may be some conict, etc.
The ability to provide a quick view of

potential deals that need to be reported on
nancial reports.
The ability to provide variant alerts based
on certain transaction types and/or deals
with specic customers/partners.
Tracking of all types of service requests for
a stakeholder that is deemed important.
Ability to track all forms of electronic communications in a non-reputable manner.
Section 404 Management Assessment of Internal Controls

Each annual report must include an
internal control report, which has to:
State that management is responsible
for creating and maintaining an adequate internal control structure and
procedures for nancial reporting.
Contain an assessment of the effectiveness of the internal control
structure and the procedures of the
issuer for nancial reporting.
The companys auditor is required to

attest to and report on the companys
internal control report as part of the
annual report.
The ability to determine service routing

rules based on dened monetary values
where specic approvals are required.
The ability to dene a managed approval
chain.
The ability to dene specialized audit trails
that captures the approval and/or rejection of service transactions and providing
centralized transaction history.
Comprehensive ad-hoc reporting.
240

Table 1. Continued
Compliance Requirements
Management Problem
Section 409 Real Time Issuer Disclosures
Issuers must disclose information on material changes in the nancial condition or

operations of the issuer on a rapid or current
basis.
Information used for reporting purposes

must be accessible in a near real-time
manner.
Section 802 Records Management and Retention

Provisions related to the destruction, alternation or falsication of records in federal
investigations and bankruptcy.
Strategy surrounding non-repudiation of

all service invocations and logging and
auditing related services.
Fine and/or imprisonment of not more than

20 years for whoever knowingly alters,
destroys, mutilates, conceals, covers up, falsies, or makes a false entry in any record,
document or tangible object with the intent
to impede, obstruct or inuence.
Comprehensive workow that denes both

happy path use cases as well as exception
handling in a centralized manner.
Audit records must be centralized and have
archival and retention strategies.
Section 906 Corporate Responsibility for Financial Reports

The CEO and CFO must each certify the
material representation of the companys
nancial condition in all nancial reporting.
Clearly dened system of records for all

nancial transactions.
1.1. Management Scenarios

The ability to management services can be accomplished in a variety of manners. A
basic model principle may be for applications and services to write log information
to a le or event logging service, which is useful in diagnostics but requires manual
intervention in most situations in order to determine whether errors have occurred
that require attention. In the vast majority of enterprises, log les are typically not
looked at until the service is down; although it may contain useful clues that could
have prevented an outage in the rst place.
Another approach that can be used is to ensure that all services within the enterprise
implement one or more common management interfaces that will allow external
management consoles to query the service for its status. A good strategy that can
be incorporated into the enterprise architecture will be for services to implement
an interface for application level pings, returning the service version and the ability
241
to shutdown a service remotely. Let us look at representative WSDL snippets that

outline this concept:
<?xml version="1.0" encoding="UTF-8"?>
<definitions
name="Quote"
targetNamespace="http://www.canaxia.com/Quote.wsdl"
xmlns="http://schemas.xmlsoap.org/wsdl/"
xmlns:soap="http://schemas.xmlsoap.org/wsdl/soap/"
xmlns:tns="http://www.canaxia.com/Quote.wsdl"
xmlns:xsd="http://www.w3.org/2001/XMLSchema"
xmlns:xsd1="http://www.canaxia.com/Quote.xsd1">
<types>
<xsd:schema
targetNamespace="http://www.canaxia.com/Quote.xsd1"
xmlns="http://schemas.xmlsoap.org/wsdl/"
xmlns:SOAP-ENC="http://schemas.xmlsoap.org/soap/
encoding/"
xmlns:xsd="http://www.w3.org/2001/XMLSchema"
xmlns:xsd1="http://www.canaxia.com/Quote.xsd1">
</xsd:schema>
</types>
<message name="ShutdownResponse">
</message>
<message name="ShutdownRequest">
<part name="username" type="xsd:string"/>
<part name="password" type="xsd:string"/>
</message>
<message name="VersionResponse">
</message>
<message name="PingResponse">
</message>
<portType name="ManagementPortType">
<operation name="Ping">
<documentation>Verify that the service is listening
and ready to accept messages</documentation>
<output message="tns:PingResponse"/>
</operation>
<operation name="Shutdown">
<documentation>Shuts down the service</documentation>
<input message="tns:ShutdownRequest"/>
<output message="tns:ShutdownResponse"/>
</operation>
<operation name="Version">
<documentation>Displays the version of
the service</documentation>
242

<output message="tns:VersionResponse"/>
</operation>
</portType>
<binding name="ManagementPortTypeBinding"
type="tns:ManagementPortType">
<soap:binding style="rpc"
transport="http://schemas.xmlsoap.org/soap/http"/>
<operation name="Ping">
<soap:operation soapAction="capeconnect:Quote:
ManagementPortType#Ping"/>
<input>
<soap:body use="literal"/>
</input>
<output>
</output>
</operation>
<operation name="Shutdown">
ManagementPortType#Shutdown"/>
<input>
</input>
<output>
</output>
</operation>
<operation name="Version">
ManagementPortType#Version"/>
<input>
</input>
<output>
</output>
</operation>
</binding>
<service name="Quote">
<port binding="tns:QuoteBinding" name="QuotePort">
<soap:address
location="http://localhost:8000/ccx/Quote"/>
243
</port>
</service>
</definitions>
Code 1. WSDL to dene management interfaces
Within the WSDL document, we dened the shutdown operation as taking a
username and password to prevent arbitrary shutdown by unauthorized personnel as
well as a version service that can return meta-information about the service itself.
This approach allows application developers to include entry-level management
capabilities into their services by implementing the enterprise dened service
interfaces. Common interfaces used consistently throughout the enterprise makes
the task of adding management capabilities easily scoped and, more importantly,
repeatable. An enterprise can begin to create their own framework that can proactively
check the status of a service via simple polling constructs.
The proposed service interfaces are useful in creating a strategy whereby internal
management applications can be custom developed to understand the state of the
services network and while an improvement over simple logging, it will still introduce
problems of their own including who can call the services, in what context and how
often.
The real question that should be asked is whether application developers should
be implementing management functionality into services at construction time or
whether this concern should be externally managed. Enterprises are under constant
pressure to externalize systems, deliver quicker. This requires IT to leverage agile
methods for developing software and deferring many tasks to later phases that never
actually happen. Development teams would love visibility into the run-time behavior
of their service offerings but usually their requirements come second to business
demand.
Teams may be able to develop and deploy services quickly without management
capabilities. However, this often leads to disruptions of new development and missed
deadlines because the teams must triage and diagnose run-time service problems with
very little real-time information. The enterprise should consider approaches where
management is externalized from the services themselves.
The ideal abstraction would suggest having an enterprise framework that has
responsibility for management of services. A robust framework should provide
support for answering the following questions regarding management of services:
What services have to be managed?
What are the properties of the service?
244
Figure 2. Management Separation of Concerns (OSI Model)

What if some service(s) fail?
How to properly secure and manage security of services?
Protocols used for exchanging management information including operations,
notications and protocols?
Relationships between managed services?
The ability to provide management of services is best realized by using a frameworkbased approach that can encapsulate the interactions with sets of managed services.
The framework itself may be reminiscent of a typical object-oriented system in that
services under management are analogous to an object hierarchy.
2. Systems Management
Managing service-oriented architectures can be thought of as a new layer of
functionality on top of traditional management frameworks (i.e. Tivoli, BMC,
CA/Unicenter) that can understand and interact with the underlying infrastructure
and provide a mapping layer to how services behave on it. Traditional management
infrastructures provide visible and control of the underlying infrastructure but do not
245
provide management at the service request level which maps directly to a business
process.
A framework that understands services can provide additional value over traditional
management infrastructures in that it can not only alert but can make informed
decisions and provide alternatives to increase service reliability. Service-oriented
management frameworks can permit or deny service requests based on the identity
of the requestor, its content and/or its context. Frameworks in this space can reroute
requests to alternative service providers, provide load balancing across servers or
even data centers, transform messages and provide analytics surrounding run-time
performance and utilization.
Systems Management is comprised of the following components:
Logging,
Auditing,
Monitoring,
Alerting,
Provisioning.
Management of XML-based services is enabled via the principle of introspection.
While it is possible to have services that do not use XML, without the ability to
inject into the architecture third party components become difcult to manage and
usually require serialization, rebuilding and recompiling to change. Introspection of
XML can occur at run-time with default implementations and can be used to rapidly
instrument services without the service implementation changing.
2.1. Logging
Access to services from consuming applications should be logged at a minimum. These
logs can be used for various operational processes including services provisioning,
monitoring, performance evaluations, and business trend analysis.
Listed in Table 2 are some of the elements that we believe should be centrally logged
upon every service invocation.
Corporate security policies and software development standards should dictate which
elements of a log message are required vs. optional. From a software development
perspective, the framework component that handles capturing of logging records
246

Table 2. Logging Properties
Property
Service Name
Data type
String
Description
Name of the service
Service Protocol
Credential
String
String
Credential Token
String
Transport protocol used to invoke the service.

Credential of the calling service consumer. Could be username,
thumbprint or other credential type.
If service request required authentication and/or authorization,
the log should store the unique identier provided by the
security service.
Request Start
Timestamp
Request Finish
Timestamp
Log Timestamp
Timestamp
Timestamp
In order to achieve scalability, it may be necessary to

defer logging to background processes while retaining full
traceability.
Request
sage
Mes-
Blob
Request message may exceed the limits of typical datatypes

especially if store used is a relational database.
Response Message
Severity
Blob
Same as above. Ideally store could shred request/response

XML into normalized elds for reporting purposes.
If service invocation resulted in an error, may want to indicate
problem. Severity may be indicated as follows:
Error
Timestamp
String
Time the service request was started. Should be logged to the

100 millisecond.
Time the service request was completed.
Warning
Critical
Informational
Diagnostic
Signature
Binary String
Field used to store digital signature of log entry and can be

used for non-repudiation purposes. Ideally it will be signed
using an algorithm that supports one-way encryption such as
MD5 or SHA1.
needs to be built with more system qualities than anything else within the enterprise
as it must reliably keep records of security-related and business transaction events.
Minimally, the logs will serve as a useful audit trail and can be used to conrm
compliance with government regulations and internal enterprise security policies,
detect breaches in security and helping diagnose production problems.
247
2.2. Auditing
In accounting, an audit trail provides a recorded log of sequences that help in
either validating or invalidating accounting entries. For services, the audit log may
track all changes to policies, starting and stopping security related services or other
administrative actions. Auditing can be considered an extension to logging but is
also used for different purposes. Logging often contains diagnostic information and
information regarding service requests while audit logs contain information about the
management framework itself in order to maintain higher order conceptual integrity.
Auditing also requires dependencies on external services in order to maintain integrity.
Some of the services required are given in Table 3.
Table 3. Auditing Requirements
Service
Description
Public Key Infrastructure
Audit logs may require non-repudiation. Use of digital signatures

may be used as a mechanism to accomplish this.
If the audit logs use timestamps, a trusted time service is required.
Usually this time service will be indexed to the Naval Observatory
or other Stratum 1 time service.
The audit information will ultimately need to be persisted in a
form of secured database.
Secure Time Service
Secure Storage
In reality, auditing is a hybrid requirement that is someone between logging and

security and shares characteristics with both depending upon its context. Let us look
at the differences:
Auditing assists in the detection and avoidance of violations in security related
policies, while security is applied in order to prevent future denials of service
consumer, producer or administration actions.
Logging is passive and ideally accomplished out-of-band from the service
request, where auditing should always be handled in-band and fail the entire
service transaction if the auditing record cannot be created.
Auditing is active in that it can react immediately to violations such as locking
a user account, where actions may be invoked on logging entries but will be
handled at a later time. Actions on log entries is often manual, so no automated
action can be taken on it.
Auditing records are usually centralized and may have daemon processes that
perform checks on them periodically, while logging can be accomplished
decentralized and replicated centrally in the future for analysis.
248
For more information on network time protocols and stratum, see Internet RFC
1769.
2.3. Monitoring
The ability to determine the health of the services network is vital to run-time
reliability. Operational snapshots that show the health of all services by averaging
values from the services connected through a single component (gateway or agent)
is mandatory. Usually, this information is best rendered in the form of a console
that displays the health of the selected component. Components can be optionally
compared to their specied service level agreement (SLA). An ideal approach is to
color key the display as shown in Table 4.
Table 4. Monitoring Color Key
Color
Green
Run-time Description
Service(s) is experiencing no errors.
SLA Description
Service(s) performing within SLA.
Yellow
Service(s) is experiencing errors, user experience not impacted. Service may have
failover or been re-routed.
Service(s) performing close to limit of

SLA.
Red
User experience is impacted. Services are

returning errors.
Service(s) performing outside limit of

SLA.
The ability to view run-time information on various components and services within
specied time periods allows one to quickly determine the overall health of the
services network and take action accordingly. Ideally, the management framework
should provide other metrics and views, see Table 5.
Table 5. Service Metrics
Metric
Downtime
Description
Information when the service is considered no longer able to process service
requests.
Execution Failure
Metrics expressed as a percentage/count of the number of successful service

executions vs number of service failures.
The amount of processing time overhead the policies have added to the to
service requests.
Number of service invocations that are failing due to security
considerations.
Latency
Access Violations
If the service is accessed via a gateway, downtime can show the start and end times
of when the specied service started to fail and then started again to successfully
249
process service requests. For agents, this can indicate the amount of time where a
centralized facility that performs service-level pings has not completed. Execution
failures should be specied as a percentage for high-volume services and as a count
for low-volume services. For example, if the Yahoo Stock Quote service has a billion
requests per day, having a threshold count will create either too many false positives
or not show that anything is wrong. A better way in this situation may be to indicate
that monitoring should show yellow if the number of failed invocations goes over
1% within core business-hours and red if over 2%.
Latency is an expression of how much time it takes for a request to travel from one
designated point to another. A service will have zero latency if all data in the request
can be transmitted instantly between one point and another. Adding gateways and
agents will introduce application-level latency but this will also occur based on the
underlying network topology.
Latency of service invocations occurs due to several factors including:
Propagation,
Transmission,
Routing, and
Miscellaneous processing.
Propagation is the time it takes for a service request and response to travel between
producer and consumer respectively. Transmission of service requests occurs over a
multitude of physical network components (i.e. routers, ber, T1s, ATM, and so on)
that introduce their own delays. Service components such as gateways and agents may
need to examine the payload to determine next hop or provide additional processing
such as transformation and/or validating the service request came from an authorized
user.
High latency in routing and propagation tend to indicate performance bottlenecks
in the infrastructure while latency in routing and miscellaneous processing usually
occur due to sub-optimal service run-time resources, bad services design or service
components that require optimization. Latency can be improved by implementing
pre-fetching (anticipating the need for supplemental information) and introducing
multi-threading constructs that will allow processing to occur in parallel across
multiple threads of execution.
Access violations are useful in determining whether the services network is receiving
an increased amount of failed service invocations due to security considerations and
will help determine service-level denial-of-service, malformed requests or other forms
of attacks.
250
3. Alerting
Business alerts allow you to specify an action in response to a specic type of business
event. Alerts provide notication when a service is not meeting its service-level
agreement or not seeing the level of service requests expected. An alert can watch
for specic types of business events and initiate an action in response. Alerts can be
created on both message trafc patterns and other forms of system and aggregated
data.
Alerts can be created to notify individuals and distribution lists in a variety of manners
as well as invoke predetermined services. Some of the possible alerting facilities that
should be supported in a framework include:
Email notication (user and/or distribution list),
SNMP Traps, and
Service invocation.
Alerting can also be used for technical considerations. It may make sense for the
operations staff to classify all alerts into the following alert categories:
Response time,
Transaction size,
System fault, and
Trending.
3.1. Round Trip

Response time for services measure the total elapsed time from the detection of the
service request to the completion of the response to the service request. This is known
as measuring the round trip execution time for service requests. This is useful in
detecting situations where high-volume services such as a real-time stock quoting
service exceeds its desired response time.
3.2. Transaction Size

The amount of data transmitted as part of a service request may give hints to
identifying operations that sent or receive large amounts of information. The size
251
of the payload could have the effect of slowing down overall service response times
especially in situations where XML is used. It may be useful to operations personnel
to be notied in situations where an individual service request sent a 10 MB XML
document where the typical document is 1 MB. Transaction size may also be used in
aggregation situations for operations personnel. For example, you may desire to be
notied if the average request size during any one-hour period has exceeded 50 MB.
3.3. System Fault

Alerts can be useful for tracking the availability and operational status of individual
services. Alerts may be sent if a service request could not be processed correctly
either due to the service not responding or the service returning a fault. Management
frameworks may optionally implement a heartbeat monitor that periodically directs
a predened service request to the service and alerts upon failure.
3.4. Trending
Over time service request transaction times may deviate from expected values. Using
threshold scenarios in alerting situations can sometimes result in not getting alerted
when it is important to or on the other side receiving too many false positives. One
solution to this problem is the introduction of non-threshold-based analysis. One of
the more popular approaches is the use of Bayesian Belief Networks.
Bayes theorem is named after the Reverend Thomas Bayes who worked on the
problem of computing a distribution for the parameter of a binomial distribution.
In simple terms, he dened a mathematical formula based on probability inference,
which is a means of calculating, from the number of times an event has not occurred,
the probability that it will occur in the future.
Bayesian belief networks are used to develop knowledge-based management frameworks where usage patterns are characterized by inherent uncertainty. The ability to
apply reasoning with uncertainty sometimes overlaps with conventional knowledgebased capture techniques. For example, a stock quoting service may receive an average
of 1,000 requests per minute on days where the stock market is open, say Monday thru
Friday from 9am to 4pm eastern time excluding holidays. During this period, if the
number of requests goes over 5,000 an alert can be set using traditional approaches,
but what happens on Saturdays if the request volume is 750 requests per minute
when weekend trafc is lower? Now imagine the alert threshold being correlated to
news such as peace in the Middle East. You could not dene the alerting criteria
in advance but the system could learn trading volumes and take into considerations
252
trends such as increased trading volume near year end or even inverse patterns such
as the number of buy orders is dropping but the number of sell orders is increasing.
For more information on Thomas Bayes
http://www.bayesian.org/bayesian/bayes.html.
and
Bayesian
Beliefs,
see:
4. Provisioning
A management framework should be responsible for dening services behavior based
on its exposed business and technical functionality. Service consumers in enterprises
that implement per-usage chargeback mechanisms or for external service consumers
who are charged for usage should use a management framework that supports a
metering construct. Service usage can be billed based on usage/subscription patterns.
The logic for provisioning services should be built into a subscription engine that
exposes itself as a service. Service consumers can search for services against registries
where they have been given access and then apply for a subscription to the desired
services. Likewise, subscription policies can be dened as metadata in the registry in
which the management framework is responsible for enforcing.
Figure 3. Service Invocation Producer Policies

Ultimately service consumers will access services based upon the contract based upon
their subscription. In order to effectively bill for usage, the framework will need to be
aware of the statistics of the contract, committed service levels and implied qualities
of services. When services are consumed, a separate function of the framework that
handles metering may allow/disallow access to the service, throttle the number of
253
requests within a specic time period and ultimately will create input to be used by a
billing service.
Figure 4. Service Invocation External
5. Leasing
A lease is a grant of guaranteed access over a time period. When the lease runs out,
the consumer must request a new one. Leases are negotiated between consumers
and the provider of the service as part of the service contract. Leasing helps prevent
stale caching information from queries to UDDI registries and other locator-based
services. The concept of a lease is necessary for SOA when services need to maintain
the state of information about the binding between consumers and producers. The
lease denes the amount of time for which the state may be maintained. Leasing
reduces coupling between service consumers and service producers by limiting the
amount of time consumers and providers may be bound. Without a leasing construct,
consumers could bind to a service forever and never rebind to its contract again.
When service producers need to change the implementation of a service, it can do
so when the leases held by consumers expire. This allows service implementations
to change without affecting the execution of the service consumers, because those
consumers must request a new contract and lease. When the new contract and lease
are obtained, they are not guaranteed to be identical to the previous ones.
254
Leases can be implemented so that they are either exclusive or non-exclusive. Exclusive
leases insure that no one else may access the resource during the period of the lease;
non-exclusive leases allow multiple service consumers to invoke the same service.
Leases can also be implemented using nite or innite constructs. A nite lease can
be used immediately or based on a future lease. Finite leases require service providers
to dene the exact period for which the service should be made available for discovery
in the registry. The lease period is restricted by the maximum allowable lease period
dened by the administrator.
Future leases allow a service provider to make the service discoverable once the lease
has been activated. Innite leases allow for the maximum allowable time for the lease
to change. This is useful in circumstances where a service provider wants to ask for
service consumers to periodically check back for changes in service contracts. Usually
innite leases should be constrained based on either a time period or ratio.
Leasing can be integrated into the provisioning system in which the elements are static
upon each request. This would require knowledge of how to handle leasing terms
at the service proxy and gateway. Alternatively, leasing can be implemented at the
gateway itself as a type of policy. This would allow consumers that do not understand
leasing constructs to function while maintaining loose coupling. Leasing can also be
tied into components that provide authentication, encryption and non-repudiation.
6. Billing
The billing service is responsible for creating and storing billing records. It has the
responsibility of creating billing entries into a separate billing provider system. For
internal usage, the billing service can create billing entries for chargeback purposes
into general ledger applications such as Peoplesoft, SAP or Oracle Financials. Table 6
shows the elements of a typical billing record.
Ideally, the billing service should expose an interface that can provide up to date
billing information for service consumers. The billing provider should physically
separate this interface from the billing services interface that is used to store billing
records as well as the interfaces actually used. It may be a good idea to consider
separating consumer access (query) from billing creation in order to deter fraudulent
access.
255
Table 6. Recommended Elements of a Billing Record

Element
User
Description
The user ID of the service requestor.
Record Date
Duration
Date and time of service usage.

Duration of service usage. Can be a count if aggregated, CPU time
or similar metric.
A pointer to the contract (provisioning record if applicable) which
governed service access.
Reference
Service Name
Host
Name of the service(s) consumed.

Host unique identier that the service executed on.
7. Pricing/Chargeback Models
Service level agreements establish the foundation for pricing and chargeback models.
Within most business models there will be services offered with no quality of service
guarantee and therefore may be freely available. Likewise, services that are used
in business transactions where quality of service is important, then pricing and
chargeback models become applicable.
In order to support effective pricing models, identifying the service consumer in a
non-repudiated manner may be required and may optionally support the usage of
digitally signed XML messages. The ability to collect revenue on services helps fund
future service enhancements and fund operations.
Several models for pricing & charge backs for services include:
Per User,
Per Transaction,
Percentage of Revenue,
Fixed Fee/Subscription,
Lease/License,
Business Partnership, and
Registration.
While decisions based on pricing models can be made at service construction time,
embedding this type of information within each service minimally will require
256
breaking of the provided service interfaces. If multiple models are employed, solving
for this at design-time becomes a fragile value proposition. Let us look at each of the
models in detail.
7.1. Per Transaction

Fee per use models are the most primitive of all commerce and are based upon a
pre-established relationship between two trading partners. In this model, the notion
of a charge per transaction can be implemented using payment instruments such
as debit and credit cards. Service consumers may be given discounts for various
transaction volume levels.
7.2. Fixed Fee/Subscription

Subscription models depend upon the establishment of user accounts and supporting
the appropriate validation mechanisms as part of the service request. This support
can be incorporated into an overall provisioning strategy where users specify
their anticipated service usage. A service provider may desire to create tiers of
membership levels to allow for classication of service consumers. Architecturally,
it becomes important to consider whether this model happens within the service
itself, externalized to a management framework and/or handled through out-of-band
provisioning and reconciliation.
Service providers may charge additional fees, such as a management fee that covers
the costs associated with operations personnel, bandwidth, hardware and other data
center charges. Fees for support services are also common.
7.3. Lease/License
Leasing and licensing are common for large volume usage scenarios and are usually
based upon customized agreements. Service providers may charge based on the
number of service requests or the volume of requesting components (i.e. seats)
within the service requestors infrastructure. Leasing and licensing always occur via
out-of-band provisioning mechanisms.
257
7.4. Business Partnership/Percentage of

Revenue
A visit to many software vendors web sites will uncover the notion of partnering
whereby a vendor will establish mutual usage agreements with other vendors in order
to increase the number of customers for reference purposes and as a money saving
mechanism. Bartering of services can be done based on equality or a percentage of
gross revenue (i.e. link referral model) of the requestor in third-party service branding
scenarios.
Percentage of revenue (i.e. slice of the pie) is similar to the per-transaction model in
that the amount charged to service consumers may uctuate based each billing cycle.
The difference between the per-transaction model and percentage of revenue is the
models price calculations correlate to the amount of revenue the service consumer
generates each billing cycle.
7.5. Registration
Many web search engines now charge for registering Internet web sites in order to
for the site to be generally accessible. The same thinking can be applied to services
that will be externally accessed. One model may be to use a management framework
that intercepts publish requests to a UDDI registry and charges based on a pay to be
viewed concept as shown in Figure 5. This model assumes that if a service provider
wants to offer services, they will be willing to pay a registration fee.
The pricing/chargeback models discussed in this chapter are generally applicable to
any market segment and/or industry vertical.
8. Lifecycle Management
Traditional lifecycle processes focus on methodologies used to support developers
during the development process but fall short in considering what happens at
deployment time. A complete strategy should include all aspects of a services lifecycle
and follow the path of a service from its conception to its retirement. Lifecycles for
services really have two viewpoints, the development aspect which is developer driven
and the operational aspect which is driven by operations personnel.
258
Figure 5. UDDI Registry Management

Operations staff typically manages deployment and run-time aspects of services and
develop plans for reacting quickly to deployment related problems. This strategy
may include rolling back services or in situations of disaster routing request to
alternate providers. Ongoing support for services may include infrastructure tuning
and stability improvements to help services meet service-level agreements. Over time
services will need to be retired or wholesale replaced. Usually this is the most difcult
discipline to accomplish since enterprises tend to be reluctant to change especially
when services still meet their needs. For example, there is a large portion of end-users
who still run Windows NT and Windows 2000 although its replacement has been
available for several years. The management strategy must include identication and
planning to support service obsolescence.
Lifecycle management for services can be realized in a multitude of ways and are
comprised of the following aspects:
Routing,
Transformation,
Versioning,
Deprecation,
259
Provisioning, and
Quality Assurance.
8.1. Routing
One of the tenets to Service-Oriented infrastructure (SOI) will be to eliminate single
points of failure within the enterprise. Usually this is accomplished by making sure
that a particular service has more than one instance either on a single machine and/or
on different machines. The services can also be spread across different geographic
locations. Ideally, location transparency should be incorporated into the architecture
so that clients are unaware to the physical location of each service and maintains a
single stable endpoint address.
A management framework may use load-balancing approach to make sure that
requests are evenly distributed across instances of a particular service. Different
service reception points may share a common queue provided by the framework
in which downstream service endpoints may be invoked in a round-robin fashion.
Invocation may also contain pluggable strategies to talk into account algorithms such
as afnity, weighted averaging, availability and quiescence.
To quiesce is to put a service or other computer resource into a temporarily inactive
state whereby in-progress requests are serviced but new requests are denied.
Support for afnity is important in two different situations. The rst usage of afnity
is to support service requests that are stateful by nature. Stateful services are highly
discouraged but the management framework should support it anyway. The second
usage for afnity is to increase performance of services where it may internally cache
access to downstream data sources such as relational databases, le systems or other
resources. An advanced form of afnity is known as Class afnity where groups of
services say from a particular user population (i.e. VIP customers, partners) may be
routed to a different set of services. Ideally, afnity will be dened on a per service
basis in a weak manner (for stateless services) where the vast majority of requests
for afnity are desired but not guaranteed or strongly where it is. Strong afnity
will return an error condition back to the consumer in case the target server cannot
handle the request.
Afnity is an approach to load balancing that directs multiple incoming requests
from the same consumer are always to the same producer. Service afnity disables
workload management after initial service invocation by forcing requests to always
return to the initial connection endpoint.
260
Management frameworks may support advanced load balancing where consumers

are balanced across multiple physical services. If one or services (grouping) have
zero consumers, then those services alone are considered for load balancing. One all
the services within the group have at least one active consumer, then standard load
balancing approaches are used.
8.2. Versioning and Deprecation

