A massively parallel framework using P systems and GPUs

Abstract

Since CUDA programing model appeared on the general purpose computations, the developers can extract all the power contained in GPUs (Graphics Processing Unit) across many computational domains. Among these domains, P systems or membrane systems provide a high level computational modeling framework that allows, in theory, to obtain polynomial time solutions to NP-complete problems by trading time for space, and also to model biological phenomena in the area of computational systems biology. P systems are massively parallel distributed devices and their computation can be divided in two levels of parallelism: membranes, that can be expressed as blocks in CUDA programming model; and objects, that can be expressed as threads in CUDA programming model. In this paper, we present our initial ideas of developing a simulator for the class of recognizer P systems with active membranes by using the CUDA programing model to exploit the massively parallel nature of those systems at maximum. Experimental results of a preliminary version of our simulator on a Tesla C1060 GPU show a 60X of speed-up compared to the sequential code.