ECE Seminar Series: Application-Specific HPC from an Operational Standpoint

ECE title

ECE Seminar Series Fall 2014

Thursday November 13st 1-2 PM, ITEB 336

Application-Specific HPC from an Operational Standpoint

Chris J. Michael

Naval Research Laboratory

Abstract: There are numerous situations, especially within the Department of Defense (DoD), where smaller HPC systems are specified and deployed to handle a moderately predictable workload containing less than a dozen or so special-purpose applications. Typically, these systems are designed in a naive way using rough high-level benchmarking and intuition. In most cases, this results in a system that runs the workload inefficiently, consuming precious resources and dramatically increasing the operational cost of the system.

Application-specific HPC involves designing heterogeneous systems with respect to the application workload. This is accomplished through profiling of the applications against the potential hardware candidates. Resulting systems can dramatically increase the workload efficiency when considering execution time, cost of operation, power, size, and weight. In this presentation, the basics of application-specific HPC from an operational DoD standpoint are covered. Additionally, case studies that exemplify this methodology are presented, the most thorough of which deals with all-pairs shortest paths processing for sparse graphs.

 

Short Bio: Dr. Chris J. Michael is a computer engineer with the Naval Research Laboratory located in Stennis Space Center, Mississippi. His research interests include heterogeneous computing, special-purpose computer architecture, large-volume streaming data processing, and high-performance image processing. Chris is currently working with Department of Defense sponsors to find rapidly transition able solutions to computationally heavy problems. Examples of involved computation of interest include graph processing such as centrality as well as image processing involving sparse representation theory. He received his doctorate in electrical engineering from Louisiana State University in 2010.

 

Host: Omer Khan, omer.khan@engr.uconn.edu