Change management is one of the most difcult disciplines to master. Serviceoriented architectures compound the problem in that traditional disciplines usually
assume that producers and consumers are under the same administrative control.
Once a service has been deployed to production and has gained widespread usage,
opportunities for improvement will emerge. The difculty is how to support change
to production services without forcing all consumers to upgrade at the same time.
Forced upgrades simply do not work in service-oriented architectures as the principle
of decoupling producer from consumer has the effect of separation of concerns from
a design perspective but does not address run-time considerations. In production
environments, it is a requirement that different versions of a service be simultaneously
supported. A management framework should provide a public interface to each
service and optionally upgrade/downgrade service requests to backend services as
appropriate. This strategy will allow for decommissioning older versions of a service
without consumer disruption.
Several strategies emerge to support this problem space. At development time,
if interfaces are made immutable then you can upgrade components and other
services that leverage components without breaking code that relies on old version.
Microsofts COM takes an interesting approach in that it assigns each interface a
globally unique identier (GUID). A developer cannot modify an interface and
therefore must create a new one if they wish to modify the behavior. This means that
once a client is compiled using the specied interface, the developer can be assured
that the interface will never change.
A GUID is a 128-bit integer that is unique across both space and time.
If the interface uses a document-style approach where an arbitrary number of
parameters can be passed using either an XML tagged approach as a string and/or
strongly typed object such as a property bag, then versioning can be implemented
261
inside of the service itself. Let us look at a code snippet in Java that demonstrates this
concept:
public class OrderDTO extends VectorDTO {
private static final long serialVersionUID = 1;
protected Vector bag;
}
Code 2. Data transfer object
Let us now extend to OrderDTO to contain the currency type used when placing an
order:
public class OrderDTO extends VectorDTO {
private static final long serialVersionUID = 2;
protected Vector bag;
public Currency typeOfMoney;
}
Code 3. Extending a DTO
The DTO can be passed to a method that may have the following signature:
public Boolean Process(OrderDTO order)
Within the process method, you can determine if the DTO is the version you expect by
using a method that exposes the value of serialVersionUID and can optionally upgrade
the interface as appropriate without forcing all clients to recompile. Development
time changes can be accomplished with you have administrative control over both
the client and the server but otherwise is limiting. Versioning as a problem space is
not limited to strictly producer side concerns but can also occur if you are a consumer
that needs to talk to an upgraded service but cannot modify your own code.
Versioning of services has similarities with traditional object-based approaches but also
varies in subtle ways. A management framework attempts to externalize versioning
concerns out of code and keeping this information centralized. By having versioning
handled by a framework, it affords several opportunities including:
Ability to perform version upgrades of services without reducing service
availability.
Hot deployment and maintenance of new parallel version of services.
Creation and support of multiple mirrors support each version of the service.
The ability to commission and decommission servers in both provider and
consumer modes.
262
Versioning will also afford the enterprise the ability to combine versioning constructs
with provisioning constructs. For example, a free version of the service could be
offered that supports an older version of a service while paid customers can access
newer versions of the service. The ability to establish version-based access control
is especially important for high-volume consumer driven architectures. In this same
situations, the ability to externalize versioning information within the application will
allow for centralized notication to consumers when newer versions of the service are
available and/or when currently supported services will be deprecated.
Service-oriented architectures that leverage XML as the transport make versioning
easier since a framework can provide introspection capabilities on the payload and
upgrade/downgrade as appropriate through transformation constructs. If your SOA
uses binary protocols such as classes, an intermediary may be capable (although not
usually the case) of converting objects from one type to another. It may be useful to
provide routing to the service that has the highest numerical version unless the client
species otherwise.
Upgrading of interfaces can be accomplished by software that supports either/both the
Proxy Pattern or Intercepting Filter Pattern. These patterns can also for the creation
of framework components that can intercept the message upon service invocation by
changing the message in-ight while passing the request to and from the service.
For more information on patterns, please see Gamma et al. (1994). The Intercepting
Filter pattern is covered in detail in Alur et al. (2003).
Use of document-style services is order of magnitudes easier to implement versioning
constructs over document style approaches. Document style allows for the usage of
XML and allows for situations in which versioning is not so direct. Since XML
processing allows for addition of elements, older clients will simply ignore them.
Document-style enables versioning of interfaces that use queuing approaches, since
elements may need to be accessed externally to the document itself such as priority.
8.3. Transformation
Transformations of XML payloads by management frameworks allow services to
bridge the information models and business processes within composable services
architecture. If you ask yourself, how do I get the message into the right format for
its next destination, the requirement for transformation becomes clearer.
The management framework should provide the ability to handle complex transformation of messages and the ability to map between message formats. In services that
263
are purchased off the shelf, the vocabulary used could be wide-ranging. For example,
elds named given name and surname may need to be converted to elds named rst
name and last name in order to be processed by downstream services.
Messages may also be required to be broken into a number of new messages each
with their own routing. Let us look at one example:
<?xml version="1.0"?>
<!DOCTYPE ORDER SYSTEM "order.dtd">
<?xml-stylesheet type="text/css" href="xmlorderpart.css"?>
<Orders>
<Customer>
<CustNumber>BICHE1</CustNumber>
<CompanyName>Budhoorams Auto Body</CompanyName>
<ContactName>Khaimraj Kelon</ContactName>
<ContactTitle>CEO</ContactTitle>
<CountryCode>TT</CountryCode>
<Phone>868-666-0123</Phone>
</Customer>
...
<Customer>
<CustNumber>SHOOP1</CustNumber>
<CompanyName>Little Red Zuper Car</CompanyName>
<ContactName>James Franklin</ContactName>
<CountryCode>US</CountryCode>
<Phone>860-242-8050</Phone>
</Customer>
</Orders>
Code 4. Composite order message
In the above code, we have a composite order that contains two different customers
in different countries (Trinidad and the United States). In this situation, we may
need to route the request to two different services, which will require breaking up the
document into two distinct messages as follows:
<Orders>
<Customer>
<CustNumber>BICHE1</CustNumber>
264
<CompanyName>Budhoorams Auto Body</CompanyName>

<ContactName>Khaimraj Kelon</ContactName>
<CountryCode>TT</CountryCode>
<Phone>868-666-0123</Phone>
</Customer>
</Orders>
Code 5. Order message Trinidad
<Orders>
<Customer>
<CustNumber>SHOOP1</CustNumber>
<CompanyName>Little Red Zuper Car</CompanyName>
<ContactName>James Franklin</ContactName>
<CountryCode>US</CountryCode>
<Phone>860-242-8050</Phone>
</Customer>
</Orders>
Code 6. Order message United States
The ability to convert from one message format to another is a fundamental principle
in enterprise integration. A framework can accept incoming XML data and perform
specialized XML transformations using XSLT and XPATH. Usually the least nonperforming aspect of any XML-based SOA is related to transformation. The ability
to rst externalize this from services to a framework will allow the enterprise to take
advantage of newer parsers without disturbing existing service code bases.
XML transformations can also be ofoaded to specialized XML appliances that
implement XML transformation within hardware. Hardware-based solutions have a
performance gain of factors of 20 to 1 or even greater. XML encryption and digital
signatures can also be ofoaded to specialized cards in order to provide additional
performance benets. Ideally, the framework will allow for transformation engines to
be congured using approaches such as Javas API for XML Processing (JAXP).
Datapower, Sarvega, Reactivity and Tarari are the leading vendors in this space.
265
8.4. Provisioning
Services are expressed in business terms and so should service levels. Traditional
service level agreements were based on technical constructs and had no tie to business
benet. A framework should provide a comprehensive way of translating a service
level agreement into the allocation of resourced required by consumers of the service
and may be comprised of required security proles, policies surrounding prioritization
of trafc and the number of requested that be executed during a given time period.
Service level agreements are abstract concepts that provide a mechanism for partners
to understand each others capabilities, to negotiate service parameters and assist
with providing a management goal. Service level agreements ideally provide a level
of separation between partners in managing concerns between them. A well-dened
service level agreement is given in Table 7.
Table 7. Components of a Service Level Agreement
Constraint Type Elements
Date
Start Date
End Date
Next Evaluation Date
(optional)
Function
Name
Day Time
Days of week
Hours
Process
Name (optional)
Construct
Description
Measured Item
Name
Location
Description
Species the date the service level agreement starts
and ends (i.e. 16 February 1977 to 10 September
2001) and the date in which the agreement can
be re-evaluated.
Could describe systems, services or other technical

constructs.
Should be expressed in terms of days of week (i.e.
Monday thru Friday, working days, holidays, etc.)
and hours (8am to 4pm) and include time zone
(GMT) and observation of daylight savings time.
Processes can describe sub-aspects of functional
systems and/or services.
A construct describes how the process operates
and/or message exchange patterns.
One or more items to be measured for system
qualities and determines where measurements will
occur (producer, consumer).
A service-level agreement attempts to completely dene mutual terms in a measurable

manner. For example, an agreement may contain a phrase like Starting March 1st,
2004 to March 1st, 2005, from 8am to 4pm eastern standard time, Monday thru
Friday except holidays, the average response time for the ten longest running quote
266
transactions as measured from the gateway should be less than ve seconds. In this
example, we have specied constraints on date, day/time, a group of processes and
where the agreement will be measured.
Figure 6. Provisioning Framework

A key to gaining control over service-level agreements is to not simply limit
measurement to compliance points contained within agreements but to fundamentally
build a comprehensive understanding of run-time information. The ability to
measure between producer and consumers, and even intermediaries becomes vital.
For example, the orders service from company ABC promises to deliver order items
to company 123 within three business days but uses the shipping service of company
XYZ to deliver orders. In this scenario several measurement challenges emerge.
First, company ABC can only measure its service-level by consulting either 123 or
XYZ. Second, 123 and XYZ will measure service-levels locally but will distribute
measurement information to external frameworks.
A service level agreement may be specied to a service consumer in general terms
or may be guaranteed at the time access to services is provisioned. Minimally,
provisioning engines need to support the following interface types:
Creation,
Activation,
Modication,
Suspension,
Deactivation,
Deletion, and
Interrogation.
267
Provisioning engines attempt to unify and centralize business rules around provisioning requirements. This may be implemented using a gateway approach in a distributed
manner that may synchronously or asynchronously update the provisioning system of
record. The gateway can be considered a provisioning service point or can optionally
manage a separate provisioning service.
8.5. Quality Assurance

Quality assurance is the discipline used in a systematic process of checking to
see whether a service being developed or deployed meets specied requirements.
Within most enterprises, there will exist a separate departments dedicated to quality
assurance practices. A quality assurance system is said to increase customer condence
and organizational credibility. Quality assurance systems emphasize catching defects
before they get into production.
A sound quality assurance practice will minimally test prior to production the
following elements of a service:
The service must work functionally for a single user (i.e. functionality).
The service must provide a response within a reasonable amount of time, scaling
or performing well in relation to the required number of simultaneous users.
This aspect assurance conformance to service level agreements (i.e. performance,
scalability).
The service must not crash in response to the anticipated number of simultaneous
users (i.e. reliability).
The service must never return invalid responses for any number of simultaneous
users even when under duress.
Testing services under load increases in importance especially in situations where
services will be accessed from service consumers not under the same administrative
control. Stress testing services traditionally is the method least tested but the most
important attribute especially if the service is accessed outside of the enterprise
and contains transactional, medical, nancial, or legal data. The business and legal
repercussions of providing inaccurate information to customers and partners can be
severe.
Services are usually implemented as part of a larger enterprise application. Traditional
quality assurance practices change when a service model is introduced. In the past,
practices such as regression testing entire applications were encouraged. This was
practical in that regression really tested systems under the same administrative control.
268
Since services can reside within the enterprise or external, it becomes important that
testing instead concentrate on services within administrative control due to issues of
causing unnecessary load from the external partys perspective.
SOA starts with the notion of a contract that is published by the service provider in
which consumers will leverage. Quality assurance in an SOA needs to concentrate
on making sure that the contract does not change between releases, that interfaces
over the lifetime of a service remain immutable and that developers have employed
appropriate defensive coding techniques within the application to handle invalid data
being passed to the service. Of even greater importance is guring if services adhere
to applicable industry standard schemas (where appropriate) and following enterprise
and industry vertical semantic constructs.
Management frameworks assist in quality assurance initiatives in several important
ways. First, they provide the ability to capture logging and diagnostic information
that will assist in quality assurance. The ability to quantify the amount of successful
requests vs. failed requests is a simplistic feature of a framework. Second, the
ability to have a strategy that can intercept specied requests to services external
to the enterprise or in situations where the service has usage costs and/or updates
transactional production systems becomes important. For example, if you are testing
your internal general ledger and it leverages a funds transfer service to your local
bank, it becomes viable to perform regression tests on all internal services with the
ability to simply return a predened message on behalf of the funds transfer service
without actually invoking it.
8.6. Business Processes

The discipline of business process management seeks to coordinate the end-toend activities (manual and automated) that provides value to end customers.
Enterprises have always sought to optimize their unique business processes that
provide competitive advantage and product differentiation. In order to realize this
goal, a strategy that covers the complete process lifecycle including deployment,
discovery, optimization and analysis is required.
Business people quickly understand business processes when expressed in business
terms, however the level of understanding is not realized within the enterprise due to
IT practices of digitizing business processes using their own specialized notations
such as reference architectures. Ideally, service operations that are related to business
processes should be separated from management-oriented operations and ideally
contained in their own interface denitions in order to break the coupling between
business process and technical implementation.
269
Business processes minimally include constructs that support the following notions:
Message Prioritization.
Business Activity Monitoring.
Let us discuss what these areas mean and how we can incorporate them into a services
management strategy.
8.7. Message Prioritization

The need to ensure that your most important customers receive better service than
other customers is vital to a services model. The ability to provide service request
prioritization based on requirements such as who is requesting the service and the
context of the request as it relates to business processes, contracts terms or other
business driven criteria may be necessary.
Prioritization may use implemented using queuing strategies for asynchronous service
requests, buffering or advanced forms of routing for synchronous requests. The ability
to prioritize can be based upon a supplied user credential contained within the message
where priority is this determined based on a lookup against the provisioning service,
an indicator within a header that either describes its context or indicates its priority or
based upon constructs supported at the transport level when using message queues.
8.8. Business Activity Monitoring

Business activity monitoring is a crosscutting concern and leverages the technical
aspects of monitoring and alerting. From a business perspective, the ability to
create alerts based on business processes becomes crucial in order to comply with
governmental regulations, internal audit procedures and enterprise security concerns.
Minimally, you should ask yourself do you have a mechanism to determine if your
most important customers are using your services, how are they using your services
and how often?
The idea behind business activity monitoring is simple. In the business world there
is a continuous production of information related to sales, production, logistics,
nancial operations and other business concerns that many times goes unnoticed
until it is too late. Traditional IT systems usually do not have capability to analyze
information in real time, creating alerts when problematic or anomalous situations
270
arise. Business activity monitoring can provide both real-time and historical services
activity monitoring and visualization and trafc information.
There are three high-level goals of business activity monitoring:
1. Ability to create a dashboard for executives who need aggregated information on
business activity at anytime, anywhere.
2. Monitor of business processes and activities in real-time to anticipate and manage
problems before they impact the bottom line.
3. Streamline operations with real-time access to the information line managers
need when they need it.
Business activity monitoring leverages the management frameworks ability to
introspect service request trafc and uses not only for technical considerations
but business focused events. It also incorporates disciplines found elsewhere within
the enterprise such as business intelligence built into data warehouses.
Figure 7. Business Activity Monitoring

If you wanted to analyze historical trends to determine peak usage levels for services,
then the data warehouse may t the enterprise requirement. But if say the enterprise
wanted to be notied when say order inquiry requests exceed a certain volume, the
271
ability to drill-down and see if it has a common root cause, the requirement to see
information in real-time becomes crucial. Using a service management approach not
only provides you with information but also context.
9. Management Architecture
A well-dened management strategy for services will lead to the conclusion that
execution can be distributed but management should be centralized. Management
can be accomplished using two basic methods, gateways and agents. Gateways are
a specialized proxy that intercept requests, enforce policies and forward requests
to downstream services. Agents are deployed as intercepting lters into services
containers such as Axis, Systinets WASP or in-house frameworks and enforce policies
in the address space of the service.
Let us look at the architecture of the following components:
Gateways,
Agents, and
Centralized Policies.
9.1. Gateways
In many enterprises, the infrastructure includes use of a proxy server that acts as
an intermediary between an internal computer resource and the Internet so that
the enterprise can ensure security, administrative control and caching to reduce
bandwidth usage. Proxy servers serve to separate the enterprise network from the
outside network.
Traditional proxy servers operate on the packet level. A network router functions by
examining TCP/IP packet headers and performing network routing decisions based
on rules that specify observed patterns in the packet ow. Firewalls block packet
ow by disallowing trafc that does not comply with specied policies and allowable
parameters.
There is also what is known as a reverse proxy server where users external to the
enterprise may make a request for an Internet service such as web page that resides on
an internal web server. If the request passes ltering requirements, the proxy server
272
will return the page to the user. If the proxy server implements caching, it will serve
the request out of is local store. If the request cannot be served out of its local store
will uses one of its own IP addresses to request the page from the server and forward
the response to the user on the Internet. To the Internet user, the proxy server is
invisible; all requests and responses appear to be answered directly by the proxy.
Use of TCP/IP ports and HTTP header information is insufcient for making
routing decisions based on XML-based content. Proxies that understand XML are
capable of examining trafc at the content level can serve as a mediator for both client
and server operations and perform operations itself or simply allow XML trafc to
ow through. When an intermediary can handle both situations it is referred to as a
gateway.
In services architecture, a gateway can be thought of as a specialized proxy that
understands request for services. A gateway is also analogous to a router in that it
knows where to direct a given packet of data that arrives at the gateway and a switch
that furnishes the actual path in and out of the gateway for a given packet.
Gateways when used with services will act as an intermediary for service request
and responses and enforce policies for each connected service. Gateways will require
that each service that will pass through it be registered and will create a new URL
for intercepting the request that can be published to other parties. The URL of the
service on the gateway will be the one published in a UDDI registry (Figure 8).
Figure 8. Gateway Service Registration
9.2. Agents
Agents run in context of the services address space and usually are implemented as
an intercepting lter. As agents are congured to run in the services container, they
can intercept all incoming requests and do not require generation of a proxy URL in
the same manner that gateways require. This approach also allows for services to be
globally managed vs having per-service registration requirements.
273
Figure 9. Agent Service Registration

Both gateways and agents internally can implement the pipes and lter architectural
pattern so that the same component lters can be deployed using both models.
For more about the intercepting lter pattern, see: http://java.sun.com/blueprints/
corej2eepatterns/Patterns/InterceptingFilter.html
9.3. Centralized Policies

Many IT budgets are consumed by maintenance of currently deployed services to
support new application-level policies that do not modify core business logic. The
need to externalize policies goes way beyond development time considerations and
has an effect on quality assurance efforts even for simple upgrades. IT security
departments need to govern security related standards used throughout the enterprise
and periodically need to verify compliance to the policy. Having policies embedded
within enterprise applications makes these tasks difcult.
The elements of a policy are comprised of:
Operational Rules,
Components, and
Persistent Storage.
Policies are a central concept to managing an SOA. Let us review the elements of a
policy.
9.4. Operational Rules

Policies may be dened that trigger actions from sending alerts to operations staff or
storing diagnostic information in a database for future analysis. In these situations,
policies are dened as conditional and will be invoked based upon specic conditions
occurring. Operational rules are composed of a conditional expression that is Boolean
and an action. For example, a rule can be dened to log all quotes from users who reside
274
in the state of Connecticut. Rules can be classied broadly in two categories (see
Table 8).
Table 8. Operational Rules
Rule Type
Description
General
Inspect all service invocations and take action each time rule
conditions are satised.
Aggregation
Rules that are applied based on consolidated metrics and takes

action based on aggregated view of conditions.
General rules when tied to processing instructions such as alerting allows contextual
information about specic events to be preserved but also can cause a ood in
scenarios where service failures occur in high throughput systems. General rules
should be applied in controlled situations as processing of the rules themselves could
have an adverse impact on the entire service network.
Aggregated rules provide a benet in that they are applied to summarized information
and therefore will generate less actions. This can be useful in scenarios where highthroughput services experience a failure but also have its own limitations in that
specic information about particular events may not be captured (for rule criteria,
see Table 9).
Table 9. Rule Criteria
Element
Name
Type
Description
Name of the rule.
Specify whether this rule is general or an aggregate.
Description
Action
Description of the rule.

The action(s) that should be performed if this rule is red. This
can range from creating an alert to logging.
Conditions in which this rule is red. Ideally this is best captured
using an Xpath expression.
Condition
Conditional rules may contain one or more elds within an XML message and
correspond to a specied value. Rule conditions are usually best expressed when
composed of two operands and a comparator. Optional support for XQuery
expressions should be supported in management frameworks.
For more information on XQuery, see McGovern et al. (2003).
275
9.5. Components
Policies are ultimately executed on either/both gateways and agents. Many of the
management concerns such as logging, metering and monitoring can be implemented
as both gateways and agents. However some management concerns may be limited
to one or the other. Let us look at concerns that have limited implementation in
Table 10.
Table 10. Management Attributes Matrix
Attribute
Routing
Category
Systems Management
Limitation
Gateway
Non-Repudiation
Message Prioritization
Security
Business Process
Agent
Gateway
Routing determines which service endpoint a particular request gets routed to. Since
agents run in context of the target service, the ability to change the target is already
too late. If you implement your own SOA framework and do not rely on SOAP or
other message formats, then it is possible to develop a custom message similar to the
HTTP protocol and its support for redirect headers (http status 302).
Non-repudiation is the ability to ensure that a party to a conversation (request/response) cannot deny the authenticity of their signature on a document or
the sending of a message that they originated. Since gateways serve as intermediaries
to service requests and have the ability to modify requests, they cannot be used to
provide end-to-end non-repudiation. The gateway can participate in non-repudiation
but ultimately the end service requiring non-repudiation related policies are required
to run as an agent in the address space of the service.
How to implement non-repudiation is covered in detail in Chapter 4.
Message prioritization can be implemented as either an advanced form of routing
and/or a buffering strategy. Buffering of trafc within the service in order to gain
prioritization would require adding cross-request scheduling internal to each service.
Although it is technically possible to perform this task by making worker threads go
to sleep for services that are multi-threaded, this would make service development
more complex than required and therefore is discouraged.
276
9.6. Persistent Storage

Management frameworks that leverage policy-based approaches should ideally make
sure that policies physically reside in multiple locations in order to ensure highavailability and disaster recovery. The framework themselves should treat policies
as dynamic entities that change during run-time and therefore must rely on either
synchronization/replication models.
Replication relies on background processes that copies policy data from one policy
store to another. When the background process completes replicating data from the
primary policy store to the secondary policy stores, it should synchronize its data with
the other policy stores in order to maintain consistency of policy implementation
across all gateways and agents.
Synchronization is the process of updating replicated data (policies) from one policy
store to another and ensures that all policies are identical across the enterprise. When
a new gateway or agent is added to the infrastructure, it is automatically synchronized
with all other peers. Several models for implementing synchronization exist, including
but not limited to the ones given in Table 11.
Table 11. Synchronization Models
Model
Change-Based
Cluster
Interval-Based
Member
On-Demand
Description
Receives notication when synchronized policy data changes. The
changed content is updated automatically and immediately on all
members in the synchronization loop.
This serves to synchronize all member policy services under the
same administrative federation.
Synchronization occurs based on predened interval or schedule.
This serves to synchronize a single member policy service with a
master policy service.
This form of synchronization only occurs when an administrator
specically requests it.
Synchronization can be accomplished in either a coarse-grained manner or more

granular depending upon the actual run-time services infrastructure approach.
Considerations for providing synchronization mechanisms to synchronize content
across management boundaries in some situations may be warranted.
277
10. Policy Architecture

Policies can be dened with the appropriate constructs to support business goals and
service level agreements within and across administrative domains and should map
to resources being managed (i.e. services). The ability to map higher-level policies
to low level resources requires a canonical representation to their meaning. Business
policies need to be translated into canonical form so that lower level components can
understand their intent.
In order for a framework to support policies, it must fulll the following requirements:
A canonical representation for expressing policies (this is best accomplished
using XML).
An interface that policy consumers (gateways and agents) can use to fetch
policies for services under its control.
Interfaces that support the creation, distribution, transformation, enforcement,
security and conict resolution of policies.
Interfaces that allow for storage of policies in a secure store.
Depending on how policies are distributed throughout the environment (push vs.
pull) service consumers may interact with policy managers in a publish/subscribe
model whereby they are notied when centralized policies change. Policy enforcement
points will interpret policies and perform run-time conguration changes on services
under their control.
Figure 10. Policy Architecture
278
In order to ensure integrity of the policies themselves, the policy management

architecture needs to incorporate minimally validation and conict resolution
interfaces. This will allow automated consistency checking. Interfaces are required to
translate policies to and from the canonical form so that service consumers that use
their own policy formats can plug into the service. Translation policies is applicable
when you want to have consistent management of services yet have a separate security
architecture provided by another vendor with its own set of policies (i.e. Netegrity,
Oblix, etc.).
Run-time resolution of policy conicts which may require specic knowledge about
services in order to determine costs for violating a service-level agreement or legal
consideration that can measure impact and decide accordingly may also be required.
10.1. Policy Execution

Events can trigger a sequence of processing steps from logging to sending email. One
architectural pattern that can assist in dividing the overall processing tasks tied to a
particular event into a sequence of smaller independent processing steps (lters) is
the Pipes and Filters pattern.
A common usage may be when Agile Healthcare receives a patients medical history
from a hospital and wants to ensure that the message is encrypted as well as ensure
that the message was not tampered with while in transit. It also wants to log receipt
of the message and update the medical history database with the latest information.
In Figure 11 an example pipeline is given to understand how this would work.
Figure 11. Policy Pipeline Using Pipes and Filter as Architectural Style
Filters are implemented with a generic interface that allows for reuse. The pipeline
will have several lters registered to it that will when receiving messages on the
inbound pipe, the lter will process the message and return the results back to the
outbound pipeline in a manner that allows them to be chained. Using this pattern for
components will allow you to add new lters, remove ones that are no longer needed
279
or even rearrange them into new sequences without having to touch the lter code
itself.
For more information on pipes and lters, see Buschmann et al. (1996).
Generic components can be developed and reused on both agents and gateways
when using the pipes and lters pattern. Listed in Table 12 are several recommended
components.
Table 12. Policy Pipeline Components
Name
Category
Authentication Security
Description
Validate that the service request is coming from an authorized
user.
Check to see has been granted rights to the specied service.
Authorization
Security
Certicate
Security
If the service requires client-side SSL then validate the

certicate and check against the certicate revocation list
(CRL).
SAML
Log
Security
Logging
Generate and/or validate creation of a security assertion.

Write a record to the event log based upon specied criteria.
Transform
Validate
Transformation Transform request/response using specied stylesheets.

Transformation Compare incoming XML message to specied XML schema.
11. Framework Vendors

There are several vendors that provide frameworks for managing service-oriented
architectures. Table 13 shows some of the ones the author team recommend for
enterprise consideration.
Table 13. SOA Management Frameworks
Vendor
Oblix
Product
CoreSV
Web Site
http://www.oblix.com/conuent/index.html
Blue Titan
Actional
Network Director
Looking Glass
http://www.bluetitan.com
http://www.actional.com
Digital Evolution
Infravio
Management Server
Ensemble
http://www.digev.com
http://www.infravio.com
280
12. Summary
Software as a service is important to enterprises not only to build internal systems but
also to extend the enterprise to partners, suppliers and consumers. As services allow
applications to become modular, and access to domains outside of administrative
control, traditional issues that require operational attention become even more
challenging. Services in this model must have a comprehensive management strategy
required for protable operation of the business. Management of services includes the
entire service value chain and collecting information about the health and availability
of services. All of these requirements can be accomplished in a vendor-neutral manner.
7
TRANSACTIONS
Life is an error-making and error-correcting process.
Jonas Salk, developer of the polio vaccine
As the concept of SOA has evolved as a means to integrate processes and applications at
an inter-enterprise level, traditional transaction semantics and protocols have proven
to be inappropriate. SOA-based transactions differ from traditional transactions in
that they execute over long periods, they require commitments to the transaction to
be negotiated at run-time, and isolation levels have to be relaxed. These kinds of
business-to-business transactions often require an extended transaction model that
builds on existing SOA standards and denes an interoperable transaction protocol
and message ows that help negotiate transactions guarantees at the inter-enterprise
level.
In this chapter we will look at the area of transactions as they apply to SOA and Web
Services in particular. We will show how although this is still an active area of work, it
is an extremely important one for the overall SOA environment; without transaction
capabilities, it is impossible to build complex composite applications that people can
trust to ensure consistent state changes, even in the presence of failures. However,
before examining the kinds of SOA transaction protocols that have been developed,
we need to rst examine what are often referred to as traditional ACID transaction
systems.
1. What Are ACID Transactions?

