Large Scale Distributed Simulation on the Grid

Georgios Theodoropoulos, Yi Zhang, Dan Chen, Rob Minson

University of Birmingham
Edgbaston, Birmingham, B 15 2TT, UK
{gkt,yxz,cxd,rzm} @ cs.bham.ac.uk
Stephen J. Turner, Wentong Cai Yong Xie
School of Computer Engineering Computing Laboratory
Nanyang Technological University Oxford University
Singapore 639798 Oxford, OX1 3QD, UK
{ASSJTurner,aswtcai) @ntu.edu.sg [email protected]
Brian Logan
School of Con~puterScience & Information Technology
University of Nottingham
Nottingham NG8 1BB, UK
[email protected]

Abstract projects from around the globe. One of only four such
projects is DS- rid', a collaboration between the Midlands
The developtn~rztc5/' rrmrzy conlpiex sinzulaiion applica- e-Science Centre (M~sc" at the School of Computer Sci-
tions requires collaborrrtive e f i r t from resear-chers with ence, University of Birmingham and the Parallel and Dis-
direrent domclin h o w b d g e and experfise, possibly at dij; tributed Computing Centre (PDCC3) at the School of Com-
ferenr locutic~irs. These sirnulutioiz s.ystenzs oBen require puter Engineering, Niuiya~~g Technological University, Sin-
huge computing resources and dart1 sets, which rntay be ge- gapore, with participation of the School of Computer Sci-
ographically distributed. In order. to support collabopntive ence & lnformatio~lTechnology, University of Nottingham.
model development and to cuter for the increasirzg complex- DS-Grid is concenled with Large Scale Distributed Sim-
iiy of such .systeins, it is necessay to harness distriblited ulation on the Grid. The vision of the project is a "Grid
resources over the 6zterrzet. The en~ergerzceoj' Grid frch- plug-and-play distributed simulation system", a distributed
nologies provide exciting new opponctrzitiesjbr Large scale collaborative simulation environment where researchers
distributed simulatiorz, enabling collaboration and ihe use with different domain knowledge and expertise, at differ-
of distributed computing resources, whi/e al.~ojircilitutirzg ent locations, develop, modify, assemble and execute dis-
access to geographically distributed data sets. This paper tributed simulation components over the Grid. The sirnula-
discusses the rsseurch clu~znlletzgesthaj I T Z L L Sbe
~ idd dressed tion colnponents reside in the Grid as Grid services which
bcfi~isthese opport~rrzitiestun be exploited and pre.smf,s the system will automatically discover, senlantically match
HLA-GRID-REPAST, a system tor executing large scale (with each other and with the simulation goals), and return
distribl.~redsinzulcrtions of ugent based system? over the to the user. A number of important new research challenges
Grid. have to be addressed before this vision is realised; address-
ing these challenges is the long tenn objective of the DS-
Grid collaboration.
Building on existing complementary efforts in the UK
1. Introduction
'"DS-Grid: Large Scale D~istributedSimi~latio~l
on the Grid", e-Scicnce
Sister Project FWS8286U01. UKL: hL t p : //www .cs .bham.ac.uk/
In h e Spring of 2003, the UK e-Science Core Pro- research/prujects/dsgr~d
gramme issued a call for proposals to eskiblish synergis- 'http://www.rnesc.bham.ac.uk
tic links between UK e-Science projects and leading Grid 3http://pdcc.ntu.edu.sg/

