MRI: Acquisition of an Adaptive Computing Infrastructure to Support Compute- and Data-Intensive Multidisciplinary Research

George Mason University, Fairfax VA

Investigators

Abstract

A high-performance computing (HPC) system, named Hopper will be acquired by George Mason University (GMU). This HPC system will support high-impact research and education across all the multidisciplinary research institutes, and numerous departments and academic programs at George Mason University (GMU). Among other areas, Hopper will enable research and education projects focused on advancing transportation and infrastructure systems, geography and geoinformation sciences, astrophysics, social media analytics to support disaster informatics, computational fluid dynamics, materials science, natural hazards research, data mining, computer vision, automated vehicles, bioinformatics, neuroscience, and economics. Hopper will also support GMU’s large-scale high-performance computing users who will develop, prototype, and benchmark their applications locally before starting production runs on large nationally available systems. The instrument will support a very diverse community of researchers, educators and graduate and undergraduate researchers at GMU. It will also enable research projects by high school students engaged in summer research projects, community college students whose dreams are to attain their four-year degree, adult learners seeking their first college degree or to upskill to realize their professional aspirations. In addition to advancing cutting-edge research projects, Hopper will be instrumental in preparing the next generation of data scientists, mathematicians, engineers and scientists to meet the needs of the Nation’s thriving innovation economy. Hopper will be managed centrally and shared with the entire Mason research community. The proposed system is based on an OpenStack framework and will be reconfigurable to support a diverse range of user needs at Mason. Hopper will consist of 4 OpenStack management nodes, 35 standard compute nodes, 2 high memory nodes with 3TB memory, and 6 nodes with graphics processing units (GPUs), all connected through InfiniBand (IB) switches to support low-latency high bandwidth communication between nodes and storage. The interconnect network will use HDR IB switches in a fat tree configuration and use port splitters to provide EDR IB connections to the nodes. There will be a redundant 25Gbe networking infrastructure with 100Gbe spine switches for node deployment and virtual machine (VM) access. The system will have 384 TB of Ceph based storage to back VM deployments, and 300 TB of high-speed low-latency BeeGFS storage for HPC use. Hopper will support both traditional as well as non-traditional (the long tail of science) high-performance computing users. This new system will replace GMU’s current HPC cluster, which is nearing the end of its useful life, is unable to keep up with growth in the number of users and was not designed to meet the needs of a diversity of usage modes now being requested by our faculty and student communities. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.