MRI: Collaborative Research: Acquisition of an IBM BlueGene/L Supercomputer

点击翻译按钮获取中文摘要

This collaborative project among University Corporation for Atmospheric Research, NCAR (04-21498 Loft), University of Colorado-Boulder (04-20873 Tufo), and University of Colorado-Denver (04-20985 Mandel), addressing the technical obstacles to achieving practical petascale computing in geoscience, aerospace engineering, and mathematical applications, aims at acquiring a one-cabinet BlueGene/L supercomputer with 2048 compute processors, 64 service processors, and 20 terabytes of storage. Supporting research projects in atmospherics, wildfire modeling, aircraft simulation, and high performance numerical algorithms, the project studies hardware, software, and application issues. A team composed of atmospheric scientists, computational scientists, engineers, applied mathematicians, and experts in mapping applications to highly-parallel systems proposes a tight coupling among computer architects that analyze the performance of computational kernels on new designs (often in isolation) and atmospheric and engineering modelers and designers of scalable solvers proposing new algorithms (often without direct feedback form the computer architects) to extract the highest performance from the next generation of supercomputers. Part of the team has worked with the IBM Watson BlueGene/L development team during the past year and has had access to a BlueGene/L prototype with 512 nodes. Thus, this tightly-coupled group of 12 scientists working together on computational problems, simultaneously considering architectural, algorithmic, and modeling aspects of the problems offers special value to the work.Broader Impact: The project is expected to accelerate the long-term rate of progress in computational science, which can have a profound impact on the understanding of fundamental scientific questions (such as climate change and wildfires) that are of great importance to society. In addition, providing concrete plans for collaboration and education, the project capitalizes on the large, urban minority population in the CU-Denver student body and on the active CU-Boulder recruitment plan. Although specific plans are not clear, the PIs all have a record of working with students of diverse backgrounds.