The concept of atomic transactions has played a cornerstone role in creating todays
enterprise application environments by providing guaranteed consistent outcome in
281
282
complex multiparty business operations and a separation of concerns in applications

yielding well-designed business process implementations (Gray and Reuter, 1993;
Bernstein and Newcomer, 1997).
So just what is an atomic transaction (often abbreviated to just transaction)? Put
simply, a transaction provides an all-or-nothing (atomic) property to work that is
conducted within its scope, whilst at the same time ensuring that shared resources are
isolated from concurrent users. Importantly application programmers typically only
have to start and end a transaction; all of the complex work necessary to provide the
transactions properties is hidden by the transaction system, leaving the programmer
free to concentrate on the more functional aspects of the application at hand.
Let us take a look at just how a transaction system could help in a real-world
application environment. Consider the case of an on-line cinema reservation system
(shown in Figure 1). In this gure, the cinema has many seats that can be reserved
individually, and the state of a seat is either RESERVED or UNRESERVED. The
cinema service exports two operations, reserveSeat and unreserveSeat (we will
ignore the other operations that are obviously required to make this service truly
usable). Finally we will assume that there is a transaction manager service that will be
used to manage any transactions that the cinema may require in order to process the
users requests.
Figure 1. The On-Line Cinema Booking System
Chapter 7: Transactions
283
Let us consider a very simple example: imagine that Mr. Doe wants to reserve a block
of seats for his family (1A, 1B and 1C as shown in the gure). Now, the service only
allows a single seat to be reserved through the reserveSeat operation, so this will
require Mr. Doe to call it three times, once for each seat.
Unfortunately the reservation process may be affected by failures of software or
hardware that could affect the overall consistency of the system in a number of ways.
For example, if a failure occurs after reserving 1A, then obviously none of the other
seats will have been reserved. Mr. Doe can try to complete the reservation when
(assuming) the cinema service eventually recovers, but by this time someone else may
have reserved the seats.
What Mr. Doe really wants is the ability to reserve multiple seats as an atomic
(indivisible) block. This means that despite failures and concurrent access, either all
of the seats Mr. Doe requires will be reserved for him, or none will. At rst glance
this may seem like a fairly straightforward thing to achieve, but it actually requires a
lot of effort to ensure that these requirements can be guaranteed. Fortunately atomic
transactions possess the following (ACID) properties that make them suitable for this
kind of scenario:
Atomicity: The transaction completes successfully (commits) or if it fails (aborts)
all of its effects are undone (rolled back).
Consistency: Transactions produce consistent results and preserve application
specic invariants.
Isolation: Intermediate states produced while a transaction is executing are not
visible to others. Furthermore, transactions appear to execute serially, even if
they are actually executed concurrently. Typically this is accomplished through
the use of concurrency control techniques (e.g., locks) associated with shared
resources.
Durability: The effects of a committed transaction are never lost (except by a
catastrophic failure).
A transaction can be terminated in two ways: committed or aborted (rolled back).
When a transaction is committed, all changes made within it are made durable
(forced on to stable storage, e.g., disk). When a transaction is aborted, all of the
changes are undone. Atomic transactions can also be nested, and in which case the
effects of a nested action are provisional upon the commit/abort of the outermost
(top-level) atomic transaction.
Associated with every transaction is a coordinator, which is responsible for governing
the outcome of the transaction. The coordinator may be implemented as a separate
284
service or may be co-located with the user for improved performance. It communicates
with enlisted participants to inform them of the desired termination requirements,
i.e., whether they should accept (commit) or reject (roll back) the work done within the
scope of the given transaction. For example, whether to purchase the (provisionally
reserved) ight tickets for the user or to release them. A transaction manager factory is
typically responsible for managing coordinators for many transactions. The initiator
of the transaction (e.g., the client) communicates with a transaction manager and
asks it to start a new transaction and associate a coordinator with the transaction.
Traditional transaction systems use a two-phase protocol to achieve atomicity between
participants, as illustrated in Figure 2: during the rst (preparation) phase, an
individual participant must make durable any state changes that occurred during
the scope of the transaction, such that these changes can either be rolled back
or committed later once the transaction outcome has been determined. Assuming
no failures occurred during the rst phase, in the second (commitment) phase
participants may overwrite the original state with the state made durable during
the rst phase.
Figure 2. Two-Phase Commit Protocol

In order to guarantee consensus, two-phase commit is necessarily a blocking protocol:
after returning the rst phase response, each participant who returned a commit
response must remain blocked until it has received the coordinators phase 2 message.
Until they receive this message, any resources used by the participant are unavailable
for use by other transactions, since to do so may result in non-ACID behavior. If the
coordinator fails before delivery of the second phase message these resources remain
blocked until it recovers.
As we have already mentioned, transactions are required to provide fault tolerance.
What this means is that information about running transactions (often referred to
as in-ight transactions) and the participants involved must survive failures and be
accessible during recovery. This information (the transaction log) is held in some
durable state-store. Typically the transaction log is scanned to determine whether
285
there are transactions mentioned in it that require recovery to be performed. If there

are, then the information within the log is used to recreate the transaction and the
recovery subsystem will then continue to complete the transaction.
Failures are not restricted to just the transaction coordinator. Therefore, participants
must retain sufcient information in durable store so that they too can be recovered
in the event of a failure. What information is recorded will obviously depend upon
the participant implementation.
1.1. The Synchronization Protocol

As well as the two-phase commit protocol, traditional transaction processing systems
employ an additional protocol, often referred to as the synchronization protocol. If
you recall the original ACID properties, then you will remember that Durability
is important in the case where state changes have to be available despite failures.
What this means is that applications interact with a persistence store of some kind
(e.g., a database) and this can impose a signicant overhead disk access is orders of
magnitude slower than access to main computer memory.
One apparently obvious solution to this problem would be to cache the state in main
memory and only operate on that for the duration of a transaction. Unfortunately,
you would then need some way of being able to ush the state back to the persistent
store before the transaction terminates, or risk losing the full ACID properties. This
is what the synchronization protocol does, with Synchronization participants.
Synchronizations are informed that a transaction is about to commit, so they can, for
example, ush cached state, which may be being used to improve performance of an
application, to a durable representation prior to the transaction committing. They are
then informed when the transaction has completed and in what state it completed.
Synchronizations essentially turn the two-phase commit protocol into a four-phase
protocol:
Before the transaction starts the two-phase commit, all registered Synchronizations are informed. Any failure at this point will cause the transaction to roll
back.
The coordinator then conducts the normal two-phase commit protocol.
Once the transaction has terminated, all registered Synchronizations are informed. However, this is a courtesy invocation because any failures at this stage
are ignored: the transaction has terminated so there is nothing to affect.
286
Unlike the two-phase commit protocol, the synchronization protocol does not have
the same failure requirements. For example, Synchronization participants do not need
to make sure they can recover in the event of failures; this is because any failure before
the two-phase commit protocol completes means the transaction will roll back, and
failures after it has completed cannot affect the data the Synchronization participants
were managing.
1.2. Optimizations to the Protocol

There are several variants to the standard two-phase commit protocol that are worth
knowing about because they can have an impact on performance and failure recovery.
We shall briey describe those that are the most common variants on the protocol
and found in all of the Web services transactions specications we will examine:
Presumed abort: if a transaction is going to roll back then it may simply record
this information locally and tell all enlisted participants. Failure to contact a
participant has no affect on the transaction outcome; the transaction is effectively
informing participants as a courtesy. Once all participants have been contacted
the information about the transaction can be removed. If a subsequent request
for the status of the transaction occurs there will be no information available
and the requestor can assume that the transaction has aborted (rolled back).
This optimization has the benet that no information about participants need
be made persistent until the transaction has decided to commit (i.e., progressed
to the end of the prepare phase), since any failure prior to this point will be
assumed to be an abort of the transaction.
One-phase: if there is only a single participant involved in the transaction, the
coordinator need not drive it through the prepare phase. Thus, the participant
will simply be told to commit and the coordinator need not record information
about the decision since the outcome of the transaction is solely down to the
participant.
Read-only: when a participant is asked to prepare, it can indicate to the
coordinator that no information or data that it controls has been modied
during the transaction. Such a participant does not need to be informed about
the outcome of the transaction since the fate of the participant has no affect
on the transaction. As such, a read-only participant can be omitted from the
second phase of the commit protocol.
287
1.3. Non-Atomic Transactions and

Heuristic Outcomes
We have already seen that in order to guarantee atomicity, the two-phase commit
protocol is necessarily blocking. What this means is that as a result of failures,
participants may remain blocked for an indenite period of time even if failure
recovery mechanisms exist. Some applications and participants simply cannot tolerate
this blocking.
To break this blocking nature, participants that have got past the prepare phase are
allowed to make autonomous decisions as to whether they commit or roll back: such
a participant must record this decision in case it is eventually contacted to complete
the original transaction. If the coordinator eventually informs the participant of the
transaction outcome and it is the same as the choice the participant made, then there
is no problem. However, if it is contrary, then a non-atomic outcome has obviously
happened: a heuristic outcome.
How this heuristic outcome is reported to the application and resolved is usually the
domain of complex, manually driven system administration tools, since in order to
attempt an automatic resolution requires semantic information about the nature of
participants involved in the transactions.
Precisely when a participant makes a heuristic decision is obviously implementation
dependent. Likewise, the choice the participant makes (to commit or to roll back)
will depend upon the implementation and possibly the application/environment in
which it nds itself. The possible heuristic outcomes are:
Heuristic rollback: the commit operation failed because some or all of the
participants unilaterally rolled back the transaction.
Heuristic commit: an attempted rollback operation failed because all of the
participants unilaterally committed. This may happen if, for example, the
coordinator was able to successfully prepare the transaction but then decided
to roll it back (e.g., it could not update its log) but in the meanwhile the
participants decided to commit.
Heuristic mixed: some updates (participants) were committed while others were
rolled back.
Heuristic hazard: the disposition of some of the updates is unknown. For those
which are known, they have either all been committed or all rolled back.
Heuristic decisions should be used with care and only in exceptional circumstances
since there is the possibility that the decision will differ from that determined by
288
the transaction service and will thus lead to a loss of integrity in the system. Having
to perform resolution of heuristics is something you should try to avoid, either
by working with services/participants that do not cause heuristics, or by using a
transaction service that provides assistance in the resolution process.
Now that we have described the advantages that ACID transactions offer, you may be
asking yourself why they are not sufcient for use in an SOA environment. An ACID
transaction provides failure atomicity, isolation from concurrent users etc. so would
appear to be an ideal tool for use when building complex distributed applications.
Unfortunately as we will see in the next section, this is often not the case.
2. Why ACID Is Too Strong for Web

Services
As we saw earlier, ACID transactions use a blocking two-phase commit protocol
to ensure consensus. If the transaction coordinator fails, participants may remain
blocked for extended periods of time while waiting for the coordinator to recover.
Furthermore, to ensure isolation semantics, resources acquired during an ACID
transaction may remain locked until the transaction completes; again in the event of
a failure this could be a long time. For example, imagine an online bookstore that had
to operate using ACID transactions; whenever you placed a book into your shopping
cart, the store would essentially have to reserve that book exclusively for you until
you decided whether to buy or not. That could take days or even weeks, making this
an impractical solution.
Therefore, structuring certain activities from long-running transactions can reduce
the amount of concurrency within an application or (in the event of failures) require
work to be performed again. For example, there are certain types of application where
it is known that resources acquired within a transaction can be released early, rather
than having to wait until the transaction terminates; in the event of the transaction
rolling back, however, certain compensation activities may be necessary to restore the
system to a consistent state.
Long-running activities can be structured as many independent, short-duration
transactions, to form a logical long-running transaction. This structuring allows
an activity to acquire and use resources for only the required duration of this
long-running activity. This is illustrated in Figure 3, where an application activity
(shown by the dotted ellipse) has been split into different, coordinated short-duration
transactions. Assume that the application activity is concerned with booking a taxi
289
(t1), reserving a table at a restaurant (t2), reserving a seat at the theatre (t3), and then
booking a room at a hotel (t4), and so on. If all of these operations were performed
as a single transaction then resources acquired during t1 would not be released until
the transaction has terminated. If subsequent activities t2, t3 etc. do not require those
resources, then they will be needlessly unavailable to other clients.
However, if failures and concurrent access occur during the lifetime of these
individual transactional activities then the behavior of the entire logical longrunning transaction may not possess ACID properties. Therefore, some form of
compensation may be required to attempt to return the state of the system to
consistency. For example, let us assume that t4 aborts. Further assume that the
application can continue to make forward progress, but in order to do so must now
undo some state changes made prior to the start of t4 (by t1, t2 or t3). Therefore, new
activities are started; tc1 which is a compensation activity that will attempt to undo
state changes performed, by say t2, and t3 which will continue the application once
tc1 has completed. t5 and t6 are new activities that continue after compensation,
e.g., since it was not possible to reserve the theatre, restaurant and hotel, it is decided
to book tickets at the cinema.
Figure 3. An Example of a Logical Long-Running Transaction, with

Failure
Previous transaction processing systems shared a great deal of commonality in terms
of the crux of the problem that they address and the abstractions they use to address
it. Specically, transaction processing systems were developed for particular platforms
and each system assumes that it is in sole control of the transaction domain and hence
does not generally have to interoperate with other transaction processing systems
(though interoperability with lower-level components like databases is generally well
supported via interfaces like X/Open XA, see The Open Group, 1991). Early attempts
at transaction interoperability (e.g., the Object Transaction Service from the Object
290
Management Group, see OTS) did not manage to get past the vendor lock-in
barrier, and attempts at using transactions across enterprise boundaries failed because
in such systems transactions are assumed to exhibit ACID properties.
Web services present a different kind of problem: they are specically about fostering
systems interoperability. This presents some interesting problems from a transaction
management point of view. What makes Web services so interesting is the fact that
the architecture is deliberately not prescriptive about what happens behind service
endpoints Web services are ultimately only concerned with the transfer of structured
data between parties, plus any meta-level information to safeguard such transfers (e.g.,
by encrypting or digitally signing messages) yet it is behind service endpoints that
we nd traditional transaction processing architectures supporting business activities.
Thus we are presented with a paradox. The Web services platform provides a
service-oriented, loosely coupled, and potentially asynchronous means of propagating
information between parties, whilst in the background we have traditional transaction
processing infrastructures whose behavior is neither or mutually interoperable.
Furthermore, the fact that transactions in these systems are assumed to exhibit ACID
properties potentially leads to problems when exposing resources to third parties, since
it presents opportunities to those parties to tie up resources and prevent transactions
from making progress. Thus if transactions were to be supported in the Web services
architecture then it is clear that some re-addressing of the problem is required.
As you might imagine from what we have said earlier, since transactions are an
important aspect of distributed systems in general and almost an imperative to ensure
SOAs can scale to enterprise applications, there has been quite a lot of activity in the
area of developing SOA transaction models. In the next section we will look at the
current leading specications for Web services transactions; it is important to realize
that these specications are applicable to other SOA environments.
3. A Brief History of Web Services

Transactions
So far there have been three efforts to incorporate transactions into Web services
and in the rest of this chapter we will examine the most important two. You may
wonder why you need to know about them both and the answer to that question is
straightforward: at this moment it is not possible to say which of these attempts (if
any) will become the standard for Web services transactions.
291
The rst attempt at dening a standard for Web Services transactions was the OASIS
Business Transaction Protocol (BTP) in 2001; this was then followed by the Web
Services Transactions specication (WS-Tx, now renamed WS-AtomicTransaction
(see WSAA) and WS-BusinessActivity (see WSBA)) from IBM, Microsoft and BEA
in August 2002, and more recently by the Web Services Transaction Management
specication (WS-Transaction Management) from Arjuna Technologies, Fujitsu,
IONA Technologies, Oracle and Sun in August 2003 (part of the OASIS Web
Services Composite Application Framework, see WSCAF).
Although originally having the backing of BEA, Hewlett-Packard and Oracle, the
OASIS BTP has been overtaken by the other two specications. There are a number
of technical reasons often cited for this, including the complexity of the protocol,
the fact that it was not designed solely for Web Services (often coupled with
its complexity) and lack of immediate interoperability with existing transaction
processing infrastructures. However, it is likely to be the lack of support for this
protocol from major vendors that will ultimately consign it to a niche area of history
(as seems to be the case in Web services, the political factors almost always outweigh
any technical benets). As such, in the rest of this chapter we will concentrate
solely on WS-AtomicTransaction and WS-BusinessActivity (we will refer to them
collectively as WS-Tx) and WS-TransactionManagement (WS-TXM).
In the following sections we will examine these specications. However, because of
space constraints we cannot cover them all in detail. Rather, we will look at the
commonality that exists between them and discuss the impact that the models they
provide will have on developing SOA applications.
4. The Coordination Frameworks

The rst area of commonality between WS-Tx and WS-TXM is in the area of
coordination. In general terms, coordination is the act of one entity (known as the
coordinator) disseminating information to a number of participants for some domainspecic reason. This reason could be in order to reach consensus on a decision like
in a distributed transaction protocol, or simply to guarantee that all participants
obtain a specic message, as occurs in a reliable multicast environment. When parties
are being coordinated, information known as the coordination context is propagated
to tie together operations which are logically part of the same coordinated work.
The context is critical to coordination since it contains the information necessary
for services to participate in the protocol. It provides the glue to bind all of the
applications constituent Web services together into a single coordinated application
whole.
292
Context information can ow implicitly (opaquely to the application) within normal

messages sent to the participants, or it may be an explicit action on behalf of
the client/service. This information is specic to the type of coordination being
performed, e.g., to identify the coordinator(s), the other participants in an activity,
recovery information in the event of a failure, etc. Furthermore, it may be required
that additional application specic context information (e.g., extra SOAP header
information) ow to these participants or the services which use them.
Coordination is an integral part of any distributed system, but there is no single
type of coordination protocol that can sufce for all problem domains. For example
a security coordination service will propagate differently formed contexts than a
transaction coordinator.
Despite the fact that there are many different types of distributed application that
require coordination, it should not come as a surprise to learn that each domain
typically uses a different coordination protocol. In the case of transactions, for
example, the OASIS BTP and the Object Management Groups OTS are solutions
to specic problem domains and which are not applicable to others since they
are based on different architectural styles. Given the domain-specic nature of
these protocols (i.e., loosely coupled transactional Web services and tightly coupled
transactional CORBA objects) there is no way of providing a universal solution
without jeopardizing efciency and scalability in each individual domain. However,
both of these protocols have the underlying requirement for propagating contextual
information to participants, and therefore it would make some sense if that mechanism
could be made generic, and thus re-used. On closer examination it is possible to
see that even solely within the Web services domain there are situations where
coordination is a requirement of several different types of problem domain, such as
workow management and transaction processing, but where the overall models are
very different.
Therefore, what is needed is a common Coordination Framework that allows users
and services to tie into it and customize it. A suitably designed coordination service
should provide enough exibility and extensibility to its users that allow it to be
tailored, statically or dynamically, to t any requirement.
As a result, both the WS-Tx and WS-TXM specications build upon their own
coordination frameworks: Web Services Coordination (WS-C) in the case of IBM,
Microsoft and BEA (see WSC), and Web Services Coordination Framework (WSCoordinationFramework) in the case of Oracle, Sun et al. As you might imagine,
there is a lot of commonality in the two coordination frameworks. In the following
293
sections we will look at the general architecture that both frameworks share and
indicate where differences arise.
Before we do so though, it is worth mentioning the interaction patterns that both sets
of specications assume. In order to support both synchronous request/response and
message interactions, all interactions are described in terms of correlated messages,
which an implementation may abstract at a higher level into request/response pairs.
As such, all communicated messages are required to contain response endpoint
addresses solely for the purposes of each interaction, and a correlation identier such
that incoming and outgoing invocations can be associated.
This has the immediate benet of allowing loose coupling of application entities:
the sender of a given message that requires a response need not be the same as
the ultimate receiver of that response. This allows for great exibility in choosing
service deployments, particularly in environments that may be error prone or require
dynamic changes to roles and responsibilities. However, one consequence of these
interactions is that faults and errors which may occur when a service is invoked are
communicated back to interested parties via messages which are themselves part of
the standard protocol and does not use the fault mechanisms of the underlying
SOAP-based transport.
4.1. Coordination Architecture

Both specications talks in terms of activities, which are distributed units of work,
involving one or more parties. At this level, an activity is minimally specied and is
simply created, made to run, and then completed. At the termination of an activity a
coordination protocol is executed by an associated coordinator. This is where the rst
main difference exists between WS-C and WS-CoordinationFramework: WS-C only
supports coordination at activity boundaries, whereas WS-CoordinationFramework
allows coordination to occur at arbitrary points during the lifetime of an activity; obviously this may be restricted by whether or not the specic coordinator
implementation supports it, but at least it is in the basic model.
In Figure 4, you can see that the framework could be used for propagating security,
workow, or replication contexts. Whatever coordination protocol is used, and in
whatever domain it is deployed, the same generic requirements are present:
Instantiation (or activation) of a new coordinator for the specic coordination
protocol, for a particular application instance;
Registration of participants with the coordinator, such that they will receive that
coordinators protocol messages during (some part of ) the applications lifetime;
294
Figure 4. The Architecture of a Coordination Framework

Propagation of contextual information between Web services that comprise the
application;
An entity to drive the coordination protocol through to completion.
The rst three of these points are directly the concern of the coordination frameworks
while the fourth is the responsibility of a third-party entity, usually the client
application that controls the application as a whole.
4.2. Creating a Coordinator

Both WS-C and WS-CoordinationFramework have the notion of what WS-C terms
an Activation Service and WS-CoordinationFramework calls an Activity Lifecycle
Service: this is the service that supports the creation of coordinators for specic
protocols and their associated contexts. In line with the interaction style mentioned
earlier, the process of invoking this service is done asynchronously, and so the
specications dene both the interface of the creation service itself and that of the
invoking service, so that the creation service can call back to deliver the results of the
coordinator creation (activation) namely a context that identies the protocol type
and coordinator location.
295
For simplicity we will only show the WS-C interfaces for its Activation Service in
Listing 1. The service has a one-way operation that expects to receive a CreateCoordinationContext message and the service that sent the CreateCoordinationContext
message expects to be called back with a CreateCoordinationContextResponse
message, or informed of a problem via an Error message.

<wsdl:portType name="ActivationCoordinatorPortType">
<wsdl:operation name="CreateCoordinationContext">
<wsdl:input
message="wscoor:CreateCoordinationContext"/>
</wsdl:operation>
</wsdl:portType>

<wsdl:portType name="ActivationRequesterPortType">
<wsdl:operation
name="CreateCoordinationContextResponse">
<wsdl:input
message="wscoor:CreateCoordinationContextResponse"/>
</wsdl:operation>
<wsdl:operation name="Error">
<wsdl:input message="wscoor:Error"/>
</wsdl:operation>
Listing 1. The WS-Coordination Activation Service interface
4.3. The Context

The context is critical to coordination since it contains the information necessary for
services to participate in the protocol. Since we are considering generic coordination
frameworks, contexts have to be tailored to meet the needs of specic coordination
protocols that are plugged into the framework. The format of contexts for WS-C and
WS-CoordinationFramework is specically designed to be third-party extensible and
its contents are as follows:
A coordination identier with guaranteed global uniqueness for an individual
coordinator in the form of a URI.
An address of a registration service endpoint where parties receiving a context
can register participants into the protocol.
A time-to-live value which indicates for how long the context should be
considered valid.
296
Extensible protocol-specic information particular to the actual coordination

protocol supported by the coordinator.
Since both frameworks are generic, the context is of very little use to some applications
without the ability to augment it. This is shown in Listing 2 where the schema states
that a context consists of a URI that uniquely identies the type of coordination
that is required (tns:ProtocolReferenceType), an endpoint where participants
to be coordinated can be registered (tns:CoordinatorReferenceType), and an
extensibility element designed to carry specic coordination protocol context payload
(xs:any), which can carry arbitrary XML payload.
<xs:complexType name="ContextType">
<xs:complexContent>
<xs:extension base="wsctx:ContextType">
<xs:sequence>
<xs:element name="protocol-reference"
type="tns:ProtocolReferenceType"/>
<xs:element name="coordinator-reference"
type="tns:CoordinatorReferenceType" maxOccurs="unbounded"/>
<xs:any namespace="##any" processContents="lax"
maxOccurs="unbounded"/>
</xs:sequence>
</xs:extension>
</xs:complexContent>
</xs:complexType>
Listing 2. The WS-CoordinationFramework context denition
Note that the reference to wsctx:ContextType above is because WSCoordinationFramework builds on a separate context service specication,
WS-Context; this specication denes a basic context type for use by a wide
range of services and not just coordination. WS-CoordinationFramework simply
extends this context denition.
4.4. Registering Participants

Both coordination frameworks make a distinction between an application Web service
and a coordination participant Web service. This distinction is typically made from
the perspective of the model, and implementations are free to blur the line. So, what
is meant by an application Web service? This is the service that encapsulates some
297
portion of the business logic of your application, e.g., an on-line airline reservation
service.
However, what about the participant service? Normally it is the case that only some
aspect if the work that the service does will require coordination, and in fact this may
be an aspect that the original service provider did not know about at the time the
service was designed and implemented.
For example, let us take the case of a transactional Web service. In this case, the
transactional capabilities of the service are often not exposed to users of that service,
as they are typically considered to be non-functional aspects of its work. It is
frequently the case that many transactional services or components began life as nontransactional, and were upgraded as requirements for fault-tolerance and reliability
changed. In the case of the on-line airline reservation service, for instance, the fact
that individual seats are reserved and unreserved atomically is not noticed by its users,
since that is an implementation detail (e.g., provided by the database technology that
the service implementer has used to store information about individual aircraft).
Because of this, it has been the case over many decades that transaction systems
usually separate out the work that is required to make a service transactional from
the usual business logic. This separation has been taken and generalized by the
coordination frameworks: the coordination participant is the service that controls
this work and is driven by a specic coordinator. This has obvious benets in that
a non-transactional service can be made transactional without having to change its
interface the interface is managed separately by the participant. If the application
Web service actually needs to participate in multiple different types of coordination
(e.g., security and transactions), then it would typically possess a participant for each
type.
Let us return to the coordination framework and see how this maps. Once a
coordinator has been instantiated and a corresponding context created, there is going
to be a need to be able to register participants with the coordinator. Rather than
assume that the service that created the context is also the one that handles participant
registration (that is an implementation choice, after all), both frameworks separate
this out into what WS-C calls a Registration Service and WS-CoordinationFramework
calls a Coordinator Service. Regardless of its name, this service allows participants to
register to receive protocol messages associated with a particular coordinator.
Like the activation service, the registration service assumes asynchronous communication and so species WSDL for both the service that registers and the service that
receives the result of registering. In Listing 3 we will only show the partial interfaces
for the ServiceCoordinator and the ServiceRespondant.
298
<wsdl:binding name="ServiceCoordinatorPortTypeSOAPBinding"
type="tns:ServiceCoordinatorPortType">
<soap:binding transport="http://schemas.xmlsoap.org/soap/http"
style="document"/>
<wsdl:operation name="addParticipant">
<soap:operation
soapAction="http://www.webservicestransactions.org
/wsdl/wscf/2003/03/addParticipant" style="document"/>
<wsdl:input>
<soap:header part="content" use="literal"
message="tns:ContextMessage"/>
</wsdl:input>
</wsdl:operation>
</wsdl:binding>
<wsdl:binding name="ServiceRespondantPortTypeSOAPBinding"
type="tns:ServiceRespondantPortType">
<soap:binding transport="http://schemas.xmlsoap.org/soap/http"
style="document"/>
<wsdl:operation name="participantAdded">
<soap:operation
soapAction="http://www.webservicestransactions.org
/wsdl/wscf/2003/03/participantAdded"
style="document"/>
<wsdl:input>
<soap:header part="content" use="literal"
message="tns:ContextMessage"/>
</wsdl:input>
</wsdl:operation>
</wsdl:binding>
Listing 3. Partial WSDL for the WS-CoordinationFramework registration service

Once a participant has been registered with a coordinator, it receives messages that
the coordinator sends (for example, prepare to complete and complete messages
if a two-phase protocol is used).
At this point it is interesting to note another difference between WSCoordinationFramework and WS-C: the WS-CoordinationFramework model allows
participants to asynchronously call back to the coordinator and register results before
the coordinator has even started the coordination protocol. This may seem strange
299
(and is obviously only possible if the coordinator implementation supports it), but
there are actually cases where this could help. Let us return to the case of a transactional
service, where its participant knows that even though the transaction may last for
many more hours, it is response to being told to prepare will never change; as a
result, the participant can tell the coordinator what that response is immediately (and
perhaps even be garbage collected). This can help to improve performance and in
some cases tolerance to failures.
4.5. Terminating the Coordinator

The role of terminator is generally played by the client application, which at an
appropriate point will ask the coordinator to perform its particular coordination
function with any registered participants to drive to protocol through to its
completion. On completion, the client application may be informed of an outcome
for the activity which may vary from simple succeeded/failed notication through to
complex structured data detailing the activitys status.
At this stage we have described the two coordination frameworks. By themselves
they are of extremely limited use. It is obviously when they are extended (or plugged
into) that their real power becomes evident. In the next section we will look at how
transaction models for Web services have been developed using these frameworks.
5. Web Services Transactions

Given that the traditional ACID transaction model is not appropriate for Web services,
let us pose the question, what type of model or protocol is appropriate? According to
the proponents of the two main Web services transactions specications, the answer
to that question is that that no one specic protocol is likely to be sufcient, given
the wide range of situations that Web service transactions are likely to be deployed
within. Fortunately, at present there seems to be a general consensus between the
different specications on the types of transaction model to support:
ACID transactions: Web services are for interoperability in closely coupled
environments such as corporate intranets as much as they are for the Web.
Interoperability between heterogeneous transaction service implementations is
a requirement and yet has been difcult to achieve in practice. This transaction
model is designed to support interoperability of existing transaction processing
systems via Web services, given such systems already form the backbone of
enterprise class applications. Although ACID transactions may not be suitable
300

for all Web services, they are most denitely suitable for some, and particularly
high-value interactions such as those involved in the nance sector. For example,
in a J2EE environment where the Java Transaction API is the standard way of
interacting with the transaction service (JTA), JTA-to-JTA interoperability is
supported through the JTS specication (JTS), but this is neither mandated nor
universally implemented.
Forward compensation-based transactions: this model is designed for those

business interactions that are long in duration, where traditional ACID
transactions are inappropriate. With this model, all work performed within
the scope of an application should be able to be compensated such that an
applications work is either performed successfully or undone. However, how
individual Web services perform their work and ensure it can be undone if
compensation is required, is an implementation choice. The model simply
denes the triggers for compensation actions and the conditions under which
those triggers are executed.
Both the specications from IBM et al. and Oracle et al. provide models that fall
into these categories. At the moment, the WS-TransactionManagement specication
provides an additional model that you will not nd elsewhere, so we will look at that
separately later in this chapter.
To give you an idea of how the transaction specications plug into and enhance their
respective coordination frameworks, take a look at Figure 5. As you can see, the basic
coordination framework components are activation and registration as we previously
described. However, each transaction model has an associated protocol that is made
available to an application by being plugged into the generic coordination framework.
Figure 5. An Example of the Transaction Protocol Dependency on a

Coordination Framework
What this means in practice is that the basic coordination context is augmented with
information that is specic to the transaction model. We will look at exactly what is
in this additional information in the following sections.
301
5.1. Atomic Transaction

Within this model it is assumed that all services (and associated participants) provide
ACID semantics and that any use of atomic transactions occurs in environments and
situations where this is appropriate: in a trusted domain, over short durations. This
model is supported by the WS-AtomicTransaction specication, and by the ACID
Transaction model in the WS-TransactionManagement specication.
As you would expect, in order to begin an atomic transaction, the client application
rstly locates a coordinator service that supports the right protocol. How that
location occurs is conveniently ignored by the various specications, so you
are in the hands of your favorite implementation. However, once located, the
client requests a new coordinator be created. As we saw earlier, in the WS-C
specication this would require the client to send a CreateCoordinationContext
message to the Activation Service. Importantly, the client must specify the type of
transaction required; for example, for WS-AtomicTransaction this is done by using
the http://schemas.xmlsoap.org/ws/2003/09/wsat URI in the creation message. The
activation service uses this information to make sure that it can support the desired
protocol (users often make mistakes!)
In both the WS-AtomicTransaction and WS-TransactionManagement specications,
the returned context has the type of transaction encoded within the context. As we
mentioned earlier, in order to allow participants to be enlisted with the transaction,
the context also contains a reference to the atomic transaction coordinator endpoint
(what we called the registration service earlier). We have illustrated this for the
WS-AtomicTransaction specication in Listing 4.

