MRI: Acquisition of a Computing Cluster for Portland Institute for Computational Sciences

Portland State University, Portland OR

Investigators

Abstract

This research enterprise acquires, deploys, and maintains a modern heterogeneous cluster for scientific computing. The cluster combines the latest multi-core processors with the new MIC (Many Integrated Core) architecture to provide a testbed for development of new parallel mathematical algorithms. Multi-institutional access to the equipment, managed through a new center called the Portland Institute for Computational Sciences, will catalyze interactions within a regional user base of computational scientists. This equipment will considerably strengthen open, state-wide high performance computing access in Oregon. The equipment aims to profoundly impact research into fundamentally new techniques in computational mathematics. Highly scalable dimension reduction techniques that exploit existing parallelization tools developed at US national laboratories will be designed for evolution problems posed in spacetime domains. For hyperbolic problems on complex spacetime structures, a new class of mathematically sound methods, well-suited for thread parallelism on MICs, will be studied. New algorithms to compute eigenmodes of partial differential operators, based on parallelizable approximations of operator-valued contour integrals, are considered. Another project will use massive medical data sets to validate novel statistical inferential methods leading to valuable templates of disease progression. The equipment will be extensively used to develop and run numerical simulations of convection in Earth's mantle. Further catalogued examples of projects show that the equipment's impact will be regional and beyond Oregon. Educational and training activities will promote the learning of fine-grained parallelism critical to exploit new and emerging energy-efficient processors.