Proceedings of the Sixth IEEE International Symposium on Cluster Computing and the Grid Workshops (CCGRIDW'06)
0-7695-2585-7/06 $20.00 © 2006 IEEE
 and Singapore, ~iiostnotably the woi-lc on service oriented t~ativeto develop a web enabled RTl 191. Within the XMSF
middlew'e for HLA simulations on the Grid in Singa- fraiiework, niultiple federates reside as web services on
pore (the HLA-GRID system) and the work on HLA dis- a WAN and the Federation's FOM is mapped to an XML
tributed simulations of agent-based systems in Birniingham tagset, allowing interoperation with other distributed appli-
(the HLA-REPASTsystem [6]), the project has developed cations supported by web services. The lederates communi-
HLA-GRID-REPAST, a prototype plall'orni for executing cate using the Simple Object Access Protocol (SOAP) and
HLA agent-based sin~ulationon the Grid. the Blocks Exteilsible Exchange Protocol (BEEP).
In [13] we provided an overview of tlie Another important init~ativeis HLA-GRID, origillally
HLA-GRID-REPAST syslem and presented some prelimi- developed at the Nanyang Technological University in Sin-
nary perfomlance results. This paper discusses the research gapore [5, 12, 141. In HLA-GRID, federates can be instan-
challenges in developing and executing large scale distrib- t~iitedas Grid services wh~chare used to facilitate commu-
uted siniulations on the Grid, and presents a more in depth nication between Gnd service federates a i d the RTI.
description of the HLA-GRID-REPAS'Ksystem. The rest The HLA is a mature technology which offers enor-
ol' the paper is organised as follows: Section 2 outlines the mous potential for large scale distributed simulation on the
key research challenges in Grid-aware distributed siniula- Grid. However, a number or key research challenges musl
tion. Section 3 discusses HLA-GRID-REPAST, providing be overcome before this potential can be realised. The fol-
an overview of the HLA-GRID a i d HLA-REPAST lowing sections discuss some of these challenges.
systems. Section 4 concludes the paper.
2.1 Model Discovery and Semantic
2 Research Challenges in Large Scale Simu- Matching
lation on the Grid
While the FOM and RTI provide interoperability at the
Modelling and sin~ulationis an essential tool in many ar- communication level, there is little support for interoper-
ability at the semantic level; as a result, simulation devel-
eas of science and engineering, for exmiple, for analysing
natural phenomel~aor predicting the behaviour of new sys- opment lead times are often on the order of months, since
the intei-pretation of data by each federate must be fully un-
tenis being designed. The development of such coniplex
simulation applications usually requires collaborative effort derstood and carefully checked for consiste~lcyby the sim-
ulation developer. Thus, there is a clear need for automated
from researchers with diff'erent domain knowledge and ex-
pertise, possibly at different locations. Furthermore, these federate discovery aid configuration in a Grid environment.
simulation systems often require huge computing resources The configuration and reuse of software coniponents is
and the data sets required by the simulation may also be a key problem in several fields including agent-based sys-
geographically distributed (e.g. in a supply chain simula- tems, software engineering and Web services 11 I]. Of these,
lion involving different companies, tlie most up to date daca the agents community has made the niost progress in au-
will be in the individual conipwies). In order to support tomating component configuration, but work tends to be
collaborative model development and to cater lor the in- domain specific and there are no well established standards
creasing complexity of such systems, it is necessary to har- and languages. Conlponent-based software engineering has
ness distributed resources over the Internet. The last decade niade substantial progress in modelling components (e.g.
has witnessed an explosion of interest and innovation in the CORBA, .NET and Java Beans) and supporting the huinan
field of large scale distributed simulation. This activity is designer. Components are usually well defined at the syn-
mainly centred on tlie High Level Architecture (HLA), an tactic level (using languages like UML). but semantic anno-
lEEE standard to facilitate interoperability among simula tations are very limited.
tions and promote reuse of siniulation models. Fully automated configuration is limited to specific ap-
The emergence of Grid technologies provide exciting plication domains where simple fonns of component re-
new opportunities for large scale distributed simulation, en- trieval are also supported (e.g. exact-matching). Automatic
abling collaboration and the use of distributed computing configuration and adaptation of software conlponents is cur-
resources, while also facilitating access to geographically rently the focus of research in autonomic computing (e.g.
distributed data sets. In the last few years, there has been an self managing and self healing systems). Web and Grid
i~lcreasinginterest in taking advantage of Grid technologies services support scalable inter-operation of heterogeneous
to execute HLA simulalions over the Internet. software across a wide variety of networked platforms and
A seminal work in this direction is the Extensible Model- a rapid integration cycle and offer the possibility of au-
ing & Simulation (XMSF), a collaborative ini- lomatic composilioil of components obtained by querying
web service registries. The discovery and automatic com-
/ / w w w . m c v e s ~ n s t i t u t e . c r y / . w . ~ n s f / x r i ~.shft r n l
%tttp: position of web services has been addressed through both

