XSW: Accelerating Biological Database Search on Xeon Phi

Lipeng Wang,Yuandong Chan,Xiaohui Duan,Haidong Lan,Xiangxu Meng,Weiguo Liu

DOI: https://doi.org/10.1109/IPDPSW.2014.108

2014-01-01

Abstract:In this paper we present XSW, a new parallel Smith-Waterman algorithm for searching protein sequence databases on the Xeon Phi coprocessor. In order to make full use of the compute power of the many-core Xeon Phi hardware, we have used a two-level parallelization scheme: the thread level coarse-grained and VPU level fine-grained parallelism to implement our algorithm. At the thread level, XSW employs multi-threading to implement the SIMD parallelism. At the VPU level, we have used the Knights Corner instructions to gain more data parallelism. We have also reorganized the database and made use of the parallel shuffling operations on Xeon Phi to achieve better I/O efficiency. Evaluations on real protein sequence databases show that XSW achieves the peak performance of 70 GCUPS on a single Intel Xeon Phi 7110 card. Compared to two other well parallelized Smith-Waterman algorithms: the multi-core CPU-based SWIPE and the GPU-based CUDASW++ 3.0, XSW achieves much better performance than SWIPE. And XSW achieves comparable performance but better accuracy than CUDASW++ 3.0. To our knowledge this is the first reported implementation of the Smith-Waterman algorithm on Xeon Phi. The executable binary code of XSW is available at http://sdu-hpcl.github.io/XSW/.