Analysis of P systems simulation on CUDA

Abstract

GPUs (Graphics Processing Unit) have been con- solidated as a massively data-parallel coprocessor to develop many general purpose computations, and en- able developers to utilize several levels of parallelism to obtain better performance of their applications. The massively parallel nature of certain computa- tions leads to use GPUs as an underlying architec- ture, becoming a good alternative to other paral- lel approaches. P systems or membrane systems are theoretical devices inspired in the way that liv- ing cells work, providing computational models and a high level computational modeling framework for biological systems. They are massively parallel dis- tributed, and non-deterministic systems. In this pa- per, we evaluate the GPU as the underlying archi- tecture to simulate the class of recognizer P systems with active membranes. We analyze the performance of three simulators implemented on CPU, CPU-GPU and GPU respectively. We compare them using a pre- sented P system as a benchmark, showing that the GPU is better suited than the CPU to simulate those P systems due to its massively parallel nature.