Proceedings of the Sixth IEEE International Symposium on Cluster Computing and the Grid Workshops (CCGRIDW'06)
0-7695-2585-7/06 $20.00 © 2006 IEEE
 syntactic and semantic frameworks. Syntaclic franeworlis receiving the attention on researchers in distributed simula-
model services as Business Processes without sem2uitic an- tions (e.g. [4J).
notations and use staldardised technologies for description Mechanisms for federate migration have been devel-
(e.g. WSDL, SOAP, UDDI) and (manual) configuration of' oped, e.g. [ 2 ] . For decision making, models are required
services (e.g. the orchestration languages XLANG, WFSL, which take into account locality of and the communica-
BPEL3WS). The component structure is typically process tiou patterns between the federates for their migration in the
based and is described as a worMow which consists of ac- Grid. This would also require mechanisms for monitoring
tivities with data and control flow. Semantic web technolo- and evaluating the progress of the simulation against some
gies extend these frameworks with machine understandable baseline cost models and assessing the benefits of work mi-
conceptual descriptions of web service "capabilities" (se- gration. Interaction with the various HLA services such as
mantic annotations) to support increased automation of all federation management. ownership inanagement and time
activities within the web services model. management must be considered.
For automated federate discovery and conliguration, ap- The heterogeneity oS resources on the Grid will require
propriate ontologies and languages must be defined to rep- careful evaluation of' trade-offs related to communicalion,
resent the metadata of Sederates and orchestrate simulations compute capacity and appropriateness of specific HLA fed-
by matching user requirements with appropriate federates. erates (for instance, a less desiriiblc feda-ale in terms of its
Each federate can iranifest itself as a Grid service for use properties nyay provide significantly better performance).
by the sirnulalor. The ontologies should represent descrip-
tions, classifica~ionsof descriptions, and constraints related 2.3 Support for Collaborative Model De-
to valid coniiguralions of simulations and should support velopment
both searching for sinlulation nlodels and semantic match-
ing of component models with simulation goals. To achieve To conduct si~nulationexperiments easily over geo-
this, o~itologiesfor the description of the internal behav- graphically distributed resources from different organisa-
iour of federates are required, as well as appropriate reason- lions, mechanisms that can deploy coordinated, secured
ing and matchmaking engines. Although there have been simulation executions are required. The HLA does not pro-
effo~lsto describe the inner mechanisms of co~npositeor vide any support for collaborative developn~entof simu-
stateful services, such as iederates, so f a these have been lation components and applications and hence, new Grid-
limited [I I]. XMSF advocates the use of MDA for mean- aware collaborative environments for distributed simulation
ingful inter-operation of federates. must be developed. There already exists considerable in-
terest in such Advanced Collabordtive Environments in the
2.2 Resource Management Grid community5.
Collaborative Elivironments for HLA simulations on the
A large-scale simulation executing in a dislributed en- Grid could integrate support for capturing requirements,
vironment may need computing resources from many dif- model discovery and matching, worknow specification and
ferent organisations and the availability of these comput- coordinated slitrtup of' federates at each site. For collabo-
ing resources may chkuige during the exec~~tiollof the sim- 1-ativemodel development, it is also essential to create Grid
ulation. To meet the real-time requirements demanded by services to facilitate graphic andlor text-based user interac-
interactive simulations tmd the perfolmance requireme~its tions.
demanded by analytical simulations. sophisticated resource
management mechanisms are required. 3 The IZLA-GridXePast System
HLA does not provide iuly support for resource rnilllage-
lnent and dynamic balancing of simulation load and mecha- HLA-GRID-REPASI 1s designed for executing distrib-
nisms for resource management, load monitoring, dynamic uted, large scale simulations of agent-based systems over
load balancing, check-pointing and migration in a Grid en- Grids. It integrates HLA-REPAS'I and HLA-GRID and
vironnlen t are therefore required. acts as a middleware to glue simulation code wrllten in
Load balancing of distributed simulations has been stud- REPASTwith Lhe HLA.
ied extensively. However, most existing work assumes
paralleVdistributed platforms with negligible iaterproces-
sor communication delays. Their feedback loop is based
mainly on processor load and they tend Lo largely ignore HLA-GRID is a ti-m~eworkdesigned to extend the
cornnlunication latencies. In a Grid, however, conlmuni- HLA to h e Grid. In particular, it focuses on improving the
cation delays are substantial and can be the decisive factor
of the simulation performance. This issue is increasingly

Proceedings of the Sixth IEEE International Symposium on Cluster Computing and the Grid Workshops (CCGRIDW'06)
0-7695-2585-7/06 $20.00 © 2006 IEEE
 Cliz11t side Chent side

