蛋白质序列比对算法在众核结构上的并行优化

在生物信息学中,蛋白质序列比对是最为重要的算法之一,生物技术的发展使得已知的序列库变得越来越庞大,这类算法本身又具有计算密集型的特点,这导致进行序列比对所消耗的时间也越来越长,目前的单核或者数量较少的多核系统均已经难以满足对计算速度的要求.Godson-T是一个包含诸多创新结构的众核平台,在该系统上实现了对一种蛋白质序列比对算法的并行化,并且结合蛋白质比对算法以及Godson-T结构的特征,针对同步开销、存储访问竞争以及负载均衡3个方面对算法进行了细致的优化,最终并行部分整体也获得了更优的、接近线性的加速比,并且实际性能远远优于基于AMD Opteron处理器的工作站平台.

Efficient Parallelization and Optimization of Protein Sequence Comparison Algorithm on Many-Core Architecture

In bioinformatics, a protein sequence comparison between two banks is one of most important algorithms. The sequence bank size is becoming larger and larger with the development of biotechnology, while the algorithm is also computation intensive. This leads to more and more consumption time and the single processor or multicore system, with only a few cores, are not powerful enough to reach a satisfying speed nowadays. Godson-T is a new kind of many-core architecture with lots of novel features. The parallelization of a protein sequence comparison algorithm on Godson-T is implemented. At the same time, the algorithm structure and architecture features of Godson-T are combined, and some optimization in three aspects are made: synchronization overhead, memory access contention, and load balance. The result shows that a close to linear speedup is obtained, and the performance is much better than that of the workstation platform based on the AMD Opteron processor.