Computer Science Postdoc
San Francisco Bay Area
I customized the HMMER biological search algorithm to run efficiently on NERSC-8 (Cori) HPC architecture. This optimization targets the XeonPhi Knights Landing architecture via careful control of vectorization, cache management, and threading.Modifying the file access pattern and threading model of HMMER3 has produced a 16x speedup for naive usage, and 2x speedup over the previous best practice (manual division of input files for multiple concurrent background jobs). My customized version is currently in productive use at the Joint Genome Institute where it will save 3.5 million CPU hours annually. This work was published and presented at the HICOMB workshop, held at IPDPS 2018 in Vancouver, Canada.My second project improved performance of the MPAS Ocean Core on the NERSC Cori Supercomputer. The optimization strategies focused on more efficient OpenMP barrier behavior and recycling data structures used for halo exchange communication. The resulting application was 15% faster at production scale. Results were presented at the IXPUG Fall 2018 meeting in Hillsboro, Oregon.