<CreateCoordinationContext>
<ActivationService>
<wsu:Address>
http://example.org/ws-transaction/activation
</wsu:Address>
</ActivationService>
<RequesterReference>
<wsu:Address>
http://example.org/ws-transaction/client-app
</wsu:Address>
</RequesterReference>
<CoordinationType>
http://schemas.xmlsoap.org/ws/2003/09/wsat
</CoordinationType>
302
</CreateCoordinationContext>

<wscoor:CoordinationContext
xmlns:wscoor="http://schemas.xmlsoap.org/ws/2002/08/wscoor"
xmlns:wsu="http://schemas.xmlsoap.org/ws/2002/07/utility">
<wsu:Identifier>
http://example.org/tx-id/aabb-1122-ddee-3344-ff00
</wsu:Identifier>
<wsu:Expires>2003-06-30T00:00:00-08:00</wsu:Expires>
<wscoor:CoordinationType>
http://schemas.xmlsoap.org/ws/2003/09/wsat
</wscoor:CoordinationType>
<wscoor:RegistrationService>
<wsu:Address>
http://example.org/ws-transaction/registration
</wsu:Address>
</wscoor:RegistrationService>
</wscoor:CoordinationContext>
Listing 4. Atomic transaction context
After obtaining a transaction context from the coordinator, the client application
then proceeds to interact with Web services to accomplish its business-level work.
Neither specication denes how a client can determine whether or not a service
is transactional and hence can deal with the associated context. Currently there is
no equivalent of the J2EE deployment descriptor for services, or the OMGs Object
Transaction Service for dening that a service is transactional and as such requires a
valid context to be carried on all client-service interactions. No doubt this oversight
will be removed when the likes of WS-Policy statements are extended to transactional
semantics. Until that happens, however, each invocation on a business service requires
the client to propagate the context, such that the each invocation is implicitly scoped
by the transaction.
5.1.1. Supported Protocols

Both WS-AtomicTransaction and ACID Transaction support the two-phase commit
protocol we described earlier. As you should expect by now, all interactions between
the coordinator and participants are dened in terms of one-way messages. This is
illustrates in Listing 9-5, where we show the WSDL for the two-phase commit participant (twoPCParticipantPortType) from the ACID Transaction protocol as well
as the callback WSDL for the coordinator (CoordinatorParticipantPortType).
303
<wsdl:portType name="twoPCParticipantPortType">
<wsdl:operation name="prepare">
<wsdl:input message="tns:PrepareMessage"/>
</wsdl:operation>
<wsdl:operation name="onePhaseCommit">
<wsdl:input message="tns:OnePhaseCommitMessage"/>
</wsdl:operation>
<wsdl:operation name="rollback">
<wsdl:input message="tns:RollbackMessage"/>
</wsdl:operation>
<wsdl:operation name="commit">
<wsdl:input message="tns:CommitMessage"/>
</wsdl:operation>
<wsdl:operation name="forgetHeuristic">
<wsdl:input message="tns:ForgetHeuristicMessage"/>
</wsdl:operation>
</wsdl:portType>
<wsdl:portType name="CoordinatorParticipantPortType">
<wsdl:operation name="committed">
<wsdl:input message="tns:CommittedMessage"/>
</wsdl:operation>
<wsdl:operation name="rolledBack">
<wsdl:input message="tns:RolledBackMessage"/>
</wsdl:operation>
<wsdl:operation name="vote">
<wsdl:input message="tns:VoteMessage"/>
</wsdl:operation>
<wsdl:operation name="heuristicForgotten">
<wsdl:input message="tns:HeuristicForgottenMessage"/>
</wsdl:operation>
<wsdl:operation name="heuristicFault">
<wsdl:input message="tns:HeuristicFaultMessage"/>
</wsdl:operation>
</wsdl:portType>
Listing 5. The two-phase commit participant and coordinator WSDL
Transaction termination normally uses the two-phase commit protocol. If a transaction involves only a single participant, both models support a one-phase commit
optimization similar to that in traditional transaction systems. Figure 6 shows the
state transitions of an atomic transaction.
304
Figure 6. Two-Phase Commit State Transitions

As well as the two-phase commit protocol, both WS-AtomicTransaction and ACID
Transaction support the synchronization protocol, though there are slight differences
in the name. In WS-AtomicTransaction it is called the Volatile 2PC protocol, whereas
in ACID Transaction it is given the more obvious name of the Synchronization
protocol. Once again, using the WS-TXM WSDL as an example, Listing 6 shows
the one-way interaction style common to both specications.
<wsdl:portType name="SynchronizationPortType">
<wsdl:operation name="beforeCompletion">
<wsdl:input message="tns:BeforeCompletionMessage"/>
</wsdl:operation>
<wsdl:operation name="afterCompletion">
<wsdl:input message="tns:AfterCompletionMessage"/>
</wsdl:operation>
</wsdl:portType>
<wsdl:portType name="CoordinatorParticipantPortType">
<wsdl:operation name="beforeCompletionParticipantRegistered">
<wsdl:input message=
"tns:BeforeCompletionParticipantRegisteredMessage"/>
</wsdl:operation>
<wsdl:operation name="afterCompletionParticipantRegistered">
<wsdl:input message=
"tns:AfterCompletionParticipantRegisteredMessage"/>
</wsdl:operation>
</wsdl:portType>
Listing 6. The Synchronization protocol WSDL
305
A subtle, but important difference between the WS-AtomicTransaction and ACID

Transaction models is that the former does not support heuristic outcomes. Although
this may well change as the specication develops, at present not being able to propagate heuristic results from participants to the coordinator, or from the coordinator
to the driving application, will have signicant impact on the manageability of any
application that uses WS-AtomicTransaction.
5.2. Business Activity

Most business-to-business applications require transactional support in order to
guarantee consistent outcome and correct execution. These applications often involve
long running computations, loosely coupled systems and components that do not
share data, location, or administration and it is difcult to incorporate atomic
transactions within such architectures. For example, an online bookshop may reserve
books for an individual for a specic period of time, but if the individual does not
purchase the books within that period they will be put back onto the shelf for
others to buy.
Let us term this type of transaction a business activity, in line with the WSBusinessActivity specication. The WS-TransactionManagement specication calls
this a Long Running Action. A business activity is designed specically for longduration interactions, where exclusively locking resources is impossible or impractical.
In this model services are assumed to encapsulate work that can be compensated later
(possibly by executing another business activity).
As you would expect, each business activity is managed by a coordinator and when
services do work, they enlist participants with the coordinator, such that if the activity
later cancels the activity (needs to undo the work), these participants are informed
and can then compensate for the work previously performed. What this means is
that in essence, each service has a compensation element that can be used to undo
the work later and this element is controlled by the participant (compensator).
While the full ACID semantics are not maintained by a business activity, consistency
can still be maintained through compensation, though the task of writing correct
compensating actions (and thus overall system consistency) is delegated to the
developers of the services.
Central to business activities is the notion of scopes and dening activity-to-task
relationships. A business activity may be partitioned into scopes, where a scope
is a business task or unit of work using a collection of Web services. The Long
Running Action model explicitly maps these scopes to nested activities, whereas the
306
WS-BusinessActivity model is less well dened. Such scopes can be nested to arbitrary
levels, forming parent and child relationships. A parent scope has the ability to select
which child tasks are to be included in the overall outcome protocol for a specic
business activity, and so clearly non-atomic outcomes are possible. A business activity
denes a consensus group that allows the relaxation of atomicity based on business
level decisions. If a child task experiences an error, it can be caught by the parent who
may be able to compensate and continue processing.
Nested scopes are important for a number of reasons, including:
Fault-isolation: if sub-scope fails (e.g., because a service it was using fails) then
this does not require the enclosing scope to fail, thus undoing all of the work
performed so far.
Modularity: if there is already a scope associated with a call when a new scope is
begun, then the scope will be nested within it. Therefore, a programmer who
knows that a service requires scopes can use them within the service: if the
services methods are invoked without a parent scope, then the services scopes
will simply be top-level; otherwise, they will be nested within the scope of the
client.
When a child task completes it can either leave the business activity or signal to the
parent that the work it has done can be compensated later. In the latter case, the
compensation task may be called by the parent should it ultimately need to undo the
work performed by the child.
Underpinning the business activity model are three fundamental assumptions:
1. All state transitions are reliably recorded, including application state and
coordination metadata (the record of sent and received messages);
2. All request messages are acknowledged, so that problems are detected as early as
possible. This avoids executing unnecessary tasks and can also detect a problem
earlier when rectifying it is simpler and less expensive; and
3. As with atomic transactions, a response is dened as a separate operation and
not as the output of the request. Message input-output implementations will
typically have timeouts that are too short for some business activity responses. If
the response is not received after a timeout, it is resent. This is repeated until a
response is received. The request receiver discards all but one identical request
received.
Most workow systems do not distinguish compensate activities from forward
progress activities: an activity is an activity and it just does some work. If that
work happens to compensate for some previous work then so be it. In addition,
307
most services you will nd already have compensate operations written into their
denitions, like cancel seat reservation or cancel holiday and they dont need to
be driven by some other transaction/coordination engine that then sends prepare
or commit or roll back to a participant which then has to determine how to talk
to the service to accomplish the same goal.
Although similar in their goals, there are fundamental protocol differences between
the WS-BusinessActivity and Long Running Action models. From an application
programmers perspective it is very unlikely that you will notice these differences.
However, we will now briey cover these differences.
5.2.1. WS-BusinessActivity
The
WS-BusinessActivity
model
denes
two
sub-protocols,
the
BusinessAgreementWithCoordinatorComplete and BusinessAgreementWithParticipantComplete. The only difference is that in the case of
BusinessAgreementWithCoordinatorComplete the child scope cannot
autonomously decide to end its participation in the business activity, even if it can
be compensated. Rather the child task relies upon the parent to inform it when the
child has received all requests for it to perform work which the parent does this by
sending the complete message to the child.
A child activity is initially created in the active state; if it nishes the work it was
created to do and no more participation is required within the scope of the business
agreement (such as when the activity operates on immutable data), then the child can
unilaterally send an exited message to the parent. However, if the child task nishes
and wishes to continue in the business activity then it must be able to compensate
for the work it has performed (e.g., un-reserve the seat on the ight). In this case
it sends a completed message to the parent and waits to receive the nal outcome of
the parent. This outcome will either be a close message, meaning the business activity
has completed successfully or a compensate message indicating that the parent activity
requires that the child task reverse its work.
5.2.2. Long Running Action

In this (LRA) model, each application is bound to the scope of a compensation
interaction. For example, when a user reserves a seat on a ight, the airline reservation
centre may take an optimistic approach and actually book the seat and debit the
users account, relying on the fact that most of their customers who reserve seats later
book them; the compensation action for this activity would obviously be to un-book
the seat and credit the users account. Work performed within the scope of a nested
308
LRA must remain compensatable until an enclosing service informs the individual
service(s) that it is no longer required.
Listing 7 shows the WSDL for the Compensator (CompensatorPortType) and the
coordinator that drives it (CoordinatorPortType).
<wsdl:portType name="CompensatorPortType">
<wsdl:operation name="compensate">
<wsdl:input message="tns:CompensateMessage"/>
</wsdl:operation>
<wsdl:operation name="complete">
<wsdl:input message="tns:CompleteMessage"/>
</wsdl:operation>
<wsdl:operation name="forget">
<wsdl:input message="tns:ForgetMessage"/>
</wsdl:operation>
</wsdl:portType>
<wsdl:portType name="CoordinatorPortType">
<wsdl:operation name="compensated">
<wsdl:input message="tns:CompensatedMessage"/>
</wsdl:operation>
<wsdl:operation name="completed">
<wsdl:input message="tns:CompletedMessage"/>
</wsdl:operation>
<wsdl:operation name="forgot">
<wsdl:input message="tns:ForgotMessage"/>
</wsdl:operation>
<wsdl:operation name="unknownCompensator">
<wsdl:input
message="tns:UnknownCompensatorFaultMessage"/>
</wsdl:operation>
<wsdl:operation name="cannotCompensate">
<wsdl:input
message="tns:CannotCompensateFaultMessage"/>
</wsdl:operation>
<wsdl:operation name="cannotComplete">
<wsdl:input message="tns:CannotCompleteFaultMessage"/>
</wsdl:operation>
</wsdl:portType>
Listing 7. The LRA Compensator WSDL
309
Let us consider the example of an online travel agent. The travel agent is concerned
with booking a taxi, reserving a table at a restaurant, reserving a seat at the theatre,
and then booking a room at a hotel. If all of these operations were performed as
a single transaction then resources acquired during booking the taxi (for example)
would not be released until the top-level transaction has terminated. If subsequent
activities do not require those resources, then they will be needlessly unavailable to
other clients.
Figure 7 shows how part of the night-out may be mapped into LRAs. All of the
individual activities are compensatable. For example, this means that if LRA1 fails or
the user decides to not accept the booked taxi, the work will be undone automatically.
Because LRA1 is nested within another LRA, once LRA1 completes successfully
any compensation mechanisms for its work may be passed to LRA5: this is an
implementation choice for the Compensator. In the event that LRA5 completes
successfully, no work is required to be compensated, otherwise all work performed
within the scope of LRA5 (LRA1 to LRA4) will be compensated.
Figure 7. LRA Example

A Compensator is the LRA participant that operates on behalf of a service to undo the
work it performs within the scope of an LRA. How compensation is carried out will
obviously be dependent upon the service; compensation work may be carried out by
other LRAs which themselves have Compensators.
When a service performs work that may have to be later compensated within the
scope of an LRA, it enlists a Compensator participant with the LRA coordinator.
The coordinator will send the Compensator one of the following messages when the
activity terminates:
310
Success: the activity has completed successfully. If the activity is nested then
Compensators may propagate themselves (or new Compensators) to the
enclosing LRA. Otherwise the Compensators are informed that the activity
has terminated and they can perform any necessary cleanups.
Fail: the activity has completed unsuccessfully. All Compensators that are
registered with the LRA will be invoked to perform compensation in the reverse
order. The coordinator forgets about all Compensators that indicated they
operated correctly. Otherwise, compensation may be attempted again (possibly
after a period of time) or alternatively a compensation violation has occurred
and must be logged.
Each service is required to log sufcient information in order to ensure (with best
effort) that compensation is possible.
So far we have not really considered the relationship between LRAs in an application.
Obviously LRAs may be used sequentially and concurrently, where the termination
of an LRA signals the start of some other unit of work within an application.
However, LRAs are units of compensatable work and an application may have as
many such units of work operating simultaneously as it needs to accomplish its tasks.
Furthermore, the outcome of work within LRAs may determine how other LRAs are
terminated.
An application can be structured to so that LRAs are used to assemble units of
compensatable work and then held in the active state while the application performs
other work in the scope of different (concurrent or sequential) LRAs. Only when
the right subset of work (LRAs) is arrived at by the application will that subset be
conrmed; all other LRAs will be told to cancel (complete in a failure state).
At the start of this chapter we mentioned that although the WS-Transaction and
WS-TransactionManagement specications share many things in common, the latter
specication possesses an additional protocol that the IBM, Microsoft and BEA
specications do not: the business process model. In the following section we will
briey talk about this model.
5.3. Business Process Model

The business process (BP) model is different from any of the other transaction
models we have looked at so far. This model is specically aimed at tying together
heterogeneous transaction domains into a single business-to-business transaction. So,
for example, with this model it is possible to have a long-running business transaction
311
span messaging, workow and traditional ACID transactions. The reason for this is
to allow business to leverage their existing corporate IT investment.
In the business process transaction model all parties involved in a business process
reside within business domains, which may themselves use business processes to
perform work. Business process transactions are responsible for managing interactions
between these domains. A business process (business-to-business interaction) is split
into business tasks and each task executes within a specic business domain. A business
domain may itself be subdivided into other business domains (business processes) in
a recursive manner.
Each domain may represent a different transaction model if such a federation of
models is more appropriate to the activity. Each business task (which may be modeled
as a scope) may provide implementation specic counter-effects in the event that
the enclosing scope must cancel. In addition, periodically the controlling application
may request that all business domains checkpoint their state such that they can either
be consistently rolled back to that checkpoint by the application or restarted from
the checkpoint in the event of a failure.
An individual task may require multiple services to work. Each task is assumed to be
a compensatable unit of work. However, as with the LRA model, how compensation
is provided is an implementation choice for the task.
For example, let us return to our travel agent and see how it might be mapped into the
BP model. If you look at Figure 8 you can see that the on-line travel agent interacts
with its specic suppliers, each of which resides in its own business domain. The work
necessary to obtain each component is modeled as a separate task, or Web service. In
this example, the Flight reservation task is actually composed of two sub-tasks; one
that gets the ight and the other that gets the necessary travel insurance.
Figure 8. Business Processes and Tasks for a Travel Agent
312
The user may interact synchronously with the travel agent to build up the details of
the holiday required. Alternatively, the user may submit an order (possibly with a
list of alternate requirements, such as destinations, dates, etc.) to the agent who will
eventually call back when it has been lled; likewise, the travel agent then submits
orders to each supplier, requiring them to call back when each component is available
(or is known to be unavailable).
The business process transaction model supports this synchronous and asynchronous
interaction pattern. Business domains are instructed to perform work within the
scope of a global business process. The business process has an overall manager that
may be informed by individual tasks when they have completed their work (either
successfully or unsuccessfully), or it may periodically communicate with each task to
determine its current status. In addition, each task may make period checkpoints of
its progress such that if a failure occurs, it may be restarted from that point rather
than having to start from the beginning. A business process can either terminate in a
conrmed (successful) manner in which case all of the work requested will have been
performed, or it will terminate in a cancelled (unsuccessful) manner, in which case
all of the work will be undone.
If it cannot be undone, then this fact must be logged. One key difference between
the business process transaction model and that of traditional 2PC is that it assumes
success: the BP model is optimistic and assumes the failure case is the minority and
can be handled or resolved ofine if necessary, or through replay/void/compensation,
but not always automatically, often requiring human interaction.
So far we have seen how useful transactions (ACID or not) can be when developing
distributed applications. However, throughout the previous descriptions there has
been an implicit assumption: a user can trust a transaction coordinator and a
transaction coordinator can trust the participants. In the sorts of closely coupled
environments where transactions originated these kinds of trust relationships are
possible. As you can see in the chapter or Security, in the loosely coupled SOA
environment, trust is something that cannot be taken as a given. In the next section
we will give a brief overview of the kinds of issues that could arise when trust is not
readily available.
6. Security Implications
There are several implicit security issues in the use of distributed transactions,
regardless of the model that is employed. As we have seen, in any of the transaction
models there are essentially three actors:
313
1. The participant that is driven through the termination protocol.

2. The coordinator, with whom participants register and who drives them through
the termination protocol.
3. The application logic that tells the coordinator when and how to terminate.
Whatever the application logic might want (e.g., to commit) is never guaranteed by
the coordinator, simply because failures can occur that force the transaction to roll
back. However, within the basic protocols, the coordinator will always try to do what
the application tells it and the coordinator cannot determine whether or not requests
to terminate come from valid sources. What this means is that if some illegal user
were to obtain a reference to the coordinator, it could force it to do work that is
different from that which the legal user (application) would want.
Now let us consider the case where one of the participants is deliberately going to try
to subvert the transaction protocol. By being involved in a transaction, a participant
has knowledge of its outcome and can also affect that outcome. Having knowledge
that a specic transaction has committed or rolled back, for example, could be
important (e.g., insider share trading). Likewise, being able to affect the outcome of
the transaction is obviously important too.
However, can the coordinator always be trusted? The coordinator always has the nal
say on the actual outcome of the transaction. Just because the application logic says to
roll back, does not mean that the coordinator actually has to roll back the participants
it could lie. Likewise, if the application logic says to commit and all participants say
they can commit, the coordinator could still roll them back. Even more threatening
(especially to application consistency) is the fact that the coordinator could tell some
participants to commit, and others to roll back!
What this brief overview of the security assumptions should show you is that just
because there are transaction protocols for Web services does not mean that they alone
can be used to solve all of your problems. Security and transactions have gone handin-hand for several years in traditional transaction processing systems, particularly in
loosely coupled environments where no single trust domain exists. Unfortunately at
this time there is no single standard for Web services security and neither of the Web
services transactions specications indicates how any kind of security infrastructure
should be tied in. At best they simply say security is required and reference a
specic security specication. For example, the WS-AtomicTransaction specication
is quite clear in the fact that it should be used in conjunction with the WS-Security
specication to ensure that transactions only ever span trusted domains. The chapter
on Security provides a good insight into what it means to make your Web Services
secure and you should read that chapter in conjunction with this section to understand
the implications of making a Web Services transaction protocol secure.
314
7. Interoperability Considerations
As you might expect, there are some interesting interoperability issues presented by
having multiple transactions specications. However, there are also interoperability
issues within the specications. So let us rst look at whether interoperability between
heterogeneous implementations of the same Web services transactions specication
is possible.
Unfortunately both WS-Tx and WS-TransactionManagement dene message exchange patterns for their various models that allow for vendor-specic enhancements.
For example, if we look again at the CoordinationContext structure dened by the
WS-Coordination specication and shown in Listing 2, we can see that there is an any
in the context. This is intended to allow for implementation specic enhancements
to the basic context. Individual implementations can add whatever information
they require in order to function, and this then extends to the WS-Transaction
specications that leverage the context.
Although it is possible to implement the various transaction protocols without
recourse to the extensibility elements of the contexts or messages, it is likely that some
vendors will use these to implement protocol optimizations. If these implementations
used the optimizations with some foresight for interoperability then they should be
able to continue to work in the absence of the extensibility elements (albeit in a less
optimized manner). However, we have yet to nd an implementation that does this,
i.e., those that use the extensibility, require it to be present in all messages in order
for the protocols to work. Obviously this signicantly affects interoperability!
So although interoperability by implementations of the same specications is possible,
it is not necessarily a requirement of all vendors. You should be careful when choosing
an implementation if interoperability is a requirement now or in then future.
If you thought interoperability within a specication was simple, then hopefully
it should not come as a surprise to learn that interoperability between the two
specications is more complex. Despite the fact that the specications share many
requirements and have models in common, the on-the-wire message formats and
WSDL are different. It is possible that as these specications evolve (neither of them
have yet to be ratied as a standard), they may move closer and perhaps even merge.
In the meanwhile, direct interoperability (where, for example, an ACID Transaction
two-phase participant can enroll with a WS-AtomicTransaction coordinator), is not
possible. However, there are some vendors (e.g., Arjuna Technologies, ATL) who are
working closely with both specications (and their authors) to provide products that
can opaquely to an application bridge between the different transaction protocols.
315
Although all of this might sound like we have a considerable way to go before
interoperability is a fact of life, it is worth remembering that there is another
type of interoperability that is at least as important, and which we can cater for
now: interoperability with (and between) existing legacy transaction systems. Both
WS-AtomicTransaction and ACID Transaction models are meant specically for
integration with back-end infrastructures easier.
Web Services are for interoperability as much as for the Internet. As such, interoperability of existing transaction processing systems will be an important part of
Web Services transactions: such systems already form the backbone of enterprise level
applications and will continue to do so for the Web services equivalent. Business-tobusiness activities will involve back-end transaction processing systems either directly
or indirectly and being able to tie together these environments will be the key to the
successful take-up of Web Services transactions. Leveraging existing infrastructural
investments in terms of transaction systems and services that use them are extremely
important, and both sets of specications are ideal for this purpose.
8. Summary
We have looked at the two main candidates vying for the title of Web services
transactions standard. All of the main Web services heavyweights are backing either
WS-Transaction or WS-TransactionManagement. Whether there a single standard
will ever evolve is unknown and is probably more in the realm of politics than
technology.
As we have seen, although ACID transactions are not for all Web Services because of
their blocking nature, they are still a requirement for some use cases, especially where
interoperability is important. Interoperability of heterogeneous ACID transaction
systems is now possible and has been an often unfullled requirement for many
enterprises for a long time. Although Web Services represent a new arena of
distributed system that is loosely coupled, potentially large scale and heterogeneous,
they are for interoperability as much as for the Internet. As such, interoperability of
existing transaction processing systems will be an important part of Web Services
transactions: such systems already form the backbone of enterprise level applications
and will continue to do so for the Web services equivalent.
Business-to-business activities will involve back-end transaction processing systems
either directly or indirectly and being able to tie together these environments will
be the key to the successful take-up of Web Services transactions. Web services will
operate as the glue between different environments (e.g., corporate domains) and
316
application components will be implemented using tried and tested technologies

such as J2EE, CORBA and .NET. Compensation transactions such as the WSTransactionManagement Long Running Action model or other extended transaction
models will be important in areas where ACID transactions are inappropriate.
Finally we looked at the various security and interoperability issues that are still
unresolved. Because this is still a relatively new area, it is perhaps unreasonable
to assume that all of the glitches have been solved. However, when deploying
implementations, you should be aware of the issues and be able to determine whether
they are important now or in the future. Just because these problems have not been
fully addressed by the specications does not mean that a vendor implementation
cannot (or should not) provide a good solution.
8
EVENT-DRIVEN
ARCHITECTURE
Let me tell you the secret that has led me to my goal:

My strength lies solely in my tenacity.
Louis Pasteur
Early in many of our careers, we would have remembered the days of batch
systems, transaction processing monitors such as CICS and other components of
the mainframe that were used to run many mission-critical systems. As time passed
and Moores law provided us with the ability to double computing capacity every 18
months, we began to develop systems that provided information in real-time. With
increasing computing capacity, we were able to ignore the architecture principles
learned by our forefathers and instead preferred to create systems that did processing
inline. This had the effect of making systems over time less scalable and more
susceptible to outages especially when downstream services the application depended
on were unavailable.
Event-driven architectures are one of the few instances in which the real-world
benets of an architectural approach were realized before the hype. The phrase
event-driven architecture refers to applications and services that have the ability
to react to changes in conditions, regardless of whether the change is a failure in a
downstream system or a sudden change in the marketplace such as a meltdown on
Wall Street.
317
318

