Efficient Pattern Matching on CPU-GPU Heterogeneous Systems

Abstract

Pattern matching algorithms are used in several areas such as network security, bioinformatics and text mining, where the volume of data is growing rapidly. In order to provide real-time response for large inputs, high-performance computing should be considered. In this paper, we present a novel hybrid pattern matching algorithm that efficiently exploits the computing power of a heterogeneous system composed of multicore processors and multiple graphics processing units (GPUs). We evaluate the performance of our algorithm on a machine with 36 CPU cores and 2 GPUs and study its behaviour as the data size and the number of processing resources increase. Finally, we compare the performance of our proposal with that of two other algorithms that use only the CPU cores and only the GPUs of the system respectively. The results reveal that our proposal outperforms the other approaches for data sets of considerable size.