Applications of Mobile-Agent Techniques
		in Large-Scale Distributed Simulations

Malcolm Corbin
Systems Integration Department
DRA Farnborough
GU14 6TD
United Kingdom
Tel: +44 (1252) 392757
Fax: +44 (1252) 394700
email:   mjcorbin@dra.hmg.gb

This project started in 1993 with the aim of investigating 
the potential of mobile-agent technology to improve the 
organisation of large distributed simulations involving 
some thousands of interacting entities. The recent work has 
concentrated on specifying a set of generic roles for 
agents within an overall simulation architecture (Ref 1). 
These roles fall into two main categories:
	Communications management,
	Situation monitoring and analysis.

The primary problem in communications management is that of 
discovering the existence of other relevant entities 
without having to perform a global search for them through 
an indefinitely large system. The approach we have adopted 
is to assign to each model an autonomous mobile 
"Registration Agent" which can move across a network of 
servers to perform a sorting of the entities according to 
their geographical position in the virtual simulation 
space. Once these have established themselves in the right 
locality each will be close to the registration agents for 
the other relevant entities, and the discovery process can 
be completed by an exchange of "Attention" and "Interest" 
Agents, whose characteristics can be tailored to the nature 
of the interaction being modelled. An extension of this 
work would be to control the localisation of the computer-
generated models themselves within the simulation network, 
allowing their positions to be changed dynamically so as to 
minimise the overall communications loads of the system. 

Distributed systems for monitoring and analysis of events 
will also be required in large-scale simulations, since 
there will no longer be any global data exchanges which, in 
current systems, allow these functions to be performed from 
a single site. A system based on the use of mobile agents 
has the advantage of flexibility, in being able to perform 
at remote sites the arbitrary data-interpretations and 
summarisations needed to reconstruct  views of the state of 
the simulation which are useful in an analysis.  Such 
"Collation Agents" could be constructed to recognise 
particular patterns of behaviour, or track entities through 
the system, reporting back to their originator.

The next stage will be to move towards prototype 
implementations of some of these ideas. Initial work has 
been done in collaboration with Surrey University, using 
the WAVE mobile code language (Refs 2, 3, 4), but other 
types of agent support systems are being investigated.

References

1.  Corbin M J, "An Agent-Based Architecture for the 
Support of Large-Scale Simulations", 10th European 
Simulation Multiconference, ESM'96, Budapest, June 1996, p 
1105.

2.  Corbin M J and Sapaty P S  qUsing the WAVE paradigm for 
parallel simulation in distributed systemsn  (Joubert, 
Trystram, Peters & Evans eds.) Parallel Computing: Trends 
and Applications, Proceedings of ParCoM93, Parallel 
Computing, Grenoble, September 1993, North Holland, p 501.

3.  Borst P M, Corbin M J and Sapaty P S qWAVE Processing 
of Networks and Distributed Simulationn, High Performance 
Distributed Computing-3 Conference, San Francisco, August 
1994, pp 61-69.

4.  Corbin M J and Sapaty P S qDistributed Object-Based 
Simulation in WAVEn,  Journal - Simulation: Practice & 
Theory, Vol 3, No.3, pp 157-181, Nov 1995.