Parallelization and Dynamic Code Generation for High-speed Regular Expression Matching

Abstract

Pattern matching using regular expression is widely used in computer science, various method has been studied so far. Also a method to convert regular expressions into DFA, is one of them. In this paper, we propose two methods for throughput improvement of pattern matching. The first method is the parallel matching. We defined Simultaneous Finite Automata (SFA) by the natural extension of automata theory, that can be run in parallel against each partition of target string. The second method is dynamic code generation, that generates optimized native code from DFA・SFA at run time. We have implemented the regular expression matcher based on these two methods and it has achieved speedup by a factor of 5.9 on a 6-core environment.