TIB: A Time Management Algorithm of PDES for Automatically Detecting Concurrency

Exploiting concurrency is very important for improving the performance of parallel discrete event simulation. In order to exploit more concurrency, a new time management algorithm for parallel discrete event simulation running on multi-core computers is proposed in this paper. It is based on time-index-buffering variable data structure. This structure is used to buffer variable value to increase parallelism. Combined with control algorithm, it can capture before-and-after relationships and detect concurrent events automatically. The new algorithm is robust and with well-organized code. Theoretical analysis shows that in some situations it delivers more concurrent events than traditional conservative synchronization algorithms. The performance is evaluated on multi-core computers. Experimental results demonstrate that if most events in the simulation have heavy computations in contrast to interactions, the new algorithm outperforms the traditional conservative algorithms. A real world application exhibits that the new algorithm is very useful for irregular problems.