An event is a change in the state of a resource or request for processing.
A time honored approach still holds true in service-oriented architecture of if it does

not need to be done in real-time, do it asynchronously. Asynchronous behavior within
applications is not just a technical construct but the way business itself operates.
Remembering one of our core principles, that services should model the business, we
can conclude that being event-driven is mandatory for the creation of an Enterprise
SOA.
If you have meticulously read this book, you may have noticed a subtle but crucial
difference between traditional service-oriented architectures and being event-driven.
To this point, we have achieved the goal of articulating how to gain loose coupling
amongst services but need to take it one step further. Event-driven architecture will
help us become decoupled to a much greater degree. Coupling of services occurs at
the level of the service contract (In Web Services, the WSDL document). With an
event-driven approach, there is no need for any such contract. The only connection
between event producers and consumers are the subscription and publication activities
between them.
Let the reader be warned, this chapter may require multiple passes before one fully
understands its content. In this chapter, we will cover:
Service Design,
Enterprise Service Bus,
Business Rules,
Pools,
Patterns,
Agents,
Callbacks,
Finite State Machines,
Notications,
Brokered Notications,
Security concerns, and
Practical Considerations.
Chapter 8: Event-Driven Architecture
319
1. Overview
Event-driven architecture as a paradigm focuses on the notion of an event as its
primary abstraction. Events can be generally classied as any occurrence, activity
or change in state. Events also have attributes such as location (where it occurred),
interval (how often it occurred) and time (when it occurred). Events themselves
are usually processed with some form of delay, which affords the opportunity to
ignore particular events, delay response or immediately start processing. Abstracting
business activities into a event-driven paradigm will help enterprises scale and align
services with real world business process. Within the vast majority of enterprises,
event handling within applications occurs via messaging paradigms.
Traditional applications may use queuing products and interact with other applications on a request-reply basis. Applications in this scenario send a message (event)
to other applications via queues and wait for reply before continuing processing.
Application level queues in this scenario may decouple the transport layer from
a connectivity perspective but does not provide features to aid in scalability or
availability since both applications become coupled to each other in a point-to-point
interaction.
Usage of queues does not guarantee asynchronous application interaction. The
event-driven approach to service development has similarities with traditional objectoriented development but departs in subtle ways. The rst departure comes by the
techniques used to capture and model the business domain. The traditional objectoriented approach may start with a UML class diagram that captures objects and
their attributes. In this approach, the attributes determine their state and methods
describe an objects behavior. In an event-driven approach, modeling may start by
capturing the participants in an event, the roles they play in the specied scenario
and modeling of state transitions.
Classes and their relationships traditionally captured are discarded in pure eventdriven architectures. Within an event-driven approach, state is determined by event
binding amongst participants and not within an object itself. Likewise, constraints
on state transitions are modeled as conditional rejections. Events themselves have a
transactional nature to them when put into a state context in that they can become
classied as reversible or irreversible states.
Let us dive deeper into the details of being event-driven.
320
2. Events
Events can be loosely classied into the following categories:
Descriptive,
Prescriptive,
Factual, and
Assumptive.
Let us look into the details of each event classication.
2.1. Descriptive
Descriptive events are declarative in their approach and may be typically used to
classify scenarios. Events can be further grouped into hierarchies and can also serve as
aliases. For example, a participant (customer) in an event-driven scenario interacting
with a consumer order entry service may also be classied as an individual (if single)
or as a household (more than one person at the shipping address).
Descriptive events can also place participants into named groups (sets) as well as
leverage a set of criteria. For example, an income tax service may classify wealthy
Americans as those individuals who have an annual income greater than $40,000
per year and own a car. Events can belong to multiple groups at the same time. In
the preceding example, the subject could be separately classied as both wealthy and
American.
2.2. Prescriptive
Prescriptive events either serve to quantify or constrain. For example, a university
course registration service may constrain students (participants) to how many courses
they can take in a single semester. Constraints can either specify exact quantities or
lower and upper boundaries in terms of ranges or intervals.
Constraints are usually applied to events that have a limited duration from a time
perspective. For example, an insurance policy may constrain the reporting of physical
property damage to the effective date of the policy to ve business days after
the policy expiration date. Constraints can also be applied to contingent scenarios
321
whereby events are dependent upon a series of other events occurring. For example, a
product warranty may trigger an obligation to repair an appliance at no charge if the
product is damaged during normal product usage within one year from the date of
purchase provided that the consumer purchased additional warranty coverage within
ninety days of purchase.
2.3. Factual
Factual events are events that have occurred and can include both business and userinterface events. Business events include contractual events which specify the rights
(descriptive) and responsibilities (prescriptive) of each participant. Contractual events
are descriptive in that they incorporate denitions of the rights and responsibilities.
Contractual events are also prescriptive in they specify authorizations and obligations
on both parties. Factual events can also include workow events which demonstrate
the fulllment of responsibility.
2.4. Assumptive
Assumptive events are events that may or may not have actually occurred. They may
assumed to have occurred or anticipated to do so in the future. When the enterprise
wants to develop systems that are predictive in nature, it may allow for assumptive
events to be created. Sometimes it is useful for a semi-informed event to be created.
For example, an event monitoring service may need to may certain assumptions if it
receives multiple events that are part of a business transaction out of order. In this
situation, any actions taken by a service need to allow for events that later prove or
disprove the particular action taken.
2.5. Business Rules

Many enterprises have adopted the usage of business rules engines as a strategy of
allowing exibility in changing business rules without having to involve IT resources
to write in a proprietary language. The process of capturing and modeling business
rules is very similar to modeling of an event-driven architecture. In the event-driven
paradigm, a business rule can be thought of as a business policy as it describes how
specications should be implemented in a prescriptive manner.
Policies have several components:
Participants,
322
Roles,
Conditions,
Events, and
Classication/Scenario.
Examples of participants in business rules are customers, suppliers, supervisors, etc.,
whose activities occur in context of a role such as purchaser, approver, or authorizers.
Business rules also have conditions such as under which circumstances can an
authorization event will be triggered. Usually the entire rule will occur in some
particular context/scenario and a classication scheme applied to it such as order
entry, quoting, claims submission, etc.
Traditional business rules processing however is usually implemented in short-lived
transactional scenarios. Event-driven architecture can be considered a superset in that
it captures the lifecycle of all policies (business rules). When a rule res, it only
maintains context from its entry point but otherwise becomes isolated from other
events within the enterprise. Sometimes it is necessary to understand the state of
transactions outside of the currently executing context. Rules such as the trading
service can only accept trades from six wealthy individuals concurrently that are
submit limit orders for after hours on securities that trade on NASDAQ and the last
trade was an up tick would be difcult at best to construct using a pure rules-driven
approach and would require tons of awkward logic.
Traditional approaches to rules engines operate on the notion of forward chaining.
Rather than specifying an expression for evaluation, the rule set is run against
the current knowledge base allowing it to assert the facts for each rule has it is
preconditions satised. Rules engines generally iterate until no rules have all of their
preconditions satised. The problem with this approach is that the knowledge base
is somewhat static in that another rule in another set may invalidate conditions that
would have caused the rule to execute.
Some rules engine implementations will allow a user to leverage backward chaining.
Backward chaining starts with just a set of preconditions and no assertions. The
engine will search the knowledge base to nd the facts that satisfy the constraints.
Backward chaining is useful for constraint-based searching and can result sets can be
returned iteratively instead of running to completion. Backward chaining is closer
in spirit to event-driven architecture but forward chaining has won the popularity
contest.
To learn more about Business Rules, we recommend Ross (2003).
323
3. Agents
Imagine a situation where a client from one enterprise wants to conduct dynamic
business with another enterprise in an integrated supply chain manner. The ability to
provide run-time integration and service provisioning at run-time is crucial. In this
particular scenario, if no particular service existed that could satisfy the functionality
required by the rst enterprise, there should exist the possibility of combining existing
services together in order to fulll the request.
One challenge that has not yet emerged within the enterprise is the ability to
understand the limitations of human capability to analyze the required services within
a transaction and create compositions using approaches such as BPEL manually. At
the heart of the matter is the notion of service composition as it is more complex
than it appears at rst and cannot be solved without good architectural discipline and
several degrees of automation.
Enterprise SOAs that are event-driven will introduce the notion of an agent. An
agent can launch several responses based on a business event, each of which can be
executed independently often requiring no further action from the initiating agent.
Agents can be congured statically or learn dynamically to watch for a range of events
that may or may not happen, and may occur in an order that is difcult to design for
upfront. Building applications and services that can be assembled in arbitrary ways
usually requires heroic efforts when not using an event-driven approach. Until now,
only a few services were constructed using event-driven approaches such as trading
applications but have now became realizable for the enterprise to incorporate into all
enterprise applications.
Before we jump into business events, agents and other advanced concepts, let us
revisit some of the principles we discussed in previous chapters and put them into an
event-driven context.
In Chapter 4, we learned techniques for describing and registering services. Many
typical implementations of registries permit services to be referred to by function
signatures that are somewhat opaque and provide virtually no indication of the nature
of the services being managed. Services themselves in their description, require a level
of expressiveness in order to work in an agent-based model. The description of a
service needs to include its meaning and intent rather than simply ascribing a name
to it.
324
The term Ontology is borrowed from philosophy and represents a systematic account
of existence.
For example, a semantically poor description of a service interface would be that it
takes as input an integer and three strings are returned as output. A semantically rich
description for this service interface would be that it takes the name of a company
and returns (i) the stock exchange it trades, (ii) the email address of the CEO, and
(iii) the CEOs shoe size.
Semantic is the study of meaning and the changes in meaning.

One of the more important attributes of an enterprise SOA is that services should be
composable. The dynamic composition of services requires the location of services
based on their capabilities and the recognition of those services that can be matched
to create a composite offering. Approaches that XML markups such as BPEL when
taken to their logical conclusion simply will result in yet another legacy. BPEL relies
on the nature of static binding of services in order to form composition and does not
provide the ability to discover possible execution paths at run-time creating another
form of coupling. Semantic understanding is the only thing that will assist in solving
this problem.
For example, a web retailer wants to sell products similar to those on Amazon.com
and wants to expose a web services interface. Likewise, a consumer is shopping for
products related to ancient Chinese folklore and specically interested in learning
more about the legend of Fong Sai Yuk. The transaction requires choreography
between two different services; an online language translation service and a dictionary
service. The language translation service can handle translating text between several
languages where as the dictionary service can return the meaning in English. The
problem with describing this type of interaction in WSDL is that it would only
contain a description of strings as input and output rather than the concept of how
they can be combined.
Current practices around service descriptions will over time become exposed as being
impoverished. The savage pursuit of describing the functional signature of a service
and its interfaces instead of characterizing their meaning will simply result in yet
another integration problem down the road. The ability for the enterprise to support
service discovery services (above and beyond registries) is an essential component to
achieving loose coupling.
325
Figure 1. Service Ontology

The service prole describes what the service does, what it requires of consumers or
other agents and what it provides for them. The service model describes how the
service works and the service grounding describes to the service can be accessed.
The service as shown in Figure 1 provides an organizational point of reference
for declaration of services. The properties presents, described by and supports are
properties of the declared service. The classes Service Prole, Service Model and
Service Grounding represent the declared ranges of those properties.
Service proles state what the service does and give the types of information required
by an agent seeking services (or its intermediaries) to determine whether the service
meets its needs. The prole is also used to state the requirements of the agent seeking
services so that an intermediary such as a matchmaking service which will pair parties
together has a convenient dual-purpose representation on which its operations can
be formed.
Service models state how the service works. For services that are composed of
several steps which occur over an extended period of time, the model will provide a
description that can be leveraged by agents. The description can be used by the agent
to perform the following tasks:
Direct the agent to perform an in-depth analysis of whether the service meets
its needs.
Create a composite service description from multiple services to perform a
specic task.
326
Under service invocation, allow the agent to coordinate the activities of different
participants.
Provide the ability for the agent to monitor the execution of the service.
Service groundings state the details of how agents can access the service and will
contain information on which communication protocols to use, message formats
and other service-specic details required to interact with the service. The service
grounding also contains specications for how to exchange data elements for abstract
types contained within the service model. This could be a specication on how to
serialize/deserialize the data type.
Current service interactions typically show a client querying a centralized registry
(nd) to discover the location of a service endpoint (bind and execute). What is
not taken into consideration is the ability of a service provider to locate potential
consumers of a service and provide pieces of the puzzle to the client in which the
required interaction has not been predened. Architectures that are agent-based can
provide the opportunity for a service provider to become proactive in the service
composition process.
To learn more about agent-based approaches to services, visit http://www.daml.org.
3.1. Service Design

The rst stage of service design and development used within many enterprises starts
with a highly synchronous interaction between producers and consumers. While we
can use management approaches to load balance services to achieve scalability, it is
important to talk about proper service construction techniques. The ability for the
service itself to perform asynchronously will yield higher performance over other
approaches and sustain response times with huge variations in load.
One commonly used approach to handling huge variations in load is to overprovision
resources. Many web sites simply procure enough servers to handle the anticipated
peak load and use various load balancing approaches to cause even utilization
between them. Usually this strategy works well in situations when resources used are
commodity in nature, low cost and readily available but otherwise will fail when the
ratio of peak to average load is very high. It would be unrealistic to purchase hundreds
of machines to support peak loads when the average load could be served by only a
few machines. The traditional approach of building farms in this scenario will not
meet the scalability requirements but will also result in wasted expense. The cost of
managing a large number of servers over time can grow signicantly higher than the
multiplicative cost of the servers themselves.
327
The mindset within the enterprise needs to move away from the mindset of developing
infrastructure to support peak loads to developing services that are mission-critical to
be well-conditioned to handle load. A well-conditioned service has the characteristic
that when the number of requests exceeds its intended capacity, it will minimally not
over commit usage of operating system and downstream resources that ultimately
degrade the responses of all clients.
Services that can detect overload conditions and attempt to adapt their resources to
them can use strategies to shed load in a predictable fashion by returning a predened
response to the consumers when it is saturated or by degrading the quality of service
based on either amount of data returned and/or actual response times. Ideally, this
should be a congurable behavior within each service instance. The ability to inform
service consumers that the service is overloaded is far better than dropping service
requests.
The architecture used to develop services should also address:
Handling large variations in service utilization.
The ability to support massive concurrency of service requests.
Decoupling event handling and load management from service logic.
Allowing policies for load management to be congurable.
The ability to provide classication in a generalized manner to service requests.
There are many approaches that can be used in creating highly adaptable services that
realize the above goals. Some of the techniques that can be used are:
Resource Pools,
Multi-threaded Designs,
Soft References.
In order to understand how they can be implemented within an enterprise, we will
need to revisit several software engineering concepts and see how they can be applied
to an enterprise SOA.
3.2. Pools
Pooling mechanisms are useful in situations where resources are limited and can cause
performance bottlenecks where there are not enough resources to meet the demands
328
of incoming service requests. Connections to relational databases are a good usage

of a pooled approach since the performance costs of creating and destroying the
connection is reduced.
Pools can provide functionality to allow clients to reserve resources in advance of their
usage and return the resource to the pool when it is no longer needed. Pools can also
solve for resource related slowdowns in situations when there are more connections
to a resource, the longer it takes to create new connections.
Pools are used in situations where users of the pool assume that all resources contained
within it are interchangeable. For example, if a user needs to query a relational database
and the database connection is in a state that will accept queries, it does not matter
to the client application which particular connection is used.
Sometimes pools are implemented as singletons where a generic pool will hold a
variety of connection types. In other situations, there may be a different pool for
each resource type (this is highly recommended). Most application servers provide
a generic pooling mechanism for relational databases but do not provide pooling
support for other connection types such as directory services, legacy applications and
so on.
public interface PoolableObjectFactory {
Object create();
void activate(Object obj);
void passivate(Object obj);
void destroy(Object obj);
Code 1. Poolable Object Factory
A pool may sometimes implement other patterns such as Factory to ensure consistent
lifecycle support. This will allow a pool to be written in a manner that will allow a
client to select pooling behavior independent from the types of resources contained
within them. An example is contained in Code 1.
Pools themselves can provide congurable constraints on entities contained within
it such as maximum number of connections, who can use them and for how long
and load balancing / failover in situations where connections to existing resources
fail. Pools can also make service startup and recovery faster as they can delay creation
of connections to other resources until they are required for timely client response
times.
329
4. Threads
Many enterprise applications leverage frameworks and application servers that provide
a simplistic approach to simultaneously support multiple concurrent requests. These
frameworks hide the complexity of developing multithreaded applications from
developers and meet their goal of making application development simpler. In
most situations, the approach used by application servers and frameworks provide
a sufcient level of concurrency with appropriate response times. However these
same frameworks and application servers will most often fail in situations with
unpredictable load.
Modern operating systems support the notion of a thread of execution in which the
operating system determines what will execute. A thread executes in context of a
process. Processes are used to group resources together and threads are the resource
in which the operating system scheduler will schedule for execution on the CPU.
Usages of threads are a common approach to improving the parallelism of applications.
The operating system will rapidly switch execution of threads to give the illusion of
parallel execution. Threads like services can be in any of multiple states; running,
blocked, ready, terminated. Threads can be useful in applications that are required to
do more than one task at a time but can be detrimental in any application that has a
sequential process that cannot be divided into parallel tasks as they would block until
the previous one completes.
There are two pervasive approaches used in developing multithreaded applications:
1. Thread per Request, and
2. Thread Pools.
4.1. Thread per Request

Thread per request approaches are used by most architectures that are based on
remote procedure call style of interaction. Microsofts Distributed Component
Object Model (DCOM) and Javas Remote Method Invocation (RMI) use this
approach. In the thread per request, each accepted request is dispatched to a thread
which processes the request. Locking techniques based on semaphores are utilized to
protect the integrity of internal data structures and shared resources.
330
Semaphores are a technique for coordinating or synchronizing activities in which

multiple processes compete for the same operating system resource.
Figure 2. Thread per Request

In the thread per request model (Figure 2), incoming service requests are dispatched
to separate threads. In this model, each thread is responsible for performing all
activities of the process and returning the result to the client. Each thread may use
operating system resources such as accessing les which are also incorporated into
each threads process.
Thread per request is commonly used because of its ease of programming as all tasks
can be thought of as and are executed inline. This model also provides isolation from
all other currently executing requests and only becomes noticeable when handling
shared resources or state.
The original reason for usage of threads within operating systems was in support of
timesharing. When the number of threads that need to execute concurrently grows
signicantly passed the number of processors contained within a server, the operating
system will start to thrash. The operating systems scheduler will experience scheduling
overhead and increased cache misses resulting in a rapidly increasing slowdown of all
operations. The thread per request model has serious limitations in that it provides
zero capability in terms of resource management.
331
4.2. Thread Pools

The second approach used in many application servers such as BEA Weblogic, IBM
Websphere and Oracle 10g Application Server is the notion of thread pools which
serve to bind the number of threads associated with a service. Thread pools dispatch
to a bounded pool of threads. When all threads within a pool are busy processing
requests, outstanding requests are queued until a thread within the pool becomes
free.
Many application servers may dedicate a xed number of threads (sometimes
congurable) to handle incoming client connections. When the service has more
connections than it can handle, additional clients that attempt to connect may
be dropped even though additional worker threads may be free. In this situation,
the perceived performance from the client may be dramatically different than the
performance of the actual service.
Most communication between service consumer and the services themselves will use
the TCP/IP protocol. When the service cannot handle additional connections, it
may simply ignore the request and not provide any response back to the client. The
client in this situation will usually know that the service is unavailable based solely
on network level timeouts. In situations, where the connection was processed but the
transmission of the request is overloaded due to unavailability of worker threads this
may also result in a rapid decline of perceived service response time.
The TCP/IP protocol uses an exponentially increasing timeout value for retransmitting connection attempts (SYN) and will result in long connection delays on the
part of clients. This will also have the added effect of providing good response times
to those who are already connected while penalizing those clients who have not yet
connected.
An enterprise SOA should always prioritize client access based on business policies.
Use of thread pools themselves requires choosing the right size for bounding. The
vast majority of application servers allow for an administrator to statically dene the
maximum number of threads within the pool. Dening the thread pool with too few
threads will result in underutilization of server resources and not providing maximum
throughput. By dening thread pools too high, will result in severe degradation of
response times and server resources under heavy utilization.
Usage of thread pools in many environments may also result in the consumption
of all available threads with long-running requests which will prevent requests that
consume fewer resources to be turned around quickly. It will also starve requests
332
that are of a higher priority. Many application servers will solve the former problem
by segmenting pools based on the type of request but do not necessarily provide a
solution to the latter.
The usage of queuing via dispatching may help in some situations but since most
implementations of queues use rst-in, rst-out (FIFO), there is no ability to
support the prioritization of requests based on either business priority or resource
requirements. The dispatcher approach is also limited in that it cannot preempt a
busy thread for another thread that contains a request that uses less resources or of
higher business priority.
A clever enterprise developer may be able to develop a framework that solves
for the threading issues discussed to date, but this would be missing the entire
point. The author team recommends leveraging existing frameworks and application
servers where appropriate but to consider alternative approaches to commonly
used techniques that were held over from client/server, J2EE Blueprints and other
paradigms of the past.
5. Alternative Pattern-Based
Approaches
One approach that can be used to develop intra-service communications is to consider
leveraging several commonly used design patterns such as:
Strategy,
Chain of Responsibility,
Interpreter,
Flyweight, and
Memento.
For additional information on design patterns, we recommend Shalloway and Trott
(2004).
333
5.1. Strategy Pattern

The intent of the strategy pattern is to dene a family of algorithms, encapsulate
each one and make them interchangeable. The strategy pattern lets the algorithm
vary independently from the services that use it. The implementation of a strategy
usually uses internal variables to record the state of the algorithm and may also keep
the results in this structure.
If services have potentially generic routines embedded within them such as determining business priority and/or resource utilization handling, at a minimum it
becomes difcult to re-use these algorithms and dynamically exchange them. Use
of the strategy pattern will allow new strategies to become pluggable and assist in
decoupling different layers of functionality. Strategies will also allow you to vary your
choice of policy at run-time.
Figure 3. Strategy Pattern

Usage of the strategy pattern for queuing and resource access can allow the
prioritization of requests based on resource consumption and business priority.
5.2. Chain of Responsibility Pattern

The intent of the chain of responsibility pattern is to help avoid coupling the sender
of a request to its receiver by giving more than one object a chance to handle the
334
request. Objects are placed in a chain and the request is passed to each object until
one decides to handle it. The number and types of objects that can handle the
request are not known at design time and thus can be congured dynamically. The
approached used to chain objects typically use recursive composition to allow an
unlimited number of objects (handlers) to be linked together.
Figure 4. Chain of Responsibility
Chain of responsibility reduces communication and connections between objects.

Each sender keeps a single reference to the head of the chain, while each receiver
keeps a single reference to its immediate successor in the chain.
Generic handlers can be developed to commonly used functionality without requiring
changes to underlying services. This approach is used by .NET remoting to implement
the concept of channels. The client will make a method call that gets passed to a
formatter sink that handles serialization of the request over the wire. In turn the
receiver uses a transport sink to receive all incoming requests. The transport sink
further passes along requests to the formatter sink which deserialize the request which
passes it to a dispatcher sink which in turn makes the actual method call and returns
the results to the client back through the chain.
Use of the chain of responsibility in a services paradigm can be used to add additional
processing steps into a service call without requiring recompilation and can be done to
handle business policies. Technical features such as logging, encryption, translations,
load-balancing and replication are ideally implemented using this approach.
335
5.3. Interpreter Pattern

The intent of the interpreter pattern is given a language to dene a representation
for its grammar along with an interpreter that uses the representation to interpret
sentences in the language.
Interpreter is best used in situations where your problem domain can be expressed as a
mini-language. Typically, the interpreter will translate each instruction by assigning a
grammar rule to a designated class. Rules about production can be dened as objects,
instead of classes that result in lighter weight code processing.
Figure 5. Interpreter Pattern

When services use the interpreter pattern, they can interpret the incoming request
and effectively use switch-based processing instead of threading to handle requests.
Interpreter pattern is used to take requests and determine which queue a request
should be placed. Ideally it is used with a multiple queue approach or in situations
where incoming messages contain hints of actual business priority within the message
336
payload itself. Logic can be processed and tuned via conguration over time without
requiring recompilation.
In Figure 3, the UML class diagram demonstrates a small portion of a policy
administration system that could be used within the insurance industry. Using this
approach, a developer could quickly add additional functionality such as supporting
endorsements (additional coverages) that are not part of the standard policy process
by implementing it in a separate class without requiring changes to other aspects of
the system.
5.4. Flyweight Pattern

The intent of the yweight pattern is to use sharing to support large numbers of
ne-grained objects efciently. Since many SOAs use XML, the yweight pattern
can be applied to the message in situations where the same information is included
at many different points in a document. If the same data contained within the
document is repeated at multiple locations and the data changes, every occurrence of
the data must be located and updated.
The yweight pattern has the potential to increase performance but is sometimes
counter-intuitive to object-oriented design. Principles of object orientation usually
suggest that it is best to encapsulate everything within an object so they can be
uniformly accessed, extended and re-used. Within a large enterprise, there exist
thousands of potential objects. The time to properly model all the potential objects
combined with the run-time performance of them, will almost guarantee sub par
performance. The approach of everything is an object was implemented in several
languages including Smalltalk. From a purists perspective, it allowed near perfect
architectures but fell apart in performance. Modern languages such as Java and C#
understand this simple truth and provide a mixed implementation of both objects
and intrinsic types.
In Figure 6, we show the relationship between pools of objects for a stock quoting
service. The quoting service needs to respond to quotes in a timely manner. There is
data that experiences lots of changes (market pricing, highs and lows) as well as data
that is relatively static (such as institutional investors, which indexes it is part of, or if
the security is an index, which securities are contained within it and so on).
337
Figure 6. Flyweight Pattern
5.5. Memento Pattern

The memento pattern can be used to provide solutions to multiple problems with
increasing concurrency.
The rst usage of the memento pattern is in the development of a scheduler that
gures out optimal utilization of resources as an alternative to FIFO-based queuing.
The scheduler can implement an optimization routine that requires no knowledge of
the internal implementation details of resource or how a particular schedule involving
those resources is evaluated. The routine can simply be supplied with a sequence
of requests to be executed and returns a weighting representing the quality of the
proposed execution path. It can also reorder the sequence of requests and try again
to see if it can get a better weighting. The optimizer may determine scheduling
based on prior knowledge of how long previous tasks took to run. The problem
becomes evaluating sequences from the beginning each time. The optimizer can
choose to evaluate the state periodically and store a memento which then becomes the
equivalent up to the corresponding point in the sequence and will allow optimization
execute signicantly faster.
The quote system referenced using yweights can also leverage mementos. The
quote system may maintain current information on thousands of securities. As we
previously mentioned, some of this information changes frequently while other
aspects are relatively static. When a quote is requested for a particular security, a
338
Figure 7. Memento Pattern

template is used to provide formatting of the output of the response. The memento
pattern in this situation can take the data elements that have changed, encapsulate the
translator state and store it requiring us to only retranslate a subset of the response.
Another usage of mementos may be in situations where the framework has the ability
to preempt currently running tasks. The framework can swap out currently existing
tasks and store them to persistent storage. If it is running in context of a chain
of responsibility, it can also implement check pointing type functionality whereby
the framework provides a cache of every n-th request of a particular type allowing
checkpoint restart.
6. Language Specic Constructs

Many modern languages use a virtual machine paradigm and prevent developers from
creating unstable applications via the usage of pointers. Instead, they dereference
actual memory address access via objects and performance cleanup on memory
using garbage collection approaches. Typical approaches to development when
incorporating event-driven approaches may in many situations result in resource
starvation. Let us look at three language specic constructs that are useful in an
event-driven paradigm:
1. Soft References,
2. Forking, and
3. Non-blocking I/O.
339
6.1. Soft References

In the example of our stock quoting service, we had references to thousands of
objects. In situations where memory is stressed, we do not want the service to not
respond nor return an error to the end user. We also would like to avoid out of
memory exceptions whenever possible. In the design of the stock quoting service,
it can use reference objects to create a situation in which the garbage collector can
reclaim individual security objects when heap memory runs low.
Our scheduler can implement a form of a reference queue to create a situation where it
is notied when a certain object or set of objects are reachable only through reference
objects. Upon notication, the scheduler can proceed with clean-up operations,
checkpoint state, deserialize other dependent objects within the chain and make them
eligible for garbage collection as well.
Java provides three different forms of references:
1. Soft,
2. Weak, and
3. Phantom.
Soft references are used in Java applications in situations where there are lots of
ne grained cached type objects and not knowing in advance if the service will
actually re-use cache entries. If the garbage collector reclaims objects that are softly
referenced usually do to lack of access over an extended period, the service will need
to recreate the object. Using soft references will allow a service to become aware of
its own resources. Additionally, within most virtual machines, the garbage collector
is required to clear all software references before throwing an Out of Memory Error.
Weak references are usually used in services that need to hold a reference to an
immutable object and may contain supplemental information about it. Our stock
quoting service may have an index constructed as a composite object that holds
multiple weak references to underlying securities that make up the index. If the weak
reference to an individual security gets destroyed, the index can be notied of this
event and perform any required cleanup of the associated object which could range
from displaying an older cached price to displaying a message that current pricing is
unavailable.
Phantom references are not recommended for usage in services and therefore not
discussed.
340
6.2. Forking
In certain situations, instead of taking on the difcult task of creating multithreaded
services, especially if you are not using an application server that provides a framework,
a simplistic approach may be to create new processes dynamically based on load. Web
servers such as Apache use this approach to create highly scalable, fair processing
HTTP services.
The ability to create new processes based on incoming requests provides the capability
to use third-party non-thread safe libraries for services. Creation of new processes per
request will also provide a level of isolation for each request so that if a problem arises
with a single request, it will not with others.
Services that use forking techniques will have a single control process that is responsible
for launching child processes which listen for connections and service client requests
directly. The control process will also maintain several idle server processes so that
clients do not need to wait for new child processes to be forked.
Fork is a feature of most operating systems that creates a new process. A forked
process will inherit various properties from its parent process.
6.3. Non-Blocking I/O

Event-driven architectures place an important limitation on event-handling code
by requiring it to be short and run to completion. When threads of execution
process event handlers, they must avoid stalling processing and should ensure
equality of execution across a large number of requests. This is best accomplished
using asynchronous non-blocking I/O. I/O can be incorporated at multiple layers
including when consumers connect to the service (sockets), to when the service writes
les to disk (asynchronous le systems).
Java provides non-blocking I/O support via the java.nio package.
When comparing the various resources used in a computing environment such as
CPU, Memory, Disk and I/O, I/O has been traditionally the hardest to optimize.
I/O features of most general purpose operating systems were designed to provide
maximum transparency to applications at the expense of being able to determine
predictable behavior and a sometimes not so insignicant hit to scalability.
Even with the availability of commodity hardware such as Intel, free operating systems
such as Linux and large hard drives generally available from Wal-Mart, enterprises
341
still manage to not only keep their mainframes, but upgrade them on a frequent basis.
Mainframes still exist as viable platforms to build services, not due to their blazing
speed of execution (not!) nor the ability to support COBOL, but in their ability to
handle I/O better than any other computing platform.
Mainframe architectures have had the ability to virtualize hardware resources, which
is now only starting to become enterprise-class on other platforms. Most operating
systems do not provide the ability for any applications to participate in enterprisewide resource management decisions as services will typically execute across process
and sometimes even server boundaries. The only implementation that the author
team is aware of that affords this capability is IBMs Workload Manager.
6.4. Enterprise Service Bus