Grid-enilbld HLA AL'L I Gnd services 1

I I I I

Gnd-endbled HI A API

LAN

R ~ S O L side
L~C~

Figure 1. Architecture of Proxy-based HLA simulation on the Grid.

interoperability and conlposability of HLA-compliant sinl- programs in t l ~ eDMSO RTl package. More details of
ulation components. HLA-GRID'S design, implementation and experiments can
The framework, which is illustrated in figure 1, achieves be found in 15, 12, 141.
interoperability belween difl'erent simulators (federates) by
using a Federate-Proxy-RT1 architecture, in which differ- 3.2 HLARePast
ent participants (clients) in the same simulation run their
federate codes at their local sites, and the RTIExsc and The REPASTsystem [3] is a Java-based toolkit for h e
FEDEXECare executed on the remote resource. A new development of lightweight agents and agent models. It was
proxy entity is introduced to act on behalf of the client's developed at the University of Chicago's Social Science Re-
federate code and co~nnlunicatewith the proxies of other search Coinpuling division and is derived from the Swarm
clients through the RTI. Proxies are executed at remote grid simulation toolkit. It has become a popular and influential
resources. Federate codes and their respective proxies com- loolkit, assessed by [lo] as the most elTeclive development
municate with each other through Grid services and a Grid- platfom~currently available for Lge-scale simulations of
enabled HLA library, whch provides the standard IILA social phenomena.
API to the federate codes, is implemented to translate the The system provides an inter-dependent collection of
communications into Grid services iilvocations. tools and structures which are generally useful for the sim-
HLA-GRID includes additional Grid services to sup- ulation of agents, and a sequential discrete event simulation
port the creation of (he RTI, discovery of l'ederalions, etc. kernel for the execution of the model.
The framework hides the heterogeneity of the simulators, HLA-REPAS?' is a nliddleware layer which enables the
execution platforms, and how the simulators com~nunicate execution of a federation of multiple interacting instances
with the RTI. All interpaces used in the HLA-GRID com- of REPASTmodels within the HLA as depicled in figure 2.
ply with the standard HLA interface specification. The main task of t l ~ eHLA-REPAST middleware is to
A prototype of HLA-GRID has been implemenled in detect the occurrelice of events in (he REPASTmodel and
Java using the Globus Toolkit version 3. The proto- consistently, reliably and trii~lsparelltlycommunicate them
type includes sta~dardHLAIRTI APIs to support Feder- to Lhe RTI. To achieve this, HLAREPAsT provides nlech-
ation, Time, Object, Declaration and Ownership Manage- ,anisms Tor mapping REPASTslate-trrinsitio~~s to RTI events
ment. The prototype has been tested using the benchmark (via the PublicOb j ect scheme described below), for

Proceedings of the Sixth IEEE International Symposium on Cluster Computing and the Grid Workshops (CCGRIDW'06)
0-7695-2585-7/06 $20.00 © 2006 IEEE
 Events from local schedule
and external buffer modlfy

Local events tngger

HLA update messages

HLA REPAST 3 HLA REPAST 4

Figure 2. An HLA-R~PAs'IFederation Figure 3. The HLA-R~PASTModel-Executive

Interface (A Single HLA-REPAST Federate)

