Parallel Simulation of Probabilistic P Systems on Multicore Platforms

Abstract

Ecologists need to model ecosystems to predict how they will evolve over time. Since ecosystems are non-deterministic phenomena, they must express the likelihood of events occurring, and measure the uncertainty of their models' predictions. One method well suited to these demands is Population Dynamic P systems (PDP systems, in short), which is a formal framework based on multienvironment probabilistic P systems. In this paper, we show how to parallelize a Population Dynamics P system simulator, used to model biological systems, on multi-core processors, such as the Intel i5 Nehalem and i7 Sandy Bridge. A comparison of three di erent techniques, discuss their strengths and weaknesses, and evaluate their performance on two generations of Intel processors with large memory sub-system di erences is presented. We show that P systems are memory bound computations and future performance optimization e orts should focus on memory tra c reductions. We achieve runtime gains of up to 2.5x by using all the cores of a single socket 4-core Intel i7 built on the Sandy Bridge architecture. From our analysis of these results we identify further ways to improve the runtime of our simulator.