The concept of an Enterprise Service Bus (ESB) is a key component of an Enterprise
SOA. Some will think it is a product that can be acquired from a vendor while others
will think it is a set of standards (this is closer to the right answer) while those in the
know, understand it is really about infrastructure capability that can be built in-house
using tools and technologies that you may already own.
The Enterprise Service Bus support message and event-based interactions amongst
services in heterogeneous environment while providing guaranteed delivery and queuing capabilities. The enterprise service bus provides multiple levels of infrastructure
support to services including:
Separation of transport (communication protocols) from service implementation
to stress both location transparency and interoperability.
Governance around the creation of implementation-independent service
interfaces.
Denition of services themselves that provide encapsulation of existing business
functionality.
Enterprise SOA mandates that both applications and infrastructure support SOA
principles. At the infrastructure level, SOA involves provisioning of the capabilities
to route and deliver service requests to the correct service provider. The infrastructure
must also support the ability to substitute one service implementation for another
without clients being aware of the change from either an interface or response time
perspective. In order for this goal to be achieved, clients must be able to invoke
services in a way that invocation is independent of the service location and the
transport used.
342
Many enterprise service buses can be implemented on top of existing messaging

infrastructures but may require changes in how they are thought about and
implemented. Traditional messaging infrastructures were described as hub-and-spoke
architectures while services are usually described in a distributed fashion. In reality,
the real problem space is the requirement of centralized control and distributed
infrastructure.
The bus provides common interface to both consumers and providers as shown in
Figure 8 as well as a centralized administration capability to control service addressing
and naming. The consumer never directly binds to a provider and therefore both can
involve independently of each other.
Figure 8. Logical View of Enterprise Service Bus

The bus will usually support at least one form of message such as request/response,
publish/subscribe, one-way and so on. It will also provide capabilities to support
multiple transport protocols such as HTTP and JMS-based communications methods
with the option of plugging in additional transports. Finally the bus may expose an
interface to legacy applications without requiring them to create their own. Usually
this is accomplished through various integration methods such as Java 2 Connectors,
third-party adapters or proprietary protocols such as Siebels EAN.
343
The ability to make many existing enterprise applications understand a common

message format so they can participate in enterprise-level integrations is futile. The
enterprise also has data stored within spreadsheets, at les and other forms that
would be ideal to incorporate but becomes challenging to wrap. The enterprise also
must consider integration with current XML standards such as SWIFT, ACORD,
Parlay and others where they are represented in XML but are not widely adopted
and/or do not have standardized representations.
Loose coupling is formed by decoupling all interactions between consumer and
provider and goes beyond typical point-to-point integration provided by basic web
services. Basic web services provide zero centralized administration capability to
control service addressing and/or naming. Service names are administered by each
exposed service interface and routing is determined by the consumers proxy interface.
Consumers typically know the address of specic service implementations and directly
bind to them using specic protocols at specic addresses. Many enterprises have
been smart by requiring all services to be published in a centralized registry (such
as UDDI) but this is limited in that clients may still cache endpoints and there is
no standard that forces clients to periodically query the registry for changes (such
as leases) nor by avoiding the registry altogether and simply retrieving the WSDL
directly. Substitution of one service for another becomes difcult and may even
require changes to service consumer logic.
The bus can also implement functionality in a centralized manner to multicast events
to multiple service endpoints. If a particular service is ofine, it can optionally queue
requests until the provider comes back online. Since the consumer and producer
are not aware of each other, the bus can provide value added services such as
implementing callbacks, notications, publish and subscribe and the ability to have
long-lived transactions. A consumer and provider interact with each other by simply
placing events on the bus and continue processing without blocking.
Since both clients and services interact directly with the bus and not each other, the bus
can provide admission control functionality and can even implement functionality
to ensure that all services receive the same message at the same time or can prioritize
service requests based on who the service consumer is and other business policies.
For additional information on ESB, check out Keen et al. (2004).
344
6.5. Callbacks
Many event-based applications under the hood implement implicit invocation of
underlying services and limit asynchronous behavior at the service entry point. Usually
this occurs because event-based services make correctness reasoning more difcult
than simple procedural approaches. The ability to achieve further asynchronous
behavior within the service can be accomplished by allowing a service to register a
public interface for implementing callbacks.
Applications that subscribe to the principles of event-driven architectures when
they cannot complete an operation immediately will register a callback and defer
processing to a later time. Event-driven applications may poll for events and execute
the appropriate callback when the event occurs. Callbacks can be done at the
service interface or can even be implemented at each of the layers within an application.
A callback is a function that will be invoked whenever an event is received.

When a consumer of the service contacts a service provider and the request cannot
complete in real-time, the client may indicate a return address it wants the reply to
the service request to be sent. If an application decides to implement callbacks at
each layer in the software stack, it may decide to drive application logic by a loop
that polls for events and executes the appropriate callback when the event occurs.
Callback processing can execute indivisibly until a blocking operation occurs. Once
a blocking operation happens, the layer processing the particular event will simply
register a new callback and return control to the caller.
7. Finite State Machines

Finite State Machines (FSM) are used to model the behavior of a system that contains
a limited number of dened conditions and transitions. The notion of state is actually
encompassed into many applications. Each application may start with an initial state
that change over time as it begins to service requests from end users. At any point in
time, an application can be viewed a complex set of states. State machines have the
following attributes:
A description of the initial state.
A set of input and output events.
345
A set of states.
Functions that map states and input to output.
Functions that map states and inputs to states (This is known as state transition).
Rules or conditions that must be met to allow a state transitions.
Finite state machines are limited in that they have a limited number (nite) of possible
states. It is possible, but not practical to construct an innite state machine due to the
inability to dene innite transitions and the rules that must be met between them.
Finite state machines use an initial state as a starting point and keep track of the
current state based on the last state transition. Input events serve as triggers which
cause a rule or condition to re that govern transition from the current state to future
state. Finite state machines are traditionally used in articial intelligence applications
and services and whenever semantic networks are constructed.
Finite state machines are similar to rules-based approaches in many ways. For
example, if all the antecedents of a rule is true then the rule is red. Only one rule
can re at a time. If the potential for multiple rules to re exists, then a conict
resolution strategy is required to determine which individual rule to select. This
ultimately results in a state transition.
State may involve one or more actions.

Finite state machines may be implemented using the controller paradigm which
serves as a switching mechanism for incoming events. Worker threads serve the
controller by evaluating current state by execution of switch (case) statements and
the resulting code for that state is executed. During the execution, one or more
actions may be performed and all state transitions are noted. The controller may also
implement pre-processing steps on the input for validation and security purposes and
post-processing steps for auditing purposes.
The diagram in Figure 9 represents a simplistic view of placing an order via the stock
market. A broker will list the stock order on a stock exchange where the order will be
listed. The order may have preprocessing validation steps to validate elements of the
order such as whether the stock trades in fractions or decimals, whether the ticker
symbol is valid and if the exchange itself trades the specied security. Once the order
has been accepted by the respective exchange, it stays valid for a period of 60 days
(exchange rules) and if not executed within this time period may be automatically
346
Figure 9. State Transition Diagram

cancelled. Likewise, if the order is executed, both buying and selling parties are
notied and the clearing process begins.
Finite state machines are used in two different manners. First, the nite state machine
species the type of operations which may be invoked on a service request. Any
resulting transitions that are triggered will invoke the desired operation. This allows
the service requester to invoke the operation at that point. In this scenario, the state
machine does not specify the behavior of any service requests and resulting operations
but merely records the change of state and determines which operations can be
invoked.
Finite state machines can also specify the complete behavior of its context (such as a
business process). In this scenario, Service consumers can send requests to the owner
of the state machine (can be an orchestration engine, ESB or composite service) which
the state machines receives the event and determines what the outcome will be by
attaching actions to transitions complete specications of operations are determined.
347
Finally, nite state machines can be deterministic or non-deterministic. Deterministic

nite state machines work on the principle that the next state is uniquely determined
by a single input event. Non-deterministic nite state machines not only depend
on the current input event, but rely on an arbitrary number of future input events.
Until these future events occur, it is impossible to determine the state the nite state
machine is in.
Finite state machines can become complex over time and introduce difcult to
code behaviors even with a few states. The optimal architecture for the enterprise
assumes that there will be multiple distinct implementations of a nite state machine
each with a specic task. One analogy to draw a parallel to is in the military. The
decision-making processes that are used for launching airplanes off the ight deck of
a carrier are distinct in nature from how the United States Marines may protect an
embassy under siege.
For additional information on nite state machines, see Lawson (2003).
8. Event Notication
Event notication is similar to publish/subscribe in several ways but can also be
considered its big brother. Publish/subscribe has been around for several decades and
is based on the paradigm of message routing. Messages are delivered to receivers based
on a set of criteria related to either metadata contained within the message or the
message content. The message sender has no awareness of message receivers and the
forwarding and delivery of messages is handled by middleware such as CORBA.
One or more notications are emitted by an event producer and received or retrieved
by one or more event consumers possibly through a broker.
Event notication unlike publish/subscribe adheres to a contract and uses the notion
of an event-sink to request asynchronous delivery of specic events from an event
source. Events can be considered an advanced form of callbacks in that they can
leverage two different models: the push model and pull model.
The push model allows for suppliers of events to push an event into the event channel
which is delivered to all consumers who registered interest in a specied event. The
pull model occurs when the consumer of a service requests an event from the event
channel which in turns requests the event from the supplier.
348
Figure 10. Push and Pull Models

A channel is a proxy to the producer and consumer
The notication model extends basic event handling by providing reliability, priority,
expiry, earliest delivery time, maximum events per consumer, throttling and can
even implement policies related to message delivery order or when messages such be
discarded.
Notication services unlike events may implement policies that specify the order in
which events are delivered including:
Any order,
First in First Out,
Deadline,
Business Priority, and
Last in First Out.
Alternatively, may decide to implement a discard policy that tells it when to throw
away messages. The stock order entry service may use this approach for consumers
who request trades on various securities and choose to cancel the order if it cannot
execute it within ve seconds or other arbitrary number according to business policy.
349
Throttling allows a registered consumer or provider to not become overloaded in the

event of unanticipated messages. Functionality of throttling may be implemented to
specify the maximum amount of time events in a sequence will be collected before
being delivered to an event consumer. This will allow service providers and consumers
to make intelligent choices in optimizing usage of resources. For example, if a service
uses a relational database that is low on memory and multiple queries for a particular
customer are batched, the service will be able to respond faster as much of the data
may already be in the database servers cache.
By using notication approaches, consumers are no longer tightly coupled to service
providers. In fact, the only form of coupling that can exist in this model between
consumer and service provider comes from actual subscription information in the
form of a contract being shared amongst channels. Suppliers of events have the ability
to query the channel for the desired event types desired by consumers. A supplier
should not create events to place on the channel if no consumer has registered an
interest in them. Consumers can also determine when new event types are being
offered by suppliers so that can subscribe to new events as they become available.
The subscription mechanism will help reduce network trafc and the load on the
infrastructure by ensuring that only trafc that is intended to be consumed will be
produced. This helps in situations where providers may be sending out spurious
events that no one actually desires.
Subscribers can alternatively use subscriptions to group one or more topics from
publishers. The range of topics can be categorized in a tree like fashion where the
subscriber can express an interest in a portion of the tree instead of each individual
element. Publishers can provide notications to subscribers to receive information on
descendant topics without having prior knowledge of their existence.
8.1. Brokered Notication

Brokered notication allows publication of messages from entities that are not
themselves service providers and includes standard message contracts and exchanges
that allow both providers and consumers to participate in brokered interactions.
An event broker is an entity which routes notications. Brokers typically aggregate
and publish events from other producers. An event broker can also apply some
transformation to the notications it processes.
Brokers will be typically used in building of congurations where intermediaries are
used to communicate with external parties in a federated fashion. Brokers within
350
a federation can ideally implement sharing of administrative workload increasing

scalability of all members within a federation.
8.2. Security Concerns

Event-driven architectures pose additional concerns above and beyond security
mechanisms discussed so far. For example, notication messages themselves can
contain empty bodies which should be signed so that content cannot be added by
third-parties in transit.
The authors strongly recommend that all communications between services leverage
strong message level authentication and encryption.
Depending upon the frequency of message exchange, different approaches to message
security may be warranted. One off and low volume message exchanges can use
public key infrastructure (PKI) for ensuring the integrity and condentiality between
publisher and consumer but will run into scalability and performance issues when
volume increases. For high volume event-driven interactions, a better approach is to
consider establishment of a secure context for events by incorporating the notion of
trust.
The authors recommend implementation of WS-Trust and WS-SecureConversation
for high performance event-driven architectures.
In addition to traditional attacks such as impersonation, message alteration or the
ability for third-parties to intercept/read data not intended for them, event-driven
also creates additional risk including but not limited to:
Message Order alteration,
Availability attacks,
Replay, and
Redirection Attacks.
8.3. Message Order Alteration

Imagine a situation where an event-driven approach is used in a stock trading scenario
where a read-only message such as the price of a specied stock is allowed to reach
selected parties before others. In this particular situation, the party that receives the
351
message even ten seconds before another party has a potential advantage. On the
receipt of bad news, the party may be able to sell their shares for less of a loss than
parties receiving the message later. For a stock that is quickly rising in price, the rst
party may be able buy the stock cheaper than other parties and make more money.
The ability to protect against message order alteration is difcult and may require
a mechanism for which no current specications are provided. In this particular
situation, one may have to resort to proprietary implementations of secured reliable
delivery channels and other mechanisms.
8.4. Availability Attacks

Reliable delivery is also subject to a variety of attacks. Processing overhead for
reliable delivery is higher than for general communication channels and therefore
can be used to overload the services network with useless messages whose sole
purpose is disruption. If the intent is to disrupt the network from processing
messages in a timely manner, this can be accomplished by injecting certicates that
have deep recursive relationships with other certicate authorities or by establishing
authentication requests and immediately disconnecting since authentication is almost
always synchronous in nature.
Availability attacks are difcult to secure but can be minimized by ensuring that the
architecture takes into consideration the amount of state information maintained
and that it is saved prior to any authentication related sequences.
8.5. Replay Attacks

In event-driven architectures, protection against messages being replayed is vital and
is predicted to be the most popular of attacks. Replay attacks can easily be detected
using mechanisms such as including a timestamp in the header along with a sequence
identier. Downstream services can decide to process only messages within a certain
window of time and additionally keep track of not only whether a particular identier
was used but in what order it was received.
8.6. Redirection Attacks

Consider a situation in which a service is restricted to which consumers it can respond
to. If an attacker can identify a nave service, it can have it send the data contained
within the response to the nave service. The attacker could then potentially use the
352
callback URL itself or even a message ID to encode the information it wishes to gain
access to.
Redirection can also be used to launch attacks and have the attacker cover its tracks.
Let us say that the quote service has a aw in how it handles large messages (buffer
overow) but also register a callback URL of another machine resulting in the
downstream service also potentially getting corrupted data and/or getting into an
endless loop.
Redirection attacks cannot be prevented by simple access control lists that state who
is authorized to access the service. In order to secure against this type of attack, the
security policy will need to be extended to also support the ability to specify where
responses can be authorized to be sent. Minimally, each service would need to be
designed where the services themselves exchange authorization information before
the entire message is sent (authorization should not be in the message itself ).
Redirection attacks can also allow for hijacking of credentials. It becomes important
that credentials are not transitive and can only be read by consumer and provider.
This requires that both consumer and provider use the same security provider or that
security providers are federated. Security in this situation should incorporate some
form of hashing algorithm.
If you want your services to experience hijacking, the best way to accomplish this is by
having developers be clever (being sarcastic here) in creating their own authentication
schemes such as inspecting headers and comparing IP addresses to a list of known
hosts.
To learn more about security, we recommend Ferguson and Schneier (2003).
Enterprises cannot realistically rewrite all of their applications, expose them all as
services, incorporate all the events, and change them to exactly match all the business
processes in a timely cost-effective manner so prioritization becomes crucial. The
authors have constructed a simple straightforward approach to prioritization and
have put them into four simple classications:
1. Return on Investment (Bang for the buck),
2. Canonical Form,
353
3. Integration, and
4. Retirement.
9.1. Return on Investment

The rst and foremost rule of an enterprise SOA is that it must provide business
value. Value can be calculated in terms of making the cost of individual projects
cheaper (internal rate of return) used to provide connectivity to external parties and
to allow them access to internal applications or enabling a 360-degree view of the
customer from a supply chain perspective. Regardless of the driving factor for SOA
adoption, all decisions should be prioritized based on a stated objective of achieving
a return on investment.
The ability to realize increased return on investment usually starts by modeling
interactions from the customer perspective. For example, when an order for widgets is
received via a Customer Relationship Management (CRM) application, this event can
trigger other events such as payment systems, supply chain applications, inventory,
resource scheduling and the general ledger in a loosely coupled manner allowing the
enterprise to have an up-to-the-minute understanding from the customer perspective.
9.2. Canonical Form

The word canonical historically meant according to religious law but when applied
to computing simply indicates a standard way of writing a formula. For example, the
below HTML snippets are said to be equivalent because they mean the same thing,
but could be interpreted differently depending upon who is consuming.
<color>white</color>
Code 2. HTML Example One
<color>rgb(255,255,255)</color>
Code 3. HMTL Example Two
The need to have message exchanges (especially XML-based) not only subscribe to
syntactic standards but a specication also describing the physical representation,
the canonical form is required. The requirement for canonical form aids not only
354
in interoperability but is a fundamental component to many XML-based security

schemes.
Enterprises that adopt a common and ideally industry-dened data model for
business events which allows them to isolate their systems not only from each other
but also ensures future interoperability. An enterprise having one common model
allows change to occur in one system and the only change required is the particular
interaction. Wherever appropriate, enterprises should not only use industry standard
models, but should actively participate in their creation.
Enterprises wherever possible should not only use industry standard models where
appropriate, but should actively participate in their creation.
If the canonical data model is dened in XML, the ability to convert between
canonical form and internal implementations can be realized via simple XML
transformations. The transformation can additionally be re-used in other parts of the
enterprise allowing for additional interoperability.
To learn more about Canonical XML, visit http://www.w3.org/TR/xml-c14n.
9.3. Integration
Using an event-driven approach allows one to arbitrarily add additional applications
into the supply chain without changing existing applications. New applications express
an interest in specic enterprise-level messages and handle responses accordingly. This
allows for the enterprise to add new applications at will and deal with previously
unknown situations. Customizations to existing applications can also be accomplished
in a loosely coupled manner.
9.4. Retirement
Taking event-driven approaches one step further will allow for the wholesale replacement of existing legacy applications with new ones. Minimally, when a consuming
application doesn?t need to have understanding of the producing application will
allow for its eventual replacement. Many enterprises desire to replace their CRM
investments but are in a conundrum since upgrading costs are obscene. In this
situation, the enterprise can replace the front-end of the CRM application using
355
standards-based approaches such as portals and write a services wrapper on top of the
CRM application to isolate changes in data.
The enterprise using this approach will rst replace the front-end of the CRM
application, working its way down the layers until it is able to wholesale replace the
persistence tier. Event-driven architecture should be incorporated into any retirement
strategy.
10. Summary
Event-driven architecture and service-oriented are distinct yet compatible concepts.
The agile enterprise will require both. In constructing an enterprise service-oriented
architecture, the enterprise will need to sense and respond to events as they occur
rather than carry out predetermined processes developed using information based on
historical occurrences.
In this chapter, we have learned how Service-Oriented Architectures provide the
potential to improve the efciency of IT systems and service offerings. While this
book covered the technical know-how of how SOA can be realized, it is simply not
enough to guarantee success. In order to be successful realizing an enterprise SOA
one needs to be skilled in management and enterprise architecture. Enterprise SOA
needs to be backed with the proper organizational structure, strong governance and
agile software development approaches.
To learn more about governance, visit the IT Governance Institute at:
http://www.itgi.org.
Enterprises that expect to use technology to gain and retain competitive advantage in
the marketplace must adopt a business-driven event-based service-oriented architecture. Enterprises that have already started down the path have a distinct advantage
over their peers. Enterprises that are just starting the journey to implementing
an Enterprise SOA are well positioned to be fast followers. By learning from the
experiences of others, their successes and failures, enterprises will be able to build
the extended enterprise realizing the holy grail of lower total cost of ownership and
business agility.
OUTTRO
How few there are who have courage enough to own
their faults or resolution enough to mend them.
Benjamin Franklin
One of the most challenging aspects of writing a book on a vast topic like serviceoriented architectures is knowing when the book should be published. Within
the minds of most authors, books are never really complete. Authors are doubly
challenged with the fact that they too may be consumers of books such as the one
you hold in your hand. We too get upset when we spend our hard earned money and
a book does not 100% answer all of our questions.
One of the challenges we faced while creating this book is whether it should cover
topics of the moment that corporations are currently struggling with such as whether
I should use SOAP internal to my enterprise. We took the stance, right or wrong that
questions such as these are more appropriate for magazine articles and discussions
with industry analyst rms. We felt that a book should not cover the issues of the
minute but instead present information that is timeless in nature.
The authors also wanted to cover other aspects of service-oriented architectures but
would have never nished the book if we kept adding to it. In order to be fair to both
parties, we have decided to take a different approach. For purchasers of this book
who register, we will be sending supplemental information in ebook format for no
additional cost in hopes that we can not only meet your unstated desires in spending
large sums of money to acquire knowledge, but to exceed it by leaps and bounds.
The ebook will discuss the following topics that are near and dear to your heart and
criteria for success:
Developing SOAs using the Enterprise Service Bus patterns,
Integrating SOAs with Business Rules Engines,
357
358
Extending Event-Driven Architecture to develop fault-tolerant SOA implementations,

Case study on how SOA was used at a Fortune 100 enterprise in the insurance
vertical to serve business partners and consumers, and
A catalog of SOA architecture and design patterns.
In the meantime, if you have questions you would love answered, do
not hesitate to ask. The author team has established a Yahoo Group at
http://groups.yahoo.com/group/soabook for this purpose. Feel free to join and
encourage others to participate
APPENDIX A:
UNDERSTANDING
DISTRIBUTED COMPUTING
Man has such a predilection for systems and abstract deductions that he is ready to distort
the truth intentionally, he is ready to deny the evidence of his senses only to justify his logic
Fyodor Dostoevsky
The idea of service-oriented architectures have been around for over a decade and have
been incorporated into many of the technologies that are pervasively used including
the worldwide web, email and FTP. These concepts when applied to business, allow
a new generation of enterprise applications to be created. SOA addresses many of the
shortcomings in previous old world architectures based on monolithic independent
applications.
Enterprises that leverage service-oriented architectures can bridge the traditional gap
existing between business requirements and IT capabilities by providing the ability
to nd, bind and execute to providers of choice. SOA can help the enterprise with
business goals that are based on technology in many ways including, but not limited
to:
Faster time to market,
Less expensive application integration,
Easier software construction,
Ability to leverage existing IT investments, and
And more
359
360
The rapid adoption of SOA is leading a shift in how enterprises leverage computing
resources and deliver services to users. It is helping to provide tighter integration and
alignment between business drivers and IT implementation. The ability unlock the
assets contained with enterprise applications and provide the capability to mix and
match services regardless of whether they reside within the enterprise or are provided
by third-parties will allow IT to increase their own value to the enterprise.
Let us start with understanding the technical benets of SOA and how it is based on
the fundamentals of distributed computing.
1. Distributed Computing
Distributed computing is one key component in enabling the creation of the extended
enterprise.The ability to have humans and systems interact with each other regardless
if they reside within the same company or are located on the other side of the planet
connected via the Internet has been a reality for many years now. The ability to
interconnect systems via network using thousands of personal computers and various
forms of networking technology has been commoditized.
Distributed computing is the ability to break down an application into discrete
computing components that can be executed across a network of computers so they
can work together to perform cooperative tasks. There are many approaches that can
be used to accomplish this goal.
In this chapter, you will learn about the foundational principles used to distribute
application execution across a network including:
The anatomy of a distributed application.
Remote Procedure Calls.
Object Request Brokers.
Transaction Monitors.
Message-Oriented Middleware.
Appendix A: Understanding Distributed Computing
361
1.1. Anatomy of a Distributed Application

Distributed applications are constructed using layered approaches whereby each layer
has a particular function. At the lowest level, the network layer provides connectivity
to a group of computer so they can exchange information in a seamless manner. Use
of pervasive industry standard protocols such as TCP/IP at the network layer provides
packaging and addressing support for communicating to other applications that use
the same network protocol. Higher-level services can be layered on top of the chosen
network protocol, such as services that support security and directory services. By
further extending this layer, a distributed application can be created that leverages
lower layers to perform coordinated tasks across the network.
Figure 1. Distributed Application

Interoperability is increased when all of the layers use the same protocol. In a general
sense, protocols dene the sets of rules governing communication between nodes
within a network. Protocols may specify timing, format, sequencing, error control,
reliability, routing and security.
Each layer in a distributed application will specify its own set of rules for operation
and data ow and therefore creates its own protocol.
Let us begin by taking a closer look at the foundation of a distributed application
the network layer.
362
1.1.1. Understanding the Network Layer

Fundamentals indicate that a computer network is simply a series of connections
between computers and other network devices that allow them to communicate
with each other. The size, speed, reliability and usage depend on the protocols and
implementations used. Without protocol, networks would not successfully allow
communication.
TCP/IP, one of the most widely used of protocols, actually is dened as a family
of protocols in which TCP and IP are two. Protocols at the network layer were
intentionally designed in a layered fashion and avoided monolithic approaches
[typically seen at other layers. The original inventors of the TCP/IP protocol
separated the suite of protocols into a discrete set of tasks with each layer that
corresponds to a different facet of communication.
While implementations may blur the lines of actual layers to increase performance,
the notion of layers at the TCP/IP are inherent within its architecture and is typically
represented pictorially as a stack, as shown in Figure 2.
Figure 2. TCP/IP Protocol Layers

The link layer is responsible for communicating with actual network hardware such
as Ethernet cards. This layer transmits and receives data from the network and
passes it along to the network layer. Likewise data received from the network layer is
transmitted over the network by the link layer.
The network layer has responsibility for routing data to its destination. This layer
is stateless and does not guarantee that data will reach its destination. Its sole
363
responsibility is to decide where data should be sent. Because this layer is not
concerned whether packets get to their nal destination or the order in which packets
arrive, its job is greatly simplied. If a packet arrives corrupted, the network layer
simply discards it. The network layer requires that every network interface have a
uniquely dened address. If the network layer is based on TCP/IP the uniquely
dened address is known as the IP address.
The transport layer provides data ow (state) for the application layer where required.
This layer performs guarantees of reliable delivery and ordering. Two of the more
popular protocols in use at this layer are Transmission Control Protocol (TCP) and
User Datagram Protocol (UDP). TCP provides end-to-end reliable communication.
TCP creates virtual circuits between two processes and ensures that packets are received
in the order they are sent (sequencing) and that lost packets are re-transmitted.
The application layer is where user applications interact with the network. Protocols
such as Telnet, FTP, Email, IRC and others interoperate. Applications can use
either/both TCP or UDP to communicate with other computers.
Let us look at a basic scenario on how TCP/IP is used within a Telnet application:
1. A user desires to connect to a machine via telnet named foxbat.soa.edu.
2. The machine name is converted via Domain Name Server (DNS) to an IP
address (192.168.0.6).
3. Telnet informs the transport layer it wants to start a TCP connection with
192.168.0.6 on port 23.
4. TCP establishes a conversation with foxbat and uses IP to route packets.
5. Telnet retrieves a port number from the users machine (say, 1742) and TCP
places the source and destination ports in its packet header.
6. The packet is now handed to the network layer where IP routes the packet to the
link layer.
7. The link layer takes the packet and places it on the network where it is transmitted
via routers and switches to the Internet.
8. The process is repeated, one router at a time until it reaches the network segment
in which foxbat is located on.
9. foxbats TCP layer replies in a similar manner as outlined in step 8.
10. The telnet daemon on foxbat and the telnet client exchange terminal information
and other parameters required for establishing an interactive session.
364
11. Control messages are sent in-band as an escape byte of 255 followed by a control
byte. Control messages include: echo, status, terminal type, terminal speed, ow
control, linemode and environment variables.
12. The user sees the login prompt from foxbat. After the login process is completed,
data is sent back and forth until the session is terminated.
The benet of layers, as originally envisioned by the inventors of TCP/IP, is that the
network and transport layer would only have to be written once for each protocol.
They would then provide a common interface to the network layer by writing
different device drivers for each kind of network interface.
The principle of layering is pervasive in the vast majority of widely adopted
architectures and adds simplicity. For TCP/IP, the transport layer provides a standard
interface; network services that leverage the interface do not need to be rewritten or
even recompiled if the transport layer code changes.
For more information on TCP/IP, see ftp://ftp.internic.net/rfc.
1.1.2. Building the Application Layer