conflict resolution and for inlegrating h e REPASTschedul-

ing system with the RTI (as depicted in figure 3). More
information about the HLA-REPAS'[ system can be found the sinlulation starts, the Client Federate Ambassador Ser-
in [6]. vice of each federate will be initialised and registered with
the corresponding Proxy RTI Ambassador Service.
3.3 Architecture of HLA-GRIDRePast During the simulation, the events and state changes
generated by the simulation model will be translaled into
Figure 4 presents a conceptual view of the federate ex- the corresponding RTI function calls by the underlying
ecutive based on HLA-GRID-REPAST. Each federate ex- HLA-REPAST middleware. Then the RTI calls will be
ecutive basically consists of two major parts, one on the passed to the Client RTI Ambassador, and it will translate
client side and another on the proxy side, which usually these function calls orice again into Grid service invocations
run on different machines. The client slde contains the to access the remote Proxy RTI Ambassador Service on the
following modules: the REPASTagent-based simulation proxy side.
model, the HLAREPAsT middleware, the Client RTI Am- Finally, the Proxy RTI Ambassador Service will inter-
bassador and the Client Federate Ambassador Service from act with the red RTl by executing the real RTI calls w ~ t h
HLA-GRLD. These components usually run on a local ma- respect LO the client side. The return values will be sent
chine (from the simulation model's point of view). back as the return value of the GRlD service call while the
On the proxy side, there are the Proxy RTI Ambassador runtime exceptions of Ille RTI calls will be sent back by
Service and the Proxy Federate Ambassador which inter- means of Apache Axis faults, which will be described later.
act with the real RTI hosted by a remote machine. Both On the other hid, callbacks from the RTI to a federate
the Proxy RTI Ambassador Service and the Clien Federate are translated into invocations of the Client Federate Am-
Ambassador Service are implemented as GRlD services on bassador Service by the Proxy Federate Ambassador. The
different sides, shown as the round rectangle boxes in Fig- Client Federate Ambassador Service will convey these call-
ure 4. The modules on both sides are coupled together to backs to the HLAREPAs?' middleware, and the latter will
fonn a single federate executive. then convert the federate calls into REPASTevents and put
Before the simulation starts, the Proxy RTI Ambassador hen1 inlo the event scheduler.
Services should have been started, each federate is asso- Modifications have been made to both HLA-GRID
ciated with a Proxy RTI Ambassador Service. The identity and HLA-REPAST m order to mtegrate them into
of each federate's Proxy RTl Ambassador Service will be HLA-GRID-REPAST. Two major modifications are con-
passed to the corresponding Client RTI Ambassador. When cerned with object encoding and remote exception hdi-

Proceedings of the Sixth IEEE International Symposium on Cluster Computing and the Grid Workshops (CCGRIDW'06)
0-7695-2585-7/06 $20.00 © 2006 IEEE
 A Conceptual HLA-Orid-Repast 3.3.2 Exception Handling
Federate Executive
In addition, remote exception handling has been introduced
in HLA-GRIDREPAST. In HLA-GRIDREPAST,Java
exceptions generated in Grid service calls are packed into
Apache Axis Pdults and passed back to the client side. The
RTI implementation used in HLA-GRID-REPAST is the
DMSO RTI NG 1.3. In this implementation, Java excep-
tions work as return values of RTl function calls in many
siludtions. HLA-REPASTmakes heavy use of such func-
tions. In Inany cases, the exceptions (return values) are dis-
carded immediately without further handling. This behav-
iour causes very small overhead when the simulation fed-
erates and the RTI ambassadors execute on the same ma-
chine. This is not the case in HLA-GRID, however, where
the federate simulations and the RTI ambassadors are in dif-
ferent machines connected via a Grid. In this case, excep-
tions raised by RTI calls waste substantial network band-
width. To address h i s issue, in HLA-GRID such excep-
tions are registered with the RTI Ambassador Services after
connection between the client simulation and RTI Ambas-
sador Service is established through the RTI Proxy Ambas-
sador. When RTI calls generate exceptions, the RTI Am-
bassador Service will only pass exceptions that are not reg-
dling. The former n required by the SOAP protocol while istered back to the client simulation. In this way, registered
the latter is designed for improving performance. RTI d l exceptions are handled remokiy at Ihe proxy side.
The perfailnance advantage of remote exception handling
is substantial, and is demonstrated in the next section.
33.1 Object Encoding
3.4 Evaluation of HLA-GRID-REPAST
The object encoding scheme has been modified to coop-
erate with HLA-GRID. HLA-REPAST uses a l'ublicob- To evaluate the robustness and performance of
ject scheme to translate Java expressions into HLA Sunc- HL A-GRID-REPAST we have developed an agent-
tion calls [6]. Java objects in the Publicobject scheme based federation using TILEWORLD [8] as a test case.
may be transferred between federates during a simulation. Tileworld is a well established testbed for agents. It consists
In HLA-REPAST,the Java objects are encoded into Java of a grid-like environment consisting of tiles, holes and
Byte arrays. However, Java Byte arrays cannot be used obstacles, and on or more agents which attempt to score as
directly within HLA-GRID, which, like other grid ser- many points as possible by pushing tiles to fill in the holes
vices, uses SOAP as the communication protocol between (figure 5). For HLA-GRIDAEPAsT, a TILEWORLD
clients and servers. SOAP is based on XML which does simulation developed for HLAREPAsT has been used, as
not support binary data. The problem can be solved by described in [6]. In that inlplementation, the environment
modifying HLA-GRID to support either the SOAP With of Tileworld containing the tiles, holes and obstacles of
Attachnlenrs (SwA) proposal 111 or the WS-Atrachmnis the model is simulated by a s~nglefederate while the
proposal [7]. Such a11 approach would require mnodifica- TILEWORLD agents are simulated by one or more agent
tions to the Globus Toolkit and would affect the interop- federates. In general, the simulation will contain one
erability of HLA-GRID-REPAST. Instead of modifying environment federate and at least one agent federate.
the Globus Toolkit, a new object encoding scheme for Java Experirne~italresults have been obtained by executing
objects is employed in HLA-GRIDREPASTas follows. simulations over a Grid in a LAN environment and a WAN
First, the Java object is encoded into a Java byte array as in environment between the UK (Birmingham) and Singapore
HLA-REPAST. Then, the base64 encoding scheme (from as shown in figure 6. The performance results obtained have
the Apache project) is used to encode the Java byte array been presented and discussed In [13J and they are briefly
into a Java String Object. Finally, a Java Byte array is gen- summarised here.
erated from the Java String and pa~sedto HLA-GRID for Our experiments in a LAN environment with one
encapsulation in a SOAP message. agent federate simulating different number of agents

