Parallel Discrete Event Simulation of Stochastic Reaction and Diffusion Using Reverse Computation

Abstract

Parallel discrete event simulation consumes massive memory, involves frequent allocation and deallocation, and generates huge amount of data, which makes the memory usage an important issue for performance. In this paper, we present our reverse computation approach in our multi-threaded parallel discrete event simulator used for large scale stochastic reactiondiffusion simulation. We compute the reverse control code to record what operation an event did on the state of the host Logical Process during forward processing, and undo the operation according to the corresponding code during recovering that event.As the results on two versions of calcium wave model show, the reverse computation version achieved better performance than the state-saving version-it saved about 33% and 36% execution time for the two models respectively when two processing threads were used.