Parallelization and characterization of pattern matching using GPUs

Abstract

Abstract

A bit-parallel algorithm for searching multiple patterns with various lengths

“…Work [13] GeForce GTX 480 30 n/a 50,000 20 Oh [31] GeForce GTX 480 3-15 3,180 34,915 n/a Zha [8] Tesla GT 200 9 904 33 17 Tumeo [12] Four Tesla C2050 12-48 100 190,000 16 Tumeo [9] Two Xeon X5560 3-20 100 190,000 16 AC-opt [17] Cray XMT 28 100 190,000 16 AC-opt [16] Cell/B.E. 3-4.5 4 20,000 8 Schatz [6] GeForce 8800 GTX 0.4-0.5 1* 312,500 800 Michailidis [15] FPGA 0.1 30 1 1,024 Singaraju [32] FPGA 3 n/a 1,237 13 *: 1 base pair = 2 bits complexity of the SSEF algorithm ranges from O(nm) to O(n/m) according to preprocessing effects, where n is the length of the text and m is that of the pattern.…”

“…Vasiliadis et al [13] presented a DFA-based algorithm for inspecting network packets in real-time. They designed an efficient packet buffering scheme for transferring packets from the main memory to the video memory.…”

“…Thus, MIMD machines are more tolerant to the irregularity than SIMD machines, because the former machines can issue different instructions to different processing elements. The DPAC algorithm was deployed on many parallel machines such as the graphics processing unit (GPU) [4,5,8,9,11,12,13], field programmable gate array (FPGA) [14,15], Cell Broadband Engine [16], and supercomputer [17]. Similar to the AC algorithm, which has the cost O(z) of printing the output, the DPAC algorithm has different time complexities for best and worst cases.…”

Unknown scale objects recognition using massively parallel processing with CUDA

“…Problems of irregular objects recognition [4], [5] or pattern matching [11], [12] are not so popular regarding CUDA. This paper presents such a solution for the Hough transform spanned on irregular objects.…”

Parallel String Matching

“…Kouzinopoulos and Margaritis evaluate the performance of GPU implementations of the classical matching algorithms [18] and report on a possible speedup of more than 10x. Vasiliadis et al [7] and Cascarano et al [26] present solutions for regular expression matching in GPUs, which is a superset of the string maching problem. These approaches create finite state machines from the input patterns and execute them in parallel on partitioned input data.…”