high performance computing on graphics processing units: hgpu.org

Spam has become a more and more serious problem as the wide use of E-mail. Spam filtering based on mail content is a mainstream technology to solve the spam. However, the efficiency of spam filtering algorithm is becoming a bottleneck when it is used in the training of a great amount of mail samples or on the high load mail server. With superior computing power and high memory bandwidth, GPU (graphics processing units) have become a commodity multiprocessor platform for general-purpose high-performance computing. In this paper, we present several novel design and implementation of parallel algorithms for spam filtering based on GPU. Our algorithms utilize the high parallelisms well as the high memory bandwidth of GPU, and use parallel computation to effectively reduce memory access latency. The results of lab experiment have shown that the GPU based spam filtering solution has 14X speedup over the classical solution when it deals with large-scale spam.