Proceedings of the Sixth IEEE International Symposium on Cluster Computing and the Grid Workshops (CCGRIDW'06)
0-7695-2585-7/06 $20.00 © 2006 IEEE
 ---.,
NTUlPDCC cluster
-<--

Figure 5. A 20x20 TILEWORLD

Figure 6. TILEWORLD Configuration for
HLA-GRIDXEPAS'C
showed that the perf'ormance of HLA-GRID-REPAST
with local exception handling is lower compared with
HLA-GRID-REPAST with remote exception handling.
When the number of agents is small (less than 112 in our
experiments), HLA-GRID-REPAST exhibits a higher over- main knowledge and expertise, at different locations, de-
head compared lo HLA-REPAST.As (he number of' agents velop, modify, assemble and execute distributed sin~ulation
increases and computation load becomes the dominant h c - components over the Grid. The required components re-
tor, the total simulation time for HLA-GRID-REPAST side in the Grid as Glid services which the system will au-
is relatively close to thal of HLA-REPAST. The distri- lomatically discover, semantically match (with each other
bution of 128 agents in multiple federates (1 to 16) in and with the simulation goals) and return to the user.
HLA-GRIDBEPAST yields speedup patterns similar to However a number of important new research challenges
HLA-REPAST with the relative overhead slightly increas- have to be addressed before this vision is redised, includ-
ing as the cross Grid communicatioi~increases (i.e. as the ing support for collaborative development of simulation ap-
number of federates increases). plications, advanced model and service discovery mecha-
Experiments in a WAN environment have shown nisms, novel resource management, mechanisms for fault-
that HLA-GRIDREPAST performs much worse than tolerant, coordinated, secured simulation executions, and
HLA-REPAS~C compared with the performance difference load balancing mechanisms to meet the different require-
in a LAN environment. Further experiments are currently ments of the different models.
being performed to further analyse the communication cosls
As a testbed to achieve DS-Grid's objectives, this pa-
in the WAN environment.
per has presented HLA-GRID-REPAST, a prototype plat-
form for executing large scale agent-based simulations over
4 Summary Grids.

The emergence of' Grid technologies provide exciting

new opportunities for large scale distributed simulation, en-
abling collaboration and the use of distributed computing 5 Acknowledgements
resources, while also facilitating access to geographically
distributed data sets. In the last few years, there has been
an increasing interest in t<kingadvantage of Grid technolo- A number of people have helped in many different ways
gies to execute HLA simulations over Che Internet. Con- to make this work possible. Ln particular we would like to
tributing Lo this effort, the DS-Grid e-Science project en- thank Ben Stone, Steve Pillinger, Jim Nichols, Irene Goh,
visages the development of a distributed collaborative sim- Tan Joo Sing and Mike Lees. This work is funded by the
ulation environment where reseuchers with differe~ltdo- UK e-Science Programme.

