MRI: Acquisition of an Instrument for Research in Irregularly Parallel Big Data Computation

New Mexico State University, Las Cruces NM

Investigators

Abstract

Proposal #: 13-37884 PI(s): Cook, Jonathan E. Cao, Huiping; Cook, Jeanine M.; Pontelli, Enrico; Song, Mingzhou Institution: New Mexico State University Title: MRI/Acq.: Instrument for Research in Irregularly Parallel Big Data Computation Project Proposed: This project, acquiring a computational instrument configured to support data-driven graph computations (DDGC) that might enable-to-scale and improve parallel computation that is generally irregular and hard to scale and improve. (Very regular computations--for example, fluid dynamics--already have a long history of development and are highly optimized to run on modern high performance (HPC) instruments.) The instrument, with 320 cores and a node and system architecture, is designed specifically with potentially transformative DDGC research in mind. The local node memory hierarchy includes both fast solid state secondary storage and traditional mechanical disk storage. Research projects explore ways to exploit the new layer of fast solid state storage, not just as a standalone data container, but within the memory hierarchy of the local node. The system has in its configuration both GPU and FPGA computing support. Some of the research projects exploit non-traditional, yet potentially powerful computation mechanisms. In particular, the instrument supports the following projects: - Genome Assembly and Annotation Computations, - Data Mining over Large Graphs, - Reasoning with Big Knowledge, - Hardware Acceleration for Scalable Graph-Based Computations, and - Data Driven Monitoring and Analysis of Scientific Computations. Each of these projects will make use of the general architecture of the instrument. GPU and FPGA capabilities will be used to further explore and enhance the possibilities to improve performance of data-driven graph computations. The instrument architecture is intended to enable cross-fertilization between the research projects that will, in turn, contribute to the development of new approaches that can perform DDG computations efficiently. Broader Impacts: Due to the growth of data analytics in so many areas of society, the techniques developed utilizing the instrument should be valuable across society (from economic capacity to homeland security needs). Techniques that can improve the DDG computations within many of the big data computations (from improved algorithms to newly demonstrated hardware approaches) are immediately useful. Moreover, the instrumentation enriches the research and educational activities at a minority-serving institution within an EPSCoR jurisdiction. The project offers opportunities for students and professors to participate in novel computer research.