In order to create a exible component-based architecture that provides the ability
to respond to increasingly complex business drivers and offer support to a diverse
set of applications and service providers, use of a layered approach to constructing
the application layer is warranted. Distributed applications as a core principle require
logical partitioning of an application into at least three layers. A good approach to
layering includes the separation of presentation, from business logic from data access
and storage through controlled and managed interfaces.
In a simple distributed application, a client that provides the user interface communicates with a business logic layer. This layer may leverage an application server and
custom-developed code that in turn communicates with either a database and/or le
system that is responsible for storing application data.
Let us look at the function of each of these layers.
Presentation
The presentation layer includes the user interface to an application and can use
either a rich or thin-client approach. Rich client interfaces provide full access to
the underlying operating system user interface components. Thin clients leverage
markup languages such as HTML and XML and provide benets of portability and
looser coupling at the expense of user interface expressiveness.
365
Figure 3. Application Layer

This layer may be implemented using the Model-view-controller architecture pattern
to separate the business model (data and business logic), view (presentation code) and
controller (response to user actions). If it uses a markup language, this layer can also
provide additional inter-tier interaction. Examples of this may include embedding
XML into HTML web pages.
One form of thin-client, web browsers not only understands loosely structured
HTML but can also support XHTML, which is a well-formed version of HTML.
Clients can receive XML and use XSLT to render XML into other markup formats
including well-formed HTML. This approach provides separation of presentation
from data. Alternative approaches include using of Cascading Style Sheets (CSS) that
can be used by the thin-clients to present XML-based documents.
Business Logic/Services
The business logic layer incorporates your applications business logic and provides a
well-dened interface to the presentation layer services. Business logic within many
enterprises is typically hosted within an application server that allows the logic
to support multiple simultaneous clients. The application servers themselves may
also provide support for pooling of business logic in order to increase efciency at
run-time as well as protect selected portions of business logic from those who are not
authorized to access specied processes.
In UML terms, the Actor can be a user interface (web browser, telephone, etc) or
another system. The actor will issue a service request, which is dened as an event.
This event is routed to the responsible business process component for execution
366
Figure 4. Business Logic/Services

and the results will be returned. The services tier is responsible for exposing business
logic to client applications and hosts both business and technical services. Services
are comprised of multiple components (hence component-based services) that expose
common interfaces that are accessible via multiple clients.
Data Access and Storage

Data access and storage are usually part of the resource tier that mediates access to
back-end systems of record that may include vendor packages such as Peoplesoft
and Siebel, relational databases such as Oracle and legacy mainframe applications.
Resources are exposed to various clients via the service tier.
Many relational databases provide the ability to take an XML document and
persist it to normalized relational structures (shredding). This allows an application
to work with self-describing well-formed XML documents while traditional tools
that understand relational structure can also access the data directly increasing
interoperability.
1.1.3. Operating System Components

An operating system provides its own set of services to applications and services that
run on top of it. An operating system is responsible for providing scheduling and
protection of multiple user processes ensuring equitable access to resources under its
management.
Modern operating systems use a componentized approach in their own construction
and operation. By understanding how operating systems work at a fundamental level,
367
one can understand how applications and services that will execute on top of them
can take advantage of the layered approach. Let us look at these three components.
Processes
A process is an instance of an application that is capable of running on one or more
CPUs and contains a sequence of steps that the computer executes. A process also
provides context that controls access to resources of the computer (CPU time, I/O,
etc) via operating system calls. A process can initiate one or more sub-processes,
which is called a child process. The child will refer to the process that initiated it
as its parent. A child process is a replica of the parent process and shares many of
its resources and is terminated when its parent is terminated. An application can be
made up of one ore more processes. Likewise, a process can contain one or more
applications.
Threads
A process is comprised of at least one thread of execution. All modern operating
systems support the creation for multiple threads of execution within a single process.
Each thread within a process can run independent of other threads but in practice
usually require some form of synchronization amongst them.
Network-aware server applications are almost always developed as multi-threaded
applications whereby a group of threads may be dedicated to monitoring input from
a socket connection (users who are attempting to connect) while another group of
threads may be dedicated to actual processing of logic. Synchronization is required
when threads need to coordinate the transfer of data from the business logic portion
of the application and send a response back to the requester.
When developing a multithreaded application, the number of simultaneously running
threads can only be one for one with the actual number of CPUs. When there are
more threads that desire to execute, the operating system will perform a context
switch that allows other threads the opportunity to execute. Context switching is a
core element of operating systems and occurs at both the process and thread levels.
Context switching between two threads in a single process is lighter weight than a
context switch between two processes.
Objects
Applications that were created using modern object-oriented languages are comprised
of cooperating objects.A process is composed of one or more objects and one or more
threads within a process access these objects. Objects themselves can be distributed
across multiple systems within multiple processes using technologies such as CORBA,
DCOM and RMI.
368
Common Object Request Broker Architecture (CORBA) is an architecture and

specication for creating; distributing and managing distributed program objects
within a network. CORBA allows applications at different locations to communicate
through an interface broker. Distributed Component Object Model (DCOM) is a
set of Microsoft concepts and program interfaces that provide underlying services
of interface negotiation, lifecycle management, licensing and event services. Remote
Method Invocation (RMI) is used in conjunction with applications written in Java
and uses an object parameter-passing mechanism known as object serialization. RMI
is designed to preserve the object model across a network.
For more information on CORBA, visit http://www.omg.org. For more information
on RMI, visit http://java.sun.com/products/jdk/rmi/. For more information on
COM, visit http://www.microsoft.com/com/dcom/dcom95/dcom1_3.asp.
1.2. Interprocess Communication

Operating systems execute processes within their own virtual address space and
provide isolation from other processes. By default, processes cannot communicate
with each other unless they use services provided by the operating system to do so.
There are many times when interprocess communication is warranted.
Interprocess communication (IPC) is a set of interfaces that allow developers to
create and manage individual application processes that can execute concurrently.
IPC allows an application to handle many user requests simultaneously. Furthermore,
a request from a single user may result in multiple processes executing within the
operating system on the users behalf, the processes need to communicate with each
other.
Applications may use one or more IPC approaches including, but not limited to:
Signals,
Pipes and Named Pipes,
Semaphores,
Shared Memory, and
Sockets.
Let us explore the details of each IPC mechanism.
Signals
Signals are used to signal asynchronous events between processes. A process may
369
implement a signal handler to execute when an event occurs or may use the system
default actions. Most signals received by an application can be ignored but some are
required to be implemented. For example, within Java when a user wants to get a
thread dump they can execute a kill 3 pid where pid is the operating system process
ID. Likewise, to terminate an application, you would execute kill 9 pid that would
terminate the process.
Pipes and Named Pipes

A pipe provides connectivity of the standard output (STDOUT) from one process
to the standard input (STDIN) of another process. Pipes provide a method for
one-way communication between processes in a parent-child relationship and may
be sometimes referred to as half-duplex pipes.
For two-way communications between processes, two pipes can be established, one
in each direction. When pipes are used for interprocess communication, they must
have a common parent process. In both, one-way and two-way communications, the
operating system buffers the data sent to the pipe until the receiving process reads it.
Pipes themselves are created with a xed size and cannot grow the amount of data
they hold.
Named pipes work similarly to standard pipes but are implemented as a virtual
device as part of the le system and use a First-in First Out (FIFO) approach. Unlike
standard pipes, named pipes do not require processes to work in a parent-child
relationship in order to communicate.
Semaphores
Semaphores are counters that are used to control access to shared resources accessed
by multiple concurrent processes. Semaphores are frequently used as a locking
mechanism to prevent processes from accessing a specied set of resources while
another process is performing operations on them. Semaphores are a special function
of the operating system kernel that each process can check and usually implemented
as sets. Semaphores are common used to either share a common memory space and/or
to share access to les.
Shared Memory
Shared memory provides the ability to map an area of memory into the address
of more than one process. Shared memory is usually the best performing method
for interprocess communication as processes do not need access to operating system
kernel resources to share data. For example, a client process may send data to a server
process in which would be modied and returned back to the client. Using other IPC
mechanisms would require the client to write a le and the server to read it requiring
kernel services. By using shared memory, the client would put data in shared memory
370
after checking a semaphore value, writes the data to the shared memory area, and
then resets the semaphore to signal the server that data has changed. The same process
would occur with the server notifying the client of changes in data.
Sockets
Sockets provide two-way (full duplex) method for interprocess communication
between client and server processes in a network. Sockets can also be used to
communicate between processes on the same system. A socket is an endpoint of
communication to which a name can be bound. Sockets can either be stream-based
or datagram-based. Stream-based sockets provide guaranteed delivery and ensure
sequenced unduplicated receipt of packets. Datagram sockets do not guarantee
delivery or sequence and may allow for duplicated packets to exist but are usually
faster at the network layer.
1.3. Communications Infrastructure

The infrastructure that enables processes to seamlessly communicate between each
other over a network is typically referred to as middleware. Middleware is a layer of
software that resides on the network and mediates information exchange between
two different processes. Middleware helps applications work together across disparate
operating systems and platforms and allow developers to write applications that
interface with the middleware layer rather than recreating for each application the
services provided by the middleware.
The ability to communicate across processes can be accomplished in a variety of
manners including:
Remote Procedure Calls,
Object Request Brokers,
Transaction Monitors, and
Message-Oriented Middleware.
When used within the enterprise, middleware helps applications ranging from
client/server to legacy mainframe applications to Web environments to participate in
solving business goals in a distributed manner. Let us take a closer look at the various
methods for communication in a distributed environment.
371
1.4. Remote Procedure Calls (RPC)

A procedure is a sequence of instructions that execute as part of an application.
A procedure call is a request that is made by one procedure to another procedure
for the specied service. Remote procedure calls reduce complexity in creation of
applications that execute across multiple operating systems and network protocols
(TCP/IP, IPX/SPX, etc.) by hiding the details of transport and marshalling from
developers.
Remote Procedure Calls cannot be considered middleware in a strict sense in that
it is embedded within actual applications. When client and server applications are
compiled, a local stub is created and linked into the application. These stubs are
invoked whenever the application requires access to a remote function.
RPC is typically implemented using synchronous request/reply exchange in which the
client blocks execution until the server satises the result. RPC can be implemented
to support asynchronous behavior through the use of a client-side proxy that either
executes in another thread or implements polling/yielding behavior.
For more information
wiki?ProxyPattern.
on
the
Proxy
pattern,
see
http://c2.com/cgi/
RPC can be used in client/server applications where the client can issue a service
request and wait for the response from the server before continuing its own processing
logic (blocking). Most implementations of RPC are not well suited to handle peer-topeer, asynchronous or object-oriented programming and should be avoided in these
situations.
1.5. Object Request Brokers (ORB)

Object request brokers are traditionally used in CORBA-based architectures and serve
as a broker between a clients request for a service and the service itself. Object request
brokers provide location transparency as the client can make service requests without
having to know where the service is provided in a distributed network or what the
actual interface needs to look like in order to carry out a successful invocation. The
ability to understand interface denitions at run-time allows for run-time resolution.
The function provided by ORB technology can be loosely categorized into:
Interface denition,
Location and possible activation of remote objects, and
372
Communication between clients and object.

Object request broker are analogous to a telephone network. ORB provides a
directory of services and mediate communications between connected clients and
services offered. For architectures that use an ORB-based approach, to the client all
services appear as if they are local. The ORB in a sense is a framework that provides
the cross-system communication between distributed objects.
Figure 5. Object Request Broker

Depending upon the particular implementation of the ORB, it may be compiled
into clients, executed as separate operating system processes or can even be embedded
within an operating system kernel. Object request brokers may provide several services
to both client and service provider including those shown in Table 1.
There are two major offerings of ORB-based approaches, one of which is championed
by the Object Management Group (OMG) and their CORBA specication and one
led by Microsoft known as the Component Object Model (COM).
Enterprises that adopt an ORB-based approach have to take into consideration
vendor lock in, lack of pervasive interoperability and extensive software development
and vendor acquisition costs. Usage of these approaches is good for integrating legacy
technologies but is highly discouraged for modern software development.
Product offerings by vendors who produce object request brokers include:
Iona Orbix http://www.iona.com.
Rogue Wave Noveau http://www.roguewave.com/products/nouveau.
373
Table 1. Services Provided by Object Request Brokers

Services
Lifecycle
Description
Denes how objects and components are created and destroyed.
Persistence
Provides the ability to store data used by the service using a backing
store such as a relational database and/or le system.
Allows a component to locate another component by name and
can optionally be used to leverage existing naming services such as
NIS or DCE.
Components can register interest in receiving notication of
selected events.
Naming
Events
Concurrency control
Provides management of locks to data that transactions may

compete for.
Transaction
Ensures that when a transaction is complemented, all changes are

either committed or restored to their pre-transaction state.
Supports measurement for purposes of compensation and may
provide mechanisms to restrict usage based on session, instance
creation or physical location.
Allows a component to provide a self-description that can be
consumed by other components.
Licensing
Properties
Security
Time
Provides the ability for the service to authenticate and authorize

all client related service requests.
Provides the ability to keep two components to share the notion
of time.
ObjectSpace Voyager http://www.objectspace.com/products/vgrOverview.htm.
1.6. Transaction Processing Monitors

Transaction processing monitors (TP Monitors) have been successfully used in order
processing, airline reservation, customer service applications and scenarios where
an online support environment has to concurrently operate in conjunction with
batch processes. TP Monitors provide a strong ability to increase scalability of a
distributed architecture and accomplishes this goal by multiplexing client transaction
requests onto a predened number of processing instances that support desired service
requests.
Transaction Processing Monitors provides services that increase scalability, availability
and management capability such as restarting failed processes, enforcement of
consistent in distributed data stores and dynamic load balancing. TP monitors
provide near linear scalability by adding more servers to service additional requests.
374
Figure 6. Transaction Processing Monitor

Typical TP monitor implementations have built-in support for authentication and
authorization of the clients attempting to use the services along with protecting the
data they attempt to access. TP monitors typically support both synchronous and
asynchronous communication models.
TP monitors break up complex applications into discrete small units of work
(transactions) and provide a guaranteed commit/rollback mechanism. A typical unit
of work may require performing multiple database actions (inserts, updates, deletes,
etc.) across multiple databases yet ensuring a consistent state.
TP monitors offer the following functionality:
Communication services Support for synchronous, asynchronous, storeand-forward (similar to MOM) and conversational mechanisms (similar to
RPC).
Directory services Physical data location independence to allow for crossplatform operations.
High availability Provide for routing user requests upon node failures as well
as re-initialization and restart of failed services.
Workload management In environments where workloads are comprised of
online and batch transactions, TP monitors provide scheduling capabilities that
375
take into consideration, desired policies so that execution maintains a consistent

throughput.
Enhanced security TP monitors provide security context for authorized users
including their specic privileges and can optionally check against external
security credential providers such as LDAP and relational databases.
Online administration TP monitor typically provide a console whereby an
administrator can view all operations in progress, start, restart and shutdown
services and gather statistical information on the platform.
Vendor independence TP monitors have been implemented pervasively and
support multiple operating system platforms. Implementation of the X/Open
consortiums XA (transaction) interface when communicating with resource
managers is also standards-based.
Legacy System Access Use of TP monitors provide transactional connectivity
to legacy technologies such as IMS, RDB and other legacy technologies.
The architectural framework provided by TP monitors allows enterprises to develop
scalable applications in a heterogeneous environment with clients ranging from a few
to thousands in a relatively simplistic manner.
Product offerings by vendors who produce transaction-processing monitors include:
IBM CICS http://www.ibm.com.
NCR Top End http://www.ncr.com.
BEA Tuxedo http://ww.bea.com.
1.7. Message-Oriented Middleware

(MOM)
Message-oriented middleware is software that supports asynchronous calls between
client and server applications and use message queues to provide temporary storage
when the destination server is unavailable. Message-oriented middleware works in
a services network in client/server, peer to peer and publish and subscribe. Message
exchange mechanisms allow for logical decoupling in that sender does not have to
know what platform the destination application uses.
Message-oriented middleware is best utilized for event-driven applications especially
when developed using object-oriented paradigms. The asynchronous nature of
376
message-oriented middleware unlike remote procedure calls (RPC) provides the

ability for the client to not block when making service requests. This can in many
circumstances provide for increased scalability (but reduced performance). Messageoriented middleware can have the ability to in stress situations overload the network
when clients can send data faster than the server can process the incoming requests.
The vast majority of implementations of message-oriented middleware are implemented in a proprietary manner and do not interoperate with other implementations
of MOM products without costly bridging products. Use of a single implementation of
a MOM will likely result in a dependence on the MOM vendor future enhancements.
Vendor lock in can be mitigated in some languages such as Java by coding to message
product neutral APIs such as JMS.
MOM software usually requires installation on every platform within a network that
will increase software-licensing costs. Administration of MOM software in a large
distributed network can result in an increase in staff for maintenance especially in a
heterogeneous environment. MOM may also consume additional CPU cycles and
memory on each platform and must be taken into consideration at capacity planning
time.
Product offerings by vendors who produce message-oriented middleware include:
Oracle Advanced Queueing http://www.oracle.com.
Arjuna Messaging http://www.arjuna.com.
IBM MQSeries http://www.ibm.com.
Tibco Rendezvous http://www.tibco.com.
Open Source Message Queue http://www.osmq.org.
1.8. Service Description

In order to achieve interoperability in a distributed computing environment, the
architecture requires not only communication at the lower layers of the protocol stack
but must have equivalent support at the application layer. Supporting interoperability
at the application-layer requires the ability to have a common way of describing
services. Web Services Description Language (WSDL) is an XML markup that
provides an XML grammar for describing services as collections of communication
endpoints capable of message exchange.
377
Description mechanisms will usually contain an abstract denition of endpoints and

messages produced and consumed by the specied service. The service description
should whenever appropriate separate the abstract denition of endpoints and
messages from the concrete network deployment or data format bindings.
Mechanisms that support service description have existed in distributed computing
environment for decades in the form of Interface Denition Language (IDL). IDL
denes a system of interfaces, operations those interfaces support, data types used as
parameters and return values for those operations.
1.9. Versioning
In the real world, an enterprise will create different versions of its software and services
on a periodic basis. A service over time will increase the functionality provided to
consumers. Supporting multiple versions of a service is different than the traditional
methods used in support of multiple versions of a product due to loose coupling
between layers and the separation of producer from consumer. It becomes important
to support the formal evolution of services, which can be accomplished via a variety
of mechanism including, but not limited to:
Making interfaces immutable and dening a new interface for each change in
service.
Creating a document-style interface that allows a document to be passed to the
service which contains versioning information.
Allowing interfaces to be mutable and requiring clients to stay synchronized
with changes.
Making interfaces immutable is a technique used by Microsofts Component Object
Model (COM) and helps prevent run-time binding errors by ensuring that clients
always have access to a stable interface. Interface immutability will allow a client to
bind to a specic version of an interface or if one is not specied, the client will be
bound to the most current version.
Document passing is another technique where either a serialized string or value object
is passed from the client to the service. The service itself takes on the responsibility
for checking within the document, the actual version the client passed and making
sure it is supported and syntactically valid. If it is not valid, the service itself will
throw a fault. Defensive coding will usually require making a copy of the document
before processing.
378
If you control both the producer and consumer of services then you can change
interfaces at will and upgrade accordingly. This strategy however is fraught with issues
if the service will be consumed outside of your administrative control.
Minimal support for versioning of XML-based services should adopt an approach
that leverages unique XML namespace URIs.
Techniques for each of the three choices have been discussed in previous chapters.
1.10. Operations
Distributed computing platforms usually provide a mechanism for describing network
services as collections of communication endpoints that support message exchange
along with an abstract description of actions supported by the service.
Web services for example use the Web Services Description Language (WSDL) to
describe operations across multiple services and capture this information using XML
in a WSDL document. WSDL dened services as a collection of network endpoints
and separates the abstract denition of endpoint and messages from their concrete
deployment and data format.
Service operations can occur using both RPC-style and message-passing (document
style) bindings. When operations use RPC-style, it becomes important to mirror the
underlying interfaces signature since order of parameters becomes signicant. For
document style, list of parameters do not require the sender to adhere to a particular
parameter order.
Typically, transmission of messages across a distributed computing environment can
be classied in to one of four categories:
1. One-Way,
2. Request-response,
3. Solicit-response, or
4. Notication.
The above four primitives are better known as primitives and are represented in
abstract terms. Understanding the basics of each of the four primitives is crucial to
services. For example, some endpoints may only received messages if they are the
result of a synchronous request/response operation while others can interoperate in
an asynchronous manner. Let us look in detail as to how each of these operations can
be dened using WSDL.
379
1.10.1. One-Way
A one-way operation sends data to an endpoint but does not expect any form of
response. This form of operation is typically seen in message-queue-based architectures
that allow clients to send and forget. Example grammar for a one-way operation is
shown below.
<wsdl:definitions ... >
<wsdl:portType ... > *
<wsdl:operation name="nmtoken">
<wsdl:input name="nmtoken"? message="qname"/>
</wsdl:operation>
</wsdl:portType >
</wsdl:definitions>
Code 1. One-way operation
In the above example, the portType element species a set of abstract operations. The
input element species the abstract message format for the one-way operation. The
name attribute of the input element denes a unique name for the operation. If it is
not specied, the name defaults to the name of the operation with the operation type
appended.
1.10.2. Request/Response
A request-response operation sends data to an endpoint and expects either a successful
response or a fault to be returned. This form of operation is typically seen in scenarios
where interaction between two services is synchronous in nature. Example grammar
for a request-response operation is shown below.
<wsdl:operation name="nmtoken" parameterOrder="nmtokens">
<wsdl:output name="nmtoken"? message="qname"/>
<wsdl:fault name="nmtoken" message="qname"/>*
</wsdl:operation>
</wsdl:portType >
</wsdl:definitions>
Code 2. Request/Response Operation
380
The input and output elements specify the abstract message format for the request
and response, respectively. The fault element is optional and specied the abstract
message format for any errors that may be returned by the service as a result of the
operation. Each element within the fault must be named to allow a binding to specify
the concrete format of the fault message. The name of the fault element is unique
within the set of faults dened for an operation.
The request/response operation must specify a particular binding to determine how
messages are actually sent within a single service conversation (such as HTTP
request/response).
1.10.3. Solicit/Response
A solicit/response operation receives data from an endpoint and returns a successful
response or a fault. This form of operation is similar to request/response in that it is
synchronous in nature. Example grammar for a solicit/response operation is shown
below.
<wsdl:operation name="nmtoken" parameterOrder="nmtokens">
<wsdl:fault name="nmtoken" message="qname"/>*
</wsdl:operation>
</wsdl:portType >
</wsdl:definitions>
Code 3. Solicit/Response Operation
The output and input elements specify the abstract message format for the solicited
request and response, respectively. The optional fault element species the abstract
message format for any errors that may be returned from the service.
1.10.4. Notication
A notication operation receives data from an endpoint but does not send any form
of acknowledgement. This form of operation is typically used to implement callback
mechanisms for asynchronous processing. Example grammar for a notication
operation is shown below.
381

<wsdl:operation name="nmtoken">
</wsdl:operation>
</wsdl:portType >
</wsdl:definitions>
Code 4. Notication Operation
The notication operation is the inverse of one-way. The output element species
the abstract message format for the notication operation.
1.11. Service Discovery

Services are not only required to be described in a consistent manner to ensure
interoperability, but must support a mechanism that allows them to be discovered.
The discovery of services within a distributed computing environment can occur
using many approaches including passing conguration les to locating the service via
a specialized directory service by name and/or specic attributes. In the past, directory
services have supported browsing, drill-down and search engine type operations.
Applications and services can be built upon application infrastructure that operates
in groups of interdependent servers usually known as clusters. Clusters coordinate
their actions to provide increased scalability, availability and fault-tolerant services to
applications.
Clustering architectures are classied based on how each member server within a
cluster accesses memory, disk and whether servers share a copy of the operating and
the I/O system. Clusters can be loosely dened into three categories.
1. Shared Memory,
2. Shared Disk, and
3. Shared Nothing.
Clusters that use the shared memory model have member servers within the cluster
use the same primary memory, through which all trafc to the cluster is routed.
The servers in the cluster also share a single copy of the operating system and I/O
subsystems. Shared disk clusters provide each member server with its own memory
but share common disks. In order to ensure data protection against simultaneous
write operations, use of a distributed lock manager to control access to disk-based
resources is required.
382
Shared nothing clusters provide each member server with its own memory and disks
and use disk mirroring technologies to provide access to shared state. Hybrid models
also exist where shared nothing architectures are ran on top of shared disk hardware.
While only a single member server can access the disk at a time, it provides seamless
failover in that another server can pickup the workload via checkpoint processes.
Cluster architectures have common attributes in how they operate and can be
generally classied into four categories:
1. Active/Active,
2. Failover/Failback,
3. Switchover, and
4. Impersonation.
Active/active clusters require each server to run its own workload and can assume
responsibility for another cluster member in the event of a failure. Failover/failbackbased cluster will automatically transfer the workload of a failed server to another
cluster member until the rst server recovers, at which time its workload is
automatically transferred back.
Clusters may also use switchover approaches that allow clients to nd replacements
for failed cluster member servers with a minimum of service disruption. Usually
this approach requires a server to stand in for another server by taking over the IP
address of a failed server. Impersonation allows clients to nd a replacement server
and reroutes network trafc intended for the failed server to the replacement server.
1.12. Application Services

Most distributed computing environments nowadays are built on top of application
servers that provide security, load balancing, and transactional support and mediated
access to allow for pooled operations in order to increase scalability and ensure
reliability.
Services that are built on top of application servers can be classied into four basic
models:
1. Stateless,
2. Conversational,
3. Cached, and
4. Singleton.
383
Let us look at each of the models:
1.12.1. Stateless Services

Stateless services do not maintain state in memory between service invocations.
This allows a service to have the potential for increased scalability by allowing
multiple instances of a service to be deployed across multiple servers in the services
infrastructure. If the infrastructure provides load-balancing services that detect
unresponsive services, using this model services can be made highly available.
1.12.2. Conversational Services

Conversational services maintain state between service invocations between clients
and can be constructed by either having the client on each service request pass a
session identier or by having the infrastructure create an instance that is solely used
by that client. Conversational services generally maintain transient state in memory
and may optionally persist state between long-running transactions.
Service request from clients are usually grouped into sessions. A session is associated
with the state that must be maintained by the service between service requests. A
service may choose to save shared state between invocations in order to mimic stateless
behavior.
If durability between invocations is not a requirement, there are multiple alternatives
that can improve both scalability and performance. Session state can be serialized and
passed to/from the client to the service as part of the request itself. The service can
also leave session state in memory and periodically expire session data based on lack
of activity.
When implementing conversational services, session afnity is required and may
result in changes to the underlying network since network protocols will by default
assume everything is stateless. Support for afnity is created when the session is built
and all subsequent requests are routed to the chosen server. This can be implemented
by using external IP-based mechanisms (i.e. F5 Big IP, Coyote Point, Cisco CSS and
so on) that leverage DNS to return a single address for the service and/or support
pass-through proxying in order to obtain stateful load balancing.
384
1.12.3. Cached Services

Cached services maintain state in memory and use its memory to process service
requests from multiple clients. A cached service can increase scalability within a
distributed computing environment in situations where request processing is CPU
intensive and the results are not specic to the caller. Implementations of cached
services vary in their implementation and can use alternative strategies for keeping
copies of cached data consistent with each other. Cached services may also leverage a
backing store when memory requirements are high.
One approach a cached service may implement is to have each cache ush itself at
regular intervals according to a congurable time-to-live property. Using this approach
removes any requirement for communication between the same instances of services
but does require the client to tolerate a window of potential data inconsistency and/or
staleness.
Another approach is to keep all copies of the data consistent with the backing store
by using concurrency control in the caches. In order to ensure scalability, this is
best accomplished by implementing optimistic concurrency where the service should
ush the cache after updates to reduce the possibility of concurrency exceptions.
1.12.4. Singleton Services

Singleton services are active on exactly one server within a network at a time and
handles service requests from multiple clients. Singletons are usually used in situations
where unique values may need to be generated in a sequential manner. Depending
upon the particular application server used, a singleton service may be migrated to a
new server or simply restarted upon service failure.
Potential issues arise in failure/disaster scenarios where connectivity between services
becomes disconnected. Singleton services upon migration can cause a split-brain
whereby multiple instances of a service are running but not aware of each other.
One technique that can increase availability while reducing the changes for the
split-brain scenario to occur is by partitioning data and creating a singleton service
for each partition.
We have covered many of the options you need to know about when implementing
an SOA. In the next section, we look at some of the concepts you need to consider
when putting them all together.
385
Distributed computing is difcult because it requires services to be constrained by
the real world at each service/network boundary. Whenever a service interacts with
other services and handles client requests, it also has to correctly handle a variety of
conditions that may or may not be known at code time.
A distributed computing environment can reduce the potential number of errors a
developer will face by providing layers of abstraction. For example, a service may
return ve different errors but a developer may want to abstract the details of the
error away and simply think about success or failure.
Tim Berners-Lee when inventing the worldwide web realized that broken links are
OK. The same thing can be said of developing web sites, cities or services. Successful
service-oriented architecture is always on the border of chaos. Broken services are
inherent to any complex adaptive system. The key to building an enterprise SOA are:
Get the basics right.
Prefer working software over intellectually pure pursuits of architecture.
Merciless refactoring.
Put yourself in the shoes of the city planners for New York City. If they ever sought
perfection, the city would never exist or have been a total failure. The truth of the
matter is that New York City has only had a single electric blackout, water always
ows (sometimes even overows during Super Bowl half-time), and people get ran
over by taxis, there is always construction and potholes that are in need of repair. The
strategy for creation of services cannot be thought of in a big all encompassing plan
upfront but needs to evolve over time to support the ever-changing needs of its users.
An SOA is never complete and is always in the constant state of change.
Creation of an enterprise SOA does not require perfection but does require attention
to detail, careful planning and informed opinion. The most successful services are
the ones that exist.
3. Summary
In this chapter, we have learned about:
Business rationale for service-oriented architectures.
386
The foundational principles for distributed computing.

