Performance improvement of the parallel smith waterman algorithm implementation using Hybrid MPI-OpenMP model

This paper applies the hybrid parallel model that combines both shared and distributed memory architectures to improve the performance of the Smith waterman algorithm (SW). The hybrid model uses both MPI and OpenMp as programming techniques for different memory architectures. Our improved implementation executes a parallel version of SW algorithm with a row wise computation of the alignment matrix, which mainly optimizes the memory usage. We applied the parallel SW implementation and tested the system scalability on a homogenous cluster of up to eight nodes each of twenty four cores. We used the SWISS-PROT protein knowledgebase to test our implementation which achieved a tremendous reduction in the running time using the Hybrid MPI-OpenMP over the OpenMP and sequential implementations. The Hybrid MPI-OpenMP achieved a speed up of 14X and 50X over the OpenMP and sequential implementations respectively when tested against all the SWISS-PROT protein knowledgebase entries.