Motivations for
Protocol Design and Validation
Introduction
Slides
Thank you to
Jörg Ott jo and
Carsten Bormann
For the entire content
Motivation: Why still Protocol Design?
• New applications appear all the time – more and more net-based
• Within applications, functional decomposition and distribution
makes protocol design an inherent part of system design
• Evolution of communication technology incurs new demands
• Environmental changes require reconsidering the design of
existing protocols
• Migration ( “convergence”) requires re-thinking solutions to old
problems for a new environment (e.g. IP telephony, IP TV)
• Vast variety of problems and solutions
• Simple (e.g., just use RPC) vs. complex (BGP-4 for telephone numbers)
• All layers (from wireless MAC to QoS to auto-configuration to applications)
• Closed environments (within a product) to open standards
What is Protocol Design?
Many possible views
• Mathematical modeling
• Design and correctness proofs
• Protocol engineering process
• Management and process aspects of protocol design (software engineering view)
• Building blocks and design patterns
• Mechanisms for certain functions in creating protocols
• Tool chains for protocol specification, implementation, and validation
• Automating the creation process (but not the conceptual thinking)
• …
• We are interested in
• Why some designs work better (get accepted) than others (which don’t)
• Ideas of what is known as good practice beyond the engineering literature
• Understanding relationship between functional and non-functional aspects
• Considering some non-technical real-world aspects as well
Requirements Aspects
Understanding which problem to solve
Real problems vs. thoughts about solutions in search for a problem
Understanding the requirements
Functional: features, security, …
Non-functional: scale, operational aspects, time-to-market, cost
Understanding the constraints
Functional: operational environment
Non-functional: cost, weight, energy consumption, memory, CPU,
…
Understanding the acceptable tradeoffs
Must vs. nice-to-have
Is this some special case of a more general problem?
If so: does the problem become simpler by generalizing?
If not, is the more general problem worth solving?
Some General Protocol Design Aspects (1)
• Design scope
• Part of a specific application design
• Creation of a platform for a competitive environment
• Design target
• Complete solution, e.g., for an application
• Creation of building blocks targeted at flexible re-use
• Use of building blocks or technologies to create a particular solution
• Important design decision: Make or take
• Re-use existing technologies (accept less than 100% match)
• Benefit from experience, code, etc.
• But: who has change control, how long will the technology be supported,
does it really fit, will both protocols evolve in parallel, …?
• Create new technology from scratch (accept higher risk, longer time to market)
Some General Design Aspects (2)
• Learning from solutions to related problems
• Borrow concepts and mechanisms – but only where applicable!
• Avoid mistakes. Look at real-world deployments before
borrowing
• Yet avoid the “second system syndrome”
• Remember requirements during the design phase
• Optimize for the common case (if at all)
• Don’t over engineer – Keep it simple
• Avoid options and parameters
• Remember that it needs to be implemented in the end
• (we will address these and more such issues during the course)
Some General Protocol Design Aspects (3)
Separation of concerns
• Treat and solve independent aspects independently
• Caveat: what is really independent?
(Strict) layering
• Black box, well-defined service access points (SAPs) with layer-internal protocols
• Intends to completely shield lower layers and communication details from higher
layers
Leaky abstraction
• Strict layering will not always work, particularly if things go wrong
• Expose issues rather than trying to conceal them at any cost
• Applies to protocol design, to coding (and code generation), and others
Cross-layer optimization gaining importance
• Deal with dependencies on the lower layers
• Limit: your system is not always directly connected to the weakest link (layer)
Design Validation
Protocol design is relevant to later protocol validation
• From a correctness perspective
• From a performance perspective
Correctness of a specification
• May involve formal specification as design methods
• Using your favorite modeling or specification language
• May involve formal proofs
• Mostly for “simple” protocols and problems
Performance of a specification
• Mathematical modeling and analysis
• Evaluation by means of “implementation” and simulation
Both validations provide important feedback for the design process
Implementation & Validation
Protocol implementations need to be correct and interoperable
• Beware of complexity!
• In some cases, code may be generated from specifications using tools
Again: validation
• Limited functional validation through testing
– Test cases may be generated from specifications
– Usually cover only usage scenarios of limited complexity (explosion of number of tests)
• Performance validation through emulation and field tests with measurements
Difficulty: getting even close to the real-world conditions (in the lab)
• True validation will only occur through real world deployment (“in the wild”)
• Different platforms, different implementations, different user behavior, different
environmental conditions, (different interpretations of the spec), …
– Will also tell something about the impact on the network at large
• Implementation experience provides most important feedback
Conformance vs. Interoperability
Traditional thinking:
• All implementations must conform to specification
• If specification is good, this ensures interoperability
Modern thinking:
• Implementations have errors
• Specifications have errors and ambiguities
• Interoperability is actually more important than
conformance
– This includes interoperability with erroneous, but deployed
systems
Operations and Maintenance
Rollout
Monitoring
• Protocol and device operation
• Its impact on its environment
Diagnosis, Debugging
Protocol evolution over time
• To fix bugs
• To meeting changing or new requirements
– To get rid of unnecessary requirements and constraints
• To deal with changing environmental conditions
A Note on Protocols in the Real World
Protocol design usually makes assumptions
• About the environment it will operate in
– Technical terms: packet network, delay, packet loss, MTU, range of data rate, etc.