Techniques for increasing interoperability amongst enterprise applications.
Distributed computing provides a cost effective model for enterprise level computing
where multi-tier architectures and middleware will be key factors for success of
electronic business. The rapid progress that has occurred in this area in order to assist
with seamless integration of all system components is of increasing importance.
APPENDIX B:
QUALITY ATTRIBUTES
The bitterness of poor quality lingers long after

the sweetness of meeting schedules is forgotten.
Kathleen Byle, Sandia National Laboratories
1. System Qualities
System qualities are the attributes of architecture that dene characteristics of how
the system shall behave and/or its structural aspects should be implemented. The
design of distributed computing architectures requires balance between competing
system qualities. The ve system qualities that are most important to incorporate
early as design goal are:
1. Availability,
2. Manageability,
3. Performance,
4. Scalability, and
5. Security.
Let us look at each of the ve system qualities.
1.1. Availability
Enterprise applications and services typically depend on multiple servers that may
be accessed by thousands of users and use internal networks to connect to relational
databases and legacy systems and even may leverage services outside the enterprise
387
388
connected via the Internet. This may be composed of multiple infrastructure components that must operate in a predictable uniform manner. When any component
within the enterprise applications nervous system fails, the interruption could result
in loss of business revenue, tarnish the brand of the company, and deny customer
access to data needed to perform customer self-service requests or other serious
outcomes. Creation of a highly available distributed computing architecture becomes
part of the business strategy.
Not all applications and services require 24 7 uptime with instantaneous response
times. Some can support failure with zero consequences while other applications
may be tolerant to unplanned downtime but require varying recovery approaches.
Few mission-critical applications and services must provide high availability and use
technology that supports replication to ensure instant, near real-time transparent
recovery with no signicant downtime.
Local application failures can usually be handled in an expedient manner. It becomes
progressively more difcult to avoid failure, as software will always be used in
scenarios above and beyond its original design supported. Typically, one or more of
the following reasons can cause failures:
Inadequate software testing.
Lack of a real change management discipline.
The inability to monitor the environment in real-time.
No tools that allow for performing long-term trend analysis.
Lack of quality software engineering processes.
Operating environment failure (hardware failure, disaster, re, ood, storms,
cooling).
Sudden changes in usage levels.
Based on our experience with several Fortune 100 enterprises, listed in Table 1 are
breakdown percentages of failures.
Availability is typically a measure of how often an application and/or service is
available for usage and is a percentage calculation based on how often the application
is actually available to handle service requests when compared to the total planned
run-time. The formal calculation for availability includes time for system maintenance
because an application during this period is unavailable for use.
As higher levels of availability are achieved, hardware costs for service execution
will increase due to server CPU utilization, network throughput and disk storage
Appendix B: Quality Attributes
389
Table 1. Typical Availability Failure Scenarios in Order

Reason
Inadequate software testing
Lack of a real change management discipline
Lack of quality software engineering processes
The inability to monitor the environment in real-time
No tools that allow for performing long-term trend analysis
Operating environment failure
Sudden changes in usage levels
Percentage
25
20
20
15
10
5
5
and redundancy. The discovery and identication of bottlenecks at higher levels will
also require higher-skilled developers who have a skilled understanding in software
engineering. Finally, higher levels of availability require automated comprehensive
testing of every service that may affect your application at run-time.
1.2. Manageability
Distributed applications and services make customer access and data exchange with
business partners easy but come at the expense with the difculty in diagnosing
and resolving capacity and resource access issues in a production environment.
Quality assurance also becomes a run-time requirement that may introduce the need
to perform audits and incorporate ndings into maintenance upgrades. This can
become difcult in infrastructures that are distributed across large geographic areas.
It becomes important to incorporate at design-time, interfaces that will allow not only
for operational quality assurance but the continuous measurement of service health
including such factors as resource consumption, faults, and statistics on consumer
service requests and overall service performance.
1.3. Performance
In many enterprises, there is a widely held belief that performance and scalability
are interchangeable terms when it reality they describe two different problem spaces
and can be in many situations opposing. Performance is usually not apparent until a
particular instance of a service is put under an increased load.
390
Conducting code reviews and/or using pair-programming style development can

uncover performance bottlenecks. At run-time, performance testing of services
should be conducted to identify bottlenecks checking for slow performing code
routines or contention for resources.
1.4. Scalability
Scalability must be designed into a service early in its lifecycle, as this is the hardest
attribute to support after the fact. Decisions made during design and early coding
phases will dictate how well your services scale. Scaling of services also requires
working knowledge of the deployment environment as it has dependencies on the
underlying hardware.
Hardware-based scaling can be accomplished by scaling up which requires adding
resources such as more memory, faster and/or more processors or even migrating to
a newer class of machine. Typically, this approach permits some scalability increases
without requiring changes to source and keeps administration simpler.
Scaling up can also have the effect of moving the bottleneck from one component to
another. For example, a server that is at 99 percent CPU utilization could increase
capacity by adding another CPU. However, this may shift the bottleneck from the
CPU to the system memory. Adding CPUs does not increase performance in a linear
fashion. In most servers, the performance gain curve tapers off as each additional
processor is added. Minimally, in order to take advantage of adding multiple CPUs,
your services should support multiple threads of execution.
Scaling out provides a different alternative to scaling up and hopes to leverage the
economics of using commodity hardware to distributed processing across multiple
smaller servers. Although scaling out is achieved by using many servers, the collection
functions as a single resource. By dedicating multiple machines to a common task,
service fault tolerance is increased at the expense of additional administration.
Scaling out provides a mechanism that results in near linear increases in scalability
as more servers are added. The key attribute to achieving scaling in this manner
is location transparency. Application code must not know what server is executing
the service. In some rare situations, it is still possible to scale out when applications
are aware of what servers are executing services. This situation is known as location
afnity. Location afnity requires changes to code to scale from one server to many. It
is easier to design services with location transparency at design time than refactoring
later.
391
1.5. Security
The ability to provide security in a distributed computing environment is all about
controlling access to a variety of resources (services, data, conguration, etc.). There
are four main concepts on which security practices and providers are based:
1. Authentication,
2. Authorization,
3. Data Protection, and
4. Auditing.
Authentication is the process of conrming the identity of the caller and/or service.
Before a service permits access, it may require conrming the identity of the service
requestor. The requestor establishes an identity by providing a credential that is
known only to the requestor and the authenticating service. In some situations, the
requestor may also desire to validate the identity of the authentication service, which
is known as mutual authentication.
Authorization is the process where verication of the authenticated party has the
appropriate permissions to access a specied resource. You cannot assume that because
someone has been authenticated that they are authorized for all actions.
Data protection is the process of ensuring the condentiality, integrity and nonrepudiation of data. Data, whether contained within a message in transit or on disk,
requires protection. Encryption assists in protecting the condentiality of data and
renders it useless to parties who lack knowledge of which cryptographic algorithm
was used and the key. Data integrity is realized through the use of hashing algorithms,
digital signatures and message authentication codes. Auditing is the process of logging
of security related events and can be used for forensic purposes.
2. Design vs Run-Time
Quality attributes are properties that govern how a system behaves over and above
its functionality. There are two categories of quality attributes, those that can be
determined at run-time and those that can only be estimated through inspection. It is
the prime responsibility of architects to ensure that all systems within their portfolio
have identied quality attributes and they are prioritized to those of which are most
important.
392

Table 2. Run-Time Quality Attributes
Quality Attribute Description

Availability
The amount of time that a system is up and running. This can be measured by
the length of time between failures and how quickly the system is able to resume
activity after a failure.
Performance
Reliability
Usability
A measurement of system response time.

The ability of a system to operate over time in a predictable manner. Reliability
is measured by mean time between failure (MTBF).
A measure that determines how easy it is for a user of the system to understand
and operate.
Table 3. Inspection Time Quality Attributes

Quality Attribute
Buildability
Description
The ability for a system to be constructed using the budget, time and staff
available for the project in many cases is simply too ambitious to be completed
given project constraints.
Conceptual integrity The ability of the architecture to communicate a clear, complete vision for
the system. Using agile methods, this is typically realized using metaphors. If
it does not feel right, then conceptual integrity is lost.
Integrability
Systems over time will need to integrate with other systems. The Integrability
of a system depends on the extent in which the system uses open standards,
how well APIs are designed and the usage of approaches such as serviceoriented architectures.
Modiability
A measure of how easy it is to change the system to incorporate new functional

requirements. There are two aspects to modiability, time and cost. If a system
uses an obscure or outdated technology that requires expensive consultants
or other hard to nd specialists, then even though it may be quick to change,
its modiability can still be low.
Portability
A systems platform may consist of hardware, operating systems, application

server software and databases. The ability to measure the ease to move to
different platforms.
The reusability of a system is the ability to reuse portions of the system in
other applications. Components that can be reused include: the run-time
platform, source code, libraries and processes.
The ability of a system to resist unauthorized attempts to access system
resources while providing services to authorized users.
Reusability
Security
Subsetability
For incremental development approaches, a system that can execute to

demonstrate small iterations demonstrates subsetability. This attribute becomes important if time and/or resources on the project are reduced. If the
subsetability of the architecture is high, then a subset of features may still
make it into production.
Testability
How easily can a system be tested using human effort, automated testing
tools, code reviews and inspections and other forms of ensuring system
quality.
393
In Table 2, the system qualities that can be evaluated at run-time are given, whereas
Table 3 shows the system qualities that can only be reasonably evaluated through
inspection.
The quality of an enterprise SOA is directly proportional to the quality incorporated
into the enterprise architecture. A thorough understanding of system qualities not
only for services but all components of an architecture will result in lower total cost
of ownership and increased business agility.
APPENDIX C:
REFERENCES
It is amazing what can be accomplished when nobody cares about who gets the credit
Robert Yates
Books
Alur, D., Malks, D. and Crupi, J. (2003) Core J2EE Patterns: Best Practices and Design
Strategies. Prentice Hall.
Atkinson, C. et al. (2002) Component-Based Product Line Engineering with UML.
Addison-Wesley.
Beer, S. (1979) The Heart of Enterprise. Wiley.
Bernstein, P.A. and Newcomer, E. (1997) Principles of Transaction Processing. Morgan
Kaufmann.
Buschmann, F., Meunier, R., Rohnert, H., Sommerlad, P. and Stal, M. (1996)
Pattern Oriented Software Architecture, Volume One. John Wiley and Sons.
Clements, P. and Northrop, L. (2002) Software Product Lines. Addison Wesley.
COD (1982) The Concise Oxford Dictionary. Oxford University Press.
Eeles, P. and Sims, O. (1998) Building Business Objects. Wiley.
Farley, J. (1998) Java Distributed Computing. New York: OReilly.
Ferguson, N. and Schneier, B. (2003) Practical Cryptography. John Wiley & Sons.
395
396
Enterprise Service Oriented Architectures
Gamma, E., Helm, R., Johnson, R. and Vlissides, J. (1994) Design Patterns: Elements
of Reusable Object-Oriented Software. Addison Wesley.
Gamma, E. et al. (1995) Design Patterns. Addison Wesley.
Gray, J. and Reuter, A. (1993) Transaction Processing: Concepts and Techniques.
Morgan Kaufmann.
Guttman, M. and Matthews, J. (1998/1999) Migrating to Enterprise Component
Computing. Cutter Consortium.
Harmon, P., Rosen, M. and Guttman, M. (2001) Developing E-Business Systems and
Architectures. Morgan Kaufmann.
Herzum, P. and Sims, O. (2000) Business Component Factory. Wiley.
Hubert, R. (2002) Convergent Architecture. Wiley.
Greeneld, J. and Short, K. (2004) Software Factories. Wiley.
Keen, M. et al. (2004) Patterns: Implementing an SOA Using Enterprise Service Bus.
IBM Redbook.
Lawson, M. (2003) Finite Automata. CRC Press.
McGovern, J., Bothner, J., Cagle, K., Linn, J. and Nagarajan, V. (2003) XQuery Kick
Start. Sams Publishing.
McGovern, J., Ambler, S., Stevens, M., Linn, J., Sharan, V. and Jo, E. (2003) A
Practical Guide to Enterprise Architecture. Upper Saddle, NJ: Prentice Hall.
McGovern, J., Ambler, S., Caserio, C., Narayan, N., Tyagi, R. and Biggs, M. (2004)
Agile Enterprise Architecture. Connecticut: Manning Publications.
Morgan, T. (2002) Business Rules and Information Systems: Aligning IT with Business
Goals. Boston, MA: Addison Wesley.
Pawson, R. and Matthews, R. (2002) Naked Objects. Wiley.
Rector, B. (2003) Introducing Longhorn for Developers, chapter 3. Microsoft.
Appendix C: References
397
Robinson, M., Tapscott, D. and Kalakota, R. (2000) e-Business 2.0: Roadmap for
Success. New York: Pearson Educational.
Ross, R.G. (2003) Principles of the Business Rule Approach. Addison Wesley.
Shalloway, A. and Trott, J. (2004) Design Patterns Explained: A New Perspective on
Object-Oriented Design, 2nd edition. Addison Wesley (rst edition 2001).
Sims, O. (1994) Business Objects. McGraw-Hill.
Taylor, D.A. (1995) Business Engineering with Object Technology. Wiley.
Magazines
Mehta, T. (2003) Adaptive Web Services Management Solutions. Enterprise Networks
& Servers Magazine, May.
Petzold, C. (2004) Create Real Apps Using New Code and Markup
Model. Microsoft MSDN Magazine, January (http://msdn.microsoft.com/
msdnmag/issues/04/01/Avalon/default.aspx).
Docs
Abrahams, A.S., Eyers, D.M. and Bacon, J.M., An Event-Based Paradigm for
E-commerce Application Specication and Execution. Department of Operations
and Information Management, The Wharton School, University of Pennsylvania.
Bray, T., Paoli, J., Sperberg-McQueen, C.M. and Male, E. (1998) Extensible Markup
Language (XML) 1.0 (Second Edition), World Wide Web Consortium, 10 February.
Bray, T., Hollander, D. and Layman, A. (1999) Namespaces in XML, World Wide
Web Consortium, 14 January.
Catania, N., Web Services Events, Web Services Management Framework, Hewlett
Packard.
398
Combine (2003) Component-Based Interoperable Enterprise System Development

An EU Framework 5 Project, see http://www.opengroup.org/combine/overview.htm.
Cummins, F. (2002) White Paper on Web Services Integration Architecture. OMG
Document bei/02-10-02.
Murry, B., WSMF Foundation, Web Services Management Framework, Hewlett
Packard.
OMG1 (2003) MDA Guide version 1.0, OMG document omg/2003-05-01; also
www.omg.org/mda.
OMG2 (2003) Unied Modeling Language Specication, OMG document
formal/03-03-01 (http://www.omg.org/technology/documents/formal/uml.htm).
OMG2 (2004) UML Prole for Enterprise Distributed Object Computing
(EDOC). OMG documents formal/04-02-01 through formal/04-02-08 (see
http://www.omg.org/technology/documents/formal/edoc.htm).
OMG3
(2004)
Unied
Modeling
http://www.uml.org/#UML2.0).
Language
Version
2.0
(see
OMG3 (2004) UML 2.0 Superstructure Specication. OMG Document ptc/04-1002 (see also http://www.omg.org/technology/documents/formal/uml.htm).
The Open Group (1991) Distributed Transaction Processing: The XA Specication.
Rogers, S. (2003) An Integrated Web Services Management Approach, IDC, July.
Sims, O. (2001) Making Components Work. Cutter Consortium, Executive Report
Vol. 4, No. 9.
Sims, O. (2002) A Component Architecture. Cutter Consortium, Executive Report
Vol. 5, No. 5.
Understand Enterprise Service Bus Scenarios and Solutions in Service-Oriented
Architecture, Part 1, IBM Developer Works.
Vambenepe, W., WSMF: Web Services Model, Web Services Management
Framework, Hewlett Packard.
399
Weblogic Server 7.0, Distributed Computing with BEA Weblogic Server, BEA
Systems.
Web Services RoadMap: Security in a Web Services World: A Proposed
Architecture and Roadmap, A Joint White Paper from IBM Corporation
and Microsoft Corporation April 7, 2002, Version 1.0, http://www106.ibm.com/developerworks/security/library/ws-secmap/
XUL (2004) The Open XUL Alliance, http://xul.sourceforge.net/links.html (see also
http://www.xulplanet.com/).
Web Sites
Agile
Alliance,
Agile
http://www.agilemanifesto.org
Manifesto
for
Software
Development,
Arjuna Technologies Ltd., www.arjuna.com

IBM UDDI V3 Registry Site, https://uddi.ibm.com/beta/registry.html
IBM UDDI Test Registry Site, https://uddi.ibm.com/testregistry/registry.html
IBMs XML Security Suite, http://www.alphaworks.ibm.com/tech/xmlsecuritysuite
Java Transaction API 1.0.1B, http://java.sun.com/products/jta/index.html
Java Transaction Service 1.0, http://java.sun.com/products/jts/index.html
Kerberos: The Network Authentication Protocol, http://web.mit.edu/kerberos/www/
Microsoft Technet, various pages: http://www.microsoft.com/technet
Microsoft UDDI page, http://msdn.microsoft.com/library/default.asp?url=/library/
en-us/dnuddispec/html/uddispecindex.asp
OASIS BTP Committee Specication, http://www.oasis-open.org/committees/
business-transactions/
400
Object Request Broker, http://www.sei.cmu.edu/str/descriptions/orb.html

Object Transaction Service 1.3, http://www.omg.org/technology/documents/formal/
transaction_service.htm
Portland
Pattern
Repository,
wiki?WelcomeVisitors
various
pages:
http://www.c2.com/cgi/
Public-Key Infrastructure (X.509) (pkix), http://www.ietf.org/html.charters/pkixcharter.html

Ryan, J., various articles, http://www.developer.com
Security Assertion Markup Language (SAML), http://www.oasis-open.org/
committees/security/
Techtarget, various pages, http://whatis.techtarget.com
UDDI Specications, http://uddi.org/pubs/uddi_v3.htm
WCAF, The Web Services Composite Application Framework Technical Committee,
http://www.oasis-open.org/committees/documents.php?wg_abbrev=ws-caf
WSAA, The Web Services Atomic Transaction specication,
http://msdn.microsoft.com/library/default.asp?url=/library/en-us/dnglobspec/html/
wsat.asp
WSBA, The Web Services Business Activity specication,
http://msdn.microsoft.com/library/default.asp?url=/library/en-us/dnglobspec/html/
wsba.asp
WSBPEL, http://www-106.ibm.com/developerworks/webservices/library/ws-bpel/
WSC, The Web Services Coordination specication, http://msdn.microsoft.com/
library/default.asp?url=/library/en-us/dnglobspec/html/wscoor.asp
WSDL Specications, http://uddi.org/pubs/uddi_v3.htm
WS-Security, http://www.oasis-open.org/committees/tc_home.php?wg_abbrev=wss
XACML, http://www.oasis-open.org/committees/tc_home.php?wg_abbrev=xacml
401
XML Canonicalization, http://www.w3.org/TR/xml-exc-c14n/

XML Encryption, http://www.w3.org/TR/xmlenc-core/
XML Key Management Specication 2.0 (XKMS), http://www.w3.org/TR/xkms2/
XML Signature, http://www.w3.org/TR/xmldsig-core/
Presentations
XML on Wall Street, February 2004, James McGovern and Jeff Ryan, Enterprise
SOA in Financial Services.
Tyndale-Biscoe, S., Sims, O., Sluman, C. and Wood, B. (2002) Business Modelling
for Component Systems with UML. Paper given at the EDOC 2002 Conference.
APPENDIX D:
ADDITIONAL READING
If someone is going down the wrong road, he doesnt need motivation to speed him up.
What he needs is education to turn him around.
Jim Rohn
The author team has composed a list of its favorite books that will further guide
you on the straight path of becoming a superior architect as we understand that
creating a service-oriented architecture requires many distinct disciplines in order to
be successful. The books listed below span multiple subject areas (in no particular
order) and will put you on the road leading to mastery of service-oriented architectures.
Java Web Services Architecture, James McGovern, Sameer Tyagi, Michael E. Stevens,
Sunil Mathew Morgan Kaufman, April 2003
False Prophets: The Gurus Who Created Modern Management and Why Their Ideas are
Bad for Business Today, James Hoopes, Perseus Publishing, April 2003
The Pragmatic Programmer: From Journeyman to Master, Andrew Hunt, David
Thomas, Ward Cunningham, Addison Wesley, October 1999
Agile Modeling: Effective Practices for Extreme Programming and the Unied Process,
Scott Ambler, Ron Jeffries, John Wiley & Sons, March 2002
e-Enterprise: Business Models, Architecture and Components, Faisal Hoque, Cambridge
University Press, April 2000
Software Architecture: Perspectives on an Emerging Discipline, Mary Shaw, David
Garlan, Prentice Hall, April 1996
403
404
Software Product Lines: Practices and Patterns, Paul Clements, Linda M. Northrop,
Addison Wesley, August 2001
Elements of UML Style, Scott Ambler, Cambridge University Press, December 2002
Whats the Big Idea? Creating and Capitalizing on the Best New Management Thinking,
Thomas Davenport, Laurence Prusak, H. Wilson, Harvard Business School Press,
April 2003
How to Open Locks with Improvised Tools, Hans Conkel, Harper Collins, October 1997
Zen and the Art of Motorcycle Maintenance, Robert Pirsig, Bantam Books, April 1984
The Mythical Man-Month: Essays on Software Engineering, Frederick P. Brooks,
Data Access Patterns: Database Interactions in Object-Oriented Applications , Clifton
Nock, Addison Wesley, September 2003
Domain-Driven Design: Tackling Complexity in the Heart of Software, Eric Evans,
Enterprise Integration Patterns: Designing, Building and Deploying Messaging Solutions,
Gregor Hohpe, Bobby Woolf, Addison Wesley, October 2003
Survival Is Not Enough: Why Smart Companies Abandon Worry and Embrace Change,
Seth Godin, Charles Darwin, Touchstone Books, December 2002
Smart Mobs: The Next Social Revolution, Howard Rheingold, Basic Books, October
2003
The Practical Guide to Enterprise Architecture, James McGovern, Scott W. Ambler,
Michael E. Stevens, James Linn, Vikas Sharan, Elias Jo, Prentice Hall, November
2003
APPENDIX E:
UPCOMING BOOKS
Leadership is not so much about technique and methods as it is about opening the heart.
Leadership is about inspiration of oneself and of others. Great leadership is about
human experiences, not processes. Leadership is not a formula or a program, it is a
human activity that comes from the heart and considers the hearts of others.
It is an attitude, not a routine.
Lance Secretan
Listed below are books that members of this author team are currently working on
and their anticipated release dates. We thank you for your continued support and
hope to serve you well in the future.
Agile Enterprise Architecture Fall 2006

Over the past decade, many enterprises have undertaken enterprise architecture
efforts. While many of these programs have had varying success, the vast majority
have failed failed to gain management support, failed to deliver actionable results,
or failed to result in meaningful change within the enterprise.
The lack of success in enterprise architecture is symptomatic of the lack of success
in IT organizations in general. In most companies, IT is perceived to be dying a
slow death. It is viewed as a necessary evil and not a critical part of the success of
the broader enterprise. To achieve the status of strategic partner, IT must re-focus its
architecture efforts from enabling the technology to enabling the business.
In order to achieve a chaordic balance in strategy, design and implementation, agile
approaches to enterprise architecture must be employed. Methods for performing
405
406
each of these activities will be explored along with practical considerations with a
focus on the issues, strategies and practices of enabling business agility.
In this book, readers will learn about:
IT Governance,
Portfolio Management,
Quality Assurance,
Software Risk Management,
Defect Management,
Agile COTS Assessment,
Architectural Assessment,
Process Patterns,
Agile Procurement Processes, and
Thought Leadership.
Enterprise Portal Architecture Fall 2006

Portals are a technology that allows for integrating various resources within the
enterprise (applications, databases, and so on) available to end users through a
unied user interface. Open source technology enables enterprises to construct highly
dynamic enterprise portals with increased agility and minimal limitations. This book
explains the fundamentals behind portal technologies and how to take full advantage
of offerings in this space.
In this book, one will learn:
The ability to develop a common approach to recurring problems within the
portal space leveraging pattern-based approaches.
Evaluating business requirements and planning for a successful implementation.
Development of robust portal solutions using the premier open source portal:
Liferay.
Implement federated security and single sign on across multiple portals.
Appendix E: Upcoming Books
407
Incorporate personalization and user prole concepts to tailor site content.

Integrating portals with web services.
Architectural and network diagrams that illustrate detailed portal implementations.
Integration with leading ERP, CRM and supply chain vendors.
Make better decisions empowering users to gain rapid access to crucial
information.
Practical considerations for implementing portals in large enterprises.
Reduce the risk of project failures in large integration and portal implementation
efforts.
Enterprise Open Source Spring 2007

Use of open source software within the enterprise is gaining momentum. The vast
majority of enterprises have some form of open source software used in production
environments, which may include Linux, Apache, JBoss and so on. The enterprise
architecture however needs to incorporate the best thinking of the industry that not
only includes using open source but contributing to it.
The model in which open source software gets developed has practices, which could
assist an organization in becoming agile in their software development practices, and
allows them to develop software faster, with cheaper cost and of better quality.
Two models of software development: The cathedral and the bazaar.
Value proposition of using open source to executives.
How much does free software really cost?
Paying for software in forms less tangible than money.
Making the build vs buy decision to include using open source.
The effect of open source development on project management.
Merging the open source development model into enterprise architecture.
Reducing the total cost of ownership for software projects.
408
Enterprise BPM Patterns Summer 2007

There is a lot of hype in industry trade journals about business process management
(BPM). Some believe it is a revolutionary process that will create new categories of
software used in developing enterprise applications. Others believe it is evolutionary
in that it will help leverage existing business and technology assets create new value
propositions. Along with any revolution comes confusion. Many will be confused
with the terminology used. Some will ask is it workow? While others will struggle
with the notion of orchestration and choreographed business processes.
Business process management is really a paradigm shift. The nature of processes
themselves presents severe, even insurmountable challenges for IT implementation.
In many enterprises, the processes themselves have requirements far from classical
automation. Change management and product lifecycle management as disciplines
can also be thought of as processes in their own right. The ideal manner in which
enterprises can learn about the concepts, practices and recommendations for business
process management is using a pattern-based notation.
The ability to have a common vocabulary and solution to recurring problems
within the business process management space using a pattern-based approach.
Practical considerations for implementing business process management in large
enterprises.
Reduce the risk of project failures in large integration and business process
management efforts.

Enterprise Service Oriented Architectures

Uploaded by

Copyright:

Available Formats

Enterprise Service Oriented Architectures

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Enterprise Service Oriented Architectures

Uploaded by

Copyright:

Available Formats

What are some of the main topics covered in the book?

What are some of the main topics covered in the book?

What are some of the challenges of implementing business process management?

What are some of the challenges of implementing business process management?

ENTERPRISE SERVICE ORIENTED ARCHITECTURES

Enterprise Service Oriented

Concepts, Challenges, Recommendations

Printed on acid-free paper

All Rights Reserved

The condition of an enlightened mind is a surrendered heart.

ABOUT THE SERIES

ABOUT THIS BOOK

ABOUT THE REVIEWERS

CHAPTER 1: UNDERSTANDING SERVICE-ORIENTED

1. Introducing Service-Oriented Architectures

2. Service-Based Collaboration through Federation

CHAPTER 2: COMPONENT-BASED SERVICES

1. Component-Based Software Engineering (CBSE)

4. From Requirements to Design

2. The Business Process Execution Language (BPEL)

3. A Worked Example of Web Services Orchestration

CHAPTER 4: WORKING WITH REGISTRY AND UDDI

1. Introducing the Registry

2. Universal Description, Discovery and Integration (UDDI)

CHAPTER 5: UNDERSTANDING ENTERPRISE SECURITY

1. Need for a Message Level Security Solution

4. Web Services Security (WSS)

CHAPTER 6: SOA MANAGEMENT

10. Policy Architecture

11. Framework Vendors

1. What Are ACID Transactions?

2. Why ACID Is Too Strong for Web Services

3. A Brief History of Web Services Transactions

4. The Coordination Frameworks

5. Web Services Transactions

CHAPTER 8: EVENT-DRIVEN ARCHITECTURE

5. Alternative Pattern-Based Approaches

6. Language Specic Constructs

7. Finite State Machines

APPENDIX A: UNDERSTANDING DISTRIBUTED COMPUTING

APPENDIX B: QUALITY ATTRIBUTES

APPENDIX D: ADDITIONAL READING

APPENDIX E: UPCOMING BOOKS

ABOUT THE SERIES

About the Series

About the Series

research on software specication methods, software development processes, tool

Service-oriented architectures (SOA) fundamentally changes the way enterprise

Figure 1. Service-Oriented Architecture

Functional decomposition of services and their underlying components with

ABOUT THIS BOOK

About This Book

What This Book Is Not!

About This Book

Effective XML: 50 Specic Ways to Improve your XML OReilly.

How to Use This Book

About This Book

Motivation for Writing This Book

About This Book

About the Authors