| Summary: | As the applications being modeled by discrete event simulation become more complex the resources required for these simulations become prohibitively large. Distributed simulation offers a promising solution to this problem, but creates a new set of issues and trade-offs. === This thesis focuses on distributed simulation algorithms based on a conservative approach with clustered processes. A multiprocessor based, distributed simulation testbed PARADIS, has been designed and implemented on a hypercube iPSC/2. Using this testbed, the performance of the TNE/DLTNE algorithms is investigated. A distributed knot detection algorithm and a distributed artificial deadlock algorithm were developed in order to detect, respectively, inter-processor deadlocks and artificial blocking. The later algorithm avoids PARADIS having to call the knot detection algorithm too quickly. === Empirical performance evaluation of PARADIS showed that a significant speedup over a serial simulation might be obtained. The experiments showed also that the number of processes that are assigned to a single processor and the lookahead play an important role in the speedup of the simulation.
|
|---|
