Towards Adaptive Caching for Parallel and Discrete Event Simulation

We investigate factors that impact the effectiveness of caching to speed up discrete event simulation. Walsh and Sirer have shown that a variant of function caching (staged simulation) can improve the performance of simulation in a networking application (Walsh and Sirer 2003). Consider, however, that the effectiveness of a caching scheme depends significantly on cache size, the cost of consulting the cache, the cache hit rate, and the cost of completing the computation in the case of a cache miss. We hypothesize that adaptive techniques can be used to optimize caching parameters (e.g. cache size), and demonstrate an adaptive scheme that decides whether to utilize caching at an LP depending on observed cache performance and event processing times. This paper focuses on a quantitative evaluation of these relationships using our own caching implementation with the P-Hold synthetic workload application (Fujimoto 1990) running on the GTW simulation kernel (Das et al. 1994). Experiments show that as the cache size is increased, performance improves to a point, then degrades, and also that the adaptive technique can substantially improve speedup.