使用GPU加速BLAST算法初探

A Preliminary Exploration on Parallelized BLAST Algorithm Using GPU

应用GPU通用高性能编程技术实现了一种加速BLAST算法的新方法。BLAST是目前最常用的用于生物序列查询比对的算法和软件包,其处理速度受到串行化执行和磁盘I/O等因素的影响。本文通过实验分析了BLAST软件包中的典型程序BLASTN的运行热点,并选定关键热点模块,应用CUDA编程技术对其进行并行化改造。对比实验结果表明,对于平均序列长度较大的序列库,应用GPGPU并行化可明显缩短该模块的运行时间,获得超过35倍的加速比。这说明,我们可以利用GPGPU对BLAST进行并行化加速,以满足高性能生物序列查询的需求。

In this article we present a novel approach to accelerating the BLAST algorithm by using the GPGPU technology. BLAST is the most widely used algorithm and software package for biological sequence search. It is, however, limited by serial process and heavy disk I/O operation. We analyze BLASTN, a typical BLAST tool, identify the most important hotspot of BLASTN, and use CUDA to reprogram the hotspot process. Our test results show that more than 35 times speedup has been achieved in the GPU BLASTN as compared to the CPU counterpart, which demonstrates the significance of parallelizing the BLAST algorithm using the GPGPU technology.