The Satellite project was part of my Ph.D. at the Computing Science Department at the University of Glasgow. The goal was to investigate new techniques to exploit the parallelism inherent in high-level algebraic Petri nets and to speed up the simulation of such nets.
During this research, an innovative simulation framework made of two parts and running in a cluster of workstations on top of MPI has been built:
The first part deals with a distributed simulator. New techniques based on a spatial model decomposition have been developed. To maintain causal relationships among the events in the simulation, three different simulation engines can be used in a message passing environment: conservative, optimistic (Time Warp) and synchronous.
The second part deals with a concurrent simulator. A technique based on temporal model decomposition has been implemented. This is an alternative approach to the application of multiple processors to high-level nets discrete event simulation models in a (distributed) shared-memory environment.
Research has demonstrated that optimal partitioning into minimum grain size logical processes and the identification of portions of models that yield concurrent events can speedup the distributed simulator over the sequential one.