Using a shot clock to design an efficient parallel distributed simulation

Parallel discrete event simulation represents an important area of research because large simulations are prohibitively expensive to execute on a sequential machine. Though simulation programs seem to have substantial amounts of parallelism, workable techniques to efficiently parallelize these programs have been elusive. In this paper, we show that by increasing the granularity of computation that each processor performs, the degradation in performance due to expensive communication can be mitigated. We simulate a network traffic flow problem using PVM to construct our parallel system. To further reduce the cost of communication, we use a novel windowing technique called a shot clock. The initial value of a shot clock is a lookahead value and expiration of a shot dock triggers actions which dictate the granularity of the messages passed in the system. The finest grained message is a single car and the coarsest grained message is a queue of cars. We achieve a speed up of 6 using PVM on a network of 16 workstations. Our experiments show that we can achieve better speed up using a system with faster communication.