Performance modelling of parallel BLAST using Intel and PGI compilers on an infiniband-based HPC cluster.

摘要

The Basic Local Alignment Search (BLAST) is one of the most widely used bioinformatics programs for searching all available sequence databases for similarities between a protein or DNA query and predefined sequences, using sequence alignment technique. Recently, many attempts have been made to make the algorithm practical to run against the publicly available genome databases. This paper presents our experience in mapping and evaluating both the serial and parallel BLAST algorithms onto a large Infiniband-based High Performance Cluster. The evaluation is performed using two commonly used parallel compilers, Intel and Portland's PGI. The paper also presents the evaluation methodology along with the experimental results to illustrate the scalability of the BLAST algorithm on our state-of-the-art HPC system. Our results show that BLAST runtime scalability can be achieved with up to 87% efficiency when considering the right combination of the MPI suite, the parallel compiler, the cluster interconnect and the CPU technology.