Optimization and Implementation Performance of Sequence Alignment on the Intel Xeon Phi-based Heterogeneous System

Abstract

The heterogeneous system based on different architectures becomes a convenient solution in the high performance computing research when facing the expanding sequence data in bioinformatics analysis. Intel Xeon Phi-based cluster is one of the most utilized heterogeneous systems in recent years. Without accurate results from sequence alignment, the remaining two steps in the variant analysis, variant calling and variant annotation, cannot achieve the correct consequence. However, most sequence aligners are developed facing the multicore system and cannot take advantage of Intel Xeon Phi-based cluster. This paper explored the implementation modes on the Intel Xeon Phi-based heterogeneous system, including native, offload and symmetric modes. We indicate that native mode cannot take advantage of Intel Xeon Phi-based cluster through the evaluation of scalability of various modes under sequence alignment. Although offload mode owns a promising future, it is not easy to enhance performance without comprehensive coding ability. Finally, the symmetric mode could provide a low complexity solution that supports significant improvements in performance.