Thursday, March 27, 2014, 03:30pm
Assistant Professor, Computer Science Dept, University of New Mexico
Scalable Middleware and Tools for HPC
Abstract: In the pursuit of unprecedented high-performance computing (HPC) capabilities, and the scientific and economic advances such capabilities will bring, U.S., Asian, European and even Indian governments have established initiatives to build and deploy extreme scale systems with exaflop (quintillion or 10^18 floating point operations per second) computational power. In terms of component counts, such systems are expected to comprise scales orders of magnitude larger than current systems. As a result, we face heightened challenges in the design and deployment of scalable software services for the management of such systems and tools for finding and fixing functional and performance problems in large scale applications.
In this talk, I describe our research efforts at the University of New Mexico that are focused on the design of scalable, robust middleware services and tools. Specifically, I will talk about our work in tree-based overlay networks (TBONs) that have become the foundation of several important tools at U.S. Department of Energy national laboratories. One such tool I will talk about is the Stack Trace Analysis Tool, a TBON-based tool for debugging extreme scale applications. Lastly, I will overview the work we are doing to address the responsive instantiation of large scale applications and tools, a common problem encountered in extreme scale HPC systems.
Bio: Prof. Dorian Arnold has been an assistant professor at the University of New Mexico since 2009. His research interests are in the broad areas of high performance computing and large scale distributed systems. In particular, he is interested in abstractions, mechanisms and tools that allow system non-experts to harness the power of high-performance systems in scalable, efficient, reliable ways. His research group maintains strong collaborations with the Los Alamos, Livermore, and Sandia National Laboratories -- lending his team the privilege to work with world-class scientists on leading edge HPC systems. In part due to such collaborations, his research projects were selected as Top 100 R&D technologies in 1999 and 2011. Prof. Arnold received a Ph.D. from the University of Wisconsin and an M.S. from the University of Tennessee, both in Computer Science. He received a B.S. in Mathematics and Computer Science from Regis University in Denver, CO and an A.S. in Physics, Math and Chemistry from St. John's College in the Central American country of Belize, where he grew up.
Hosted by: EECS Prof. Peter Dinda