Design And Implementation Of The CUDA-based Regular Expression Matching System

With the significant raise of influence of the network in public life, reinforcement of network content management plays a crucial role in constructing the healthy and harmonious network environment. String matching is one of the key trechnologies of network content management. Among the many implementations of string matching, regular expression is used widely because of its rich expressive power, easy to use and good expansibility. But, it's common to use software to implement the regular expression. With the continuous improvement of network bandwidth, its processing performance has increasingly become a bottleneck restricting the development of related applications. While the matching system based on the special hardware paltform such an FPGA and ASIC etc. is hard to collaborate the hardware and software and poor expansion, it also can't satisfy effectively the need of network content management. Therefore, looking for a high-speed string matching method fusing performance and flexibility arouses wide concern of industry.This paper proposed a CUDA-based regualr expression matching system which solved matching performance and scalability problem well and collaborated the hardware and software easily. Therefore, the system is a more practical solution to realize network content management. The contributions of this paper includes:○1 The entire matching process is divided into serial and parallel parts, and then tranplant the parallel part to GPU with the CUDA API and Class C provided by CUDA.○2 Consider a kind of parallelling pattern matching algorithm suited for transplantation, and realize the algorithm. On the one hand, using GPU's multi-threaded parallel processing hide memory access latency, the other hand, the best use of GPU's paralleled computing power to accelerate powerful pattern-matching process, improve the performance and efficiency of the matching system.○3 In the CPU-side segmente the file to be matched so that each thread has the same payload. The test demonstrates that the CUDA-based regualr expression matching system is faster than regualr expression library.