– Organization terms: trust, common management, configuration, interaction, etc.
• Lower layer services and characteristics to build upon
• Higher layer applications using it
Protocols may be successful or even “hyped”
– Examples today: HTTP, SIP, XML, to some extent SOAP, …
If they are, they will be used outside their specified limits
• In different environments, at different scales, for different purposes, …
People will blame the designer if they don’t work properly then
• Applicability statements are not necessarily read or adhered to
Subject Areas of Protocol Design
• General design space
• Functional building blocks
• Meta design aspects
Protocol Design is about Trade-Offs…
…given sets of requirements and environmental constraints.
“Good, fast, cheap – pick two, you cannot have all three.”
Examples
• Reliability vs. delay
• Functionality vs. bandwidth
• Extensibility vs. efficiency
• Functionality vs. simplicity
Virtually any design decision taken to achieve one goal will
counteract another
• Need to find a reasonable compromise to achieve desired
function at acceptable cost
Where Theory meets Practice…
Many design rules for protocols can be found
• Mechanisms to achieve certain functionality
• Keep it flexible and extensible
• Make it effective and efficient (optimize)
• Make it resilient
• …
To be applied wisely (not blindly)
• Considering the trade-offs
Beware of complexity
• People will blame the their device or technology if the stuff doesn’t (inter)work
• Regardless of where the problem is
• Too expensive or too difficult to use
Premature optimization is the root of all evil
• ...
Communication Partners and their Roles
Point-to-point vs. multipoint communications
• How many parties are involved in the protocol (from a semantics perspective)?
Unicasting vs. group-overlays vs. multicasting
– What type of information exchange is assumed?
Client-server vs. peer-to-peer communications
• Are the involved parties “equal” or do they have different responsibilities
– Note: peer-to-peer is more general than today’s widespread applications)
• In case of groups: are some more important than others?
– More than just two different classes of peers
• End-to-end vs. intermediaries vs. router-assist
– What kind of entities may, are, or must be involved? Are they “visible” or not?
• Communication among end systems vs. among network elements
– Transport and application vs. routing, network, maintenance protocols
• End-to-middle communications
Identifying Communication Partners
Names
• Human readable identifiers that can be remembered!
(e.g., DNS name, URI, URN)
Identifiers
• Machine-processable identifier (e.g., HI)
Addresses
• Protocol-level identifier (e.g., IP address)
Locators
• Information about the location of a partner in the network topology
Need to be managed (unique assignment)
• Or chosen randomly (and defended) in ad-hoc environments (birthday paradox)
One needs to resolved into the other
• Address books, (distributed) data bases (e.g., DNS, DHTs), protocol exchanges,
caching, (manual) configuration, …
Some Protocol Design “Knobs & Variables”
Stateful vs. stateless operation
• How much information is preserved across information exchanges
– Notion of an “association” or a “connection”
• Where is this state kept (one or both peers in the point-to-point case)?
Fixed nodes vs. mobile nodes
• impact on routing, reachability, …
Wireline vs. wireless communications
• Implications of different link layer technologies in general
Infrastructure-based vs. ad-hoc/autonomous communications
• What types of infrastructure are assumed? (e.g., routing, naming)
Security within the protocol vs. relying on security elsewhere
• Which implications (e.g., for required infrastructure such as PKI)
Functional Building Blocks
Naming and addressing
Rendezvous or invocation mechanisms
Semantics and properties of protocol operations
– Idempotent operations, delta vs. full state updates, synchronization,
Interaction paradigms
– Synchronous, asynchronous, both
– RPC-style operation vs. event notifications at any time
Reliability”
– Includes flow control, sequence preservation, etc.
– How probable is it that a certain operation will not fail.
– Example: transferring a file from A to B unchanged
– Multitude of different protocol mechanisms available
Functional Building Blocks (2)
Multiplexing
– Within the application protocol vs. using lower/requiring higher
layer mechanisms
“Multi-threading”
– Allowing multiple ongoing interactions at the same time
– E.g. lock-step vs. “windowing”
Security
– Authentication, integrity, non-repudiation (sender, receiver),
confidentiality
– Authorization of operations
(Auto)configuration
– How to get a system into a working condition
Meta Aspects of Protocol Design (1)
Independent of specific functions, yet to be provided in line with the
respective protocol
Adaptivity
– Capability of adapting to different environmental conditions (typically “QoS”)
(graceful degradation of service as long as acceptable)
– Example: play-out delay and codec adaptation with IP multimedia
Scalability
– Capability of working across a wide range of environmental parameters
• Typical example: Number of operational nodes
• Data rate, error rate, path length, delay (see above)
• Number and size of data items
Efficiency
– Maintaining a reasonable level of overhead
• Example: protocol encoding, protocol headers
Meta Aspects of Protocol Design (2)
Performance
– Number of protocol interactions, packets, bits, processing
– But don’t optimize (too early in the process)!
Security (again!)
Deployability
– One special case: robustness (against DoS, single point of failure,
etc.)
– Another special case: ability for stepwise introduction into the real
world
Evolvability
– Backward and forward compatibility
Operability and manageability
