MRI: Acquisition of a 40-core Linux cluster for oceanographic research at Earth & Space Research

点击翻译按钮获取中文摘要

Funds are provided to purchase a Multi-Processor, Linux-based Cluster ("MPC") to improve ESR's (Earth and Space Research) capacity to carry out highly processor-intensive computations for a variety of oceanographic research applications. ESR scientists presently conduct their research using networked, single-processor PCs. These facilities have fallen well behind the demands of their research interests. Access to the MPC would significantly enhance existing programs, and would open up research areas that are presently inaccessible to ESR. The proposed MPC is a 40-core (20 dual-core AMD Opteron processors) system arranged in 4 processors per node. The system scale is set by both budget and infrastructure (power and air conditioning) requirements. The MPC would be owned and operated by ESR and would be available to ESR scientists and, as load levels permit, to project collaborators at other institutions. The proposed system would significantly enhance the contribution of ESR scientists to studies of topical interest including: . tidal influence on ocean mixing and sea-ice properties in polar seas, and ice/ocean interactions at the base of Antarctic ice shelves; . analysis of the quasi-global and high-resolution distribution of upper ocean currents and mixing on daily timescales, and short-range forecast of the surface currents; and . the role of anticyclonic eddies in modifying convective ventilation. The MPC would assist in these studies by allowing expanded model runs to better represent processes that require extremely long run times on their existing single-CPU computers. Computationally demanding tasks include: modeling internal tides and density outflows in regions of steep topography (where fine grid spacing is essential); coupling tides to seasonal and longer changes in wind-forced and thermohaline flows (requiring longer integration times); explicitly representing sea ice in ocean and coupled ice-shelf/ocean models (a factor of 2-3 increase in CPU needs relative to ocean-only models); and increasing the sophistication of turbulence closure models for both high latitude and quasi-global upper-ocean studies.