Accelerating pairwise statistical significance estimation using NUMA machine

Pairwise statistical significance (PSS) has been found as a promising alternative to database statistical significance for homology detection. However, the estimation of PSS is both computationally and data intensive, which poses a big challenge on performance and scalability in terms of distributed computations. In this work, we develop the parallel accelerator for the estimation of PSS using Non-Uniform Memory Acess (NUMA) machine, which is implemented in C++ using OpenMP, MPI and hybrid paradig-ms, respectively. Through distributing the computation kernels of the estimation of PSS procedure across multiple cores, we achieve up to 22.65× speedup using 24 CPU cores compared to sequential implementation.