Design of Pattern Matching Systems: Pattern, Algorithm, and Scanner

Abstract

Pattern matching is at the core of many computational problems, e.g., search engine, data mining, network security and information retrieval. In this dissertation, we target at the more complex patterns of regular expression and time series, and proposed a general modular structure, named character class with constraint repetition (CCR), as the building block for the pattern matching algorithm. An exact matching algorithm named MIN-MAX is developed to support overlapped matching of CCR based regexps, and an approximate matching algorithm named Elastic Matching Algorithm is designed to support overlapped matching of CCR based time series, i.e., music melody. Both algorithms are parallelized to run on FPGA to achieve high performance, and the FPGA-based scanners are designed as a modular architecture which is parameterizable and can be reconfigured by simple memory writes, achieving a perfect balance between performance and deployment time.