The objective is to study how good or bad distributed simulation is for the simulation of a few specific systems; these systems are queuing networks, timed petri nets, and logic circuits. Regarding queuing networks and logic circuits, the performance of three well known distributed simulation schemes will be studied. Regarding timed petri nets, new distributed simulation algorithms will be developed and their performance studied. This approach aims to answer several questions that have remained unanswered previously - relationship between overhead and performance, ideal speedups that one can expect, and relationship between the simulator parameters (such as delays on communication lines) and performance. The essential elements of the approach are (i) simulate the distributed simulator instead of directly implementing it on a machine, which makes it easy to vary parameters such as communication delays, computation times etc. (ii) study the relationship between overhead and performance by varying computation and communication times for overhead messages, and (iii) define and measure certain notions of ideal speedup that are more realistic than simply the number of processors. The immediate benefit of this research would be to provide a better understanding of several performance issues, e.g., what is the cause of low performance -- too many overhead messages or lack of potential parallelism in the system? It is commonly believed that overhead is the main problem. This is not true in many cases. In the long term this will lead to high performance distributed simulators custom designed for specific classes of systems.
