摘要

A number of research issues arise in executing large scale High Level Architecture (H LA) based distributed simulations. Among these issues distributed simulation cloning, fault tolerance and Grid enabled architecture are particularly important and challenging. This paper presents a Decoupled Federate Architecture as the underlying infrastructure to facilitate a solution to each of the above issues.
Distributed simulation cloning has been designed to improve the performance of "what-if' analysis, by enabling the examination of alternative scenarios concurrently within the same execution session. Furthermore, simulation federates running at different locations are liable to failure. Such risk increases with the scale of a distributed simulation. The architecture has also been exploited to form the basis of a generic framework to support runtime robustness. The development of complex simulation applications often requires geographically distributed huge computing resources and data sets. Using the architecture, Web and Grid technologies have been successfully combined with existing simulation technologies thus to enhance interoperability and execution flexibility for such simulations.
Benchmark experiments have been performed to study the extra overhead incurred by the Decoupled Federate Architecture against a normal federate. Encouraging experimental results indicate that a well designed architecture can achieve performance close to the normal one in terms of latency and time synchronization. Additional experiments have been presented for evaluating the three solutions supported by the architecture.