Fast Pattern Matching on the Cell Broadband Engine.

AbstractPattern-matching algorithms, which are essential to intrusion detection and virus scanning applications, typically only make limited use of the Single Instruction Multiple Data (SIMD) capabilities available in new generations of general-purpose processors. This paper presents the initial results of a study to increase the SIMD exploitation by pattern-matching schemes consisting of a novel vectorized state-machine implementation that is able to utilize the vector-processing units in the Cell Broadband Engine almost fully for all its processing steps by storing most of the data structure in the large internal register sets. The implementation provides an extremely deterministic aggregate processing rate of 6.7 Gb/s for a single vector unit, which can be scaled up to 50 Gb/s for one Cell Broadband Engine and up to 100 Gb/s for a blade for small pattern sets. It also supports configurations in which the scan rate can be partially traded off for increasing the number of patterns supported.