Proceedings of the Sixth IEEE International Symposium on Cluster Computing and the Grid Workshops (CCGRIDW'06)
0-7695-2585-7/06 $20.00 © 2006 IEEE
 References
[I] J . Barton, S. matte, and H. F. Nielsen. SOAP messages with
atlachments. W3C Note, Dec. 2000.
[2] W. Cai, Z. Yuan. M. Y. H. Low. and S. Turner. Federate 1ni-
gration in HLA-based simulation. F~rtureGenerzaliun Com-
puter Sysretns, 31(1):87-95, Jan. 2005.
[3] N. Collies. Repast: An extensible framework for agent simu-
lation. http: //repast.sourceforge.~~et/docs/
repast\-intro\-final.doc,June 2003.
[4] K. lskr;ra,G. Albada, and P. Sloot. Time Warp cancellations
optirnisations on high latency networks. In Prr)ccerling of'
tl?e 7th IEEE Itzter-nationc~lS.ytnpositrn7 on Disrribut~dSiinlr-
brioit at7d Keal-?i'r?7eApp1icntion.s (DS-K?"03),pages 128-
137,2003.
[5] W. Jie, T. Zang, Z. Lei, W. Cai, S. Turner, and L. wmg. Con-
structing an OGSA-based GI-idcomputing platfornl. In In-
/ei-11~tiotz(11
C o i z f e ~ ~ r on
~ c eScienlific cmd Engineering Com-
putation (IC-SEC 2002), pages 738-741,2002.
[6] K. Minson and G. '~heodosupoulos. Distributing Repast
agent-based sin~ulalionswith HLA. In Europecrr Siinula-
tion Inreropernbili~Worhxkop 2004, pages 04E-SIW-046.
June 2004.
[7] H. F. Nielsen, E. Christensen. and .I. Fmell. Specifi-
cation: Ws-attachnlents. http: / /www- 106 .ibm.
com/developerworks/webserv~ces/library/
ws-attach.html,June 2002.
181 M. 8. Pollack and M. Kinguette. Introducing the 'lile\vorld:
Experimentally evaluating agent architechres. In Notione11
Confewnce on Artijiciczl Intelleyence, pages 183-1 89, 1990.
[9] J. M. Pollen, K.Brunton, D. Bri~trman,D. Drake, M. Hieb,
K. Morse, and A. Tolk. Using Web services to integrate
heterogeneous simulations in a <;rid environment. Future
Gerzer-crrionCornpurer Sysie~ns,2 1:97-106,2005.
[I01 R. Tobias and C. Hofmann. Evaluation of free Java libraries
for social-scientific agent-based simulalion. Jorttnal ofArri-
JciuE Soc.i~~lie.surld Soc.ial Simrtlcrfion, 7( 1 ), Jan. 2 0 4 .
[ I I] S. van Spluntel-, N. J. E. Wijngaards, F. M. 1'. Brazier. and
D. Richards. Automated component-based configuration:
Promises and fi~llacies. 111 PI-oce~clingsof the A d q t i v e
Ayetlrs and Multi-Agent Sy.stetn.c r,.olbhop or the, AISB 2004
Syr~~posiutn, pages 1 3 G135,2004.
1121 Y.Xie, Y. M. Tea, W. Cai, and S. 'l'umer. Service provision-
ing for HLA-based distributed simulation on the Grid. In
Proceedblgs o f ' r h Ninc,teerlth
~ ACM/lbx8E/SCS Work.shop on
Princip1e.s ($Ad\~rincc~dand Di.m-iblrred Sirntrlrrtion (PADS
2005) (fi)rnz~rly called Pornlie1 and Disrributerl Simulntion).
pages 282-29 1, Monterey, CA, USA, June 2005.
[13] Y. Zhang, G . Theodolapoulos, R. Minson, S. Turner, W. Cau,
X. Yong, and S. Logan. Grid-aware large scale distributed
si~nulationof agent-based systems. In 5td Europectrz Simulu-
ror Inreroper-ability Workshop (Euro-SI W), pages 03-SlW-
047. Toulouse, France, June 2005.
[I41 W. Zong, Y. Wang, W. Cai. and S. Turner. Grid services and
service discovery for HLA-bitsed distr-ibuted simulations. 111
The 8/h IEEE Inlen1-1~tionerl Synl1~osilcmon Dislrib~ileclSirn-
~tlcrtioncrnd Recil Time Applicutions (DSRT 2 W ) , pages
1 16- 124,2004.

Proceedings of the Sixth IEEE International Symposium on Cluster Computing and the Grid Workshops (CCGRIDW'06)
0-7695-2585-7/06 $20.00 © 2006 IEEE