Scaling Up to Large (Really Large) Systems
* Free, public seminar — all welcome. No RSVP required. *
Abstract
I will discuss the problem of developing tools and middleware for large-scale parallel environments. We are especially interested in systems, both leadership class parallel computers and clusters that have 100,000's or even millions of processors. The infrastructure that we have developed to address this problem is called MRNet, the Multicast/Reduction Network. MRNet's approach to scale is to structure control and data flow in a tree-based overlay network (TBON) that allows for efficient request distribution and flexible data reductions.
I will then present a brief overview of the MRNet design, architecture, and computational model and then discuss a few of the applications of MRNet. The applications include scalable automated performance analysis tools, STAT (a scalable stack trace analyzer running currently on millions of cores), Totalview (a popular and mature parallel debugger), and an extreme-scale cluster algorithm that we developed (called "Mr. Scan") that harnessed 32K GPUs.
About the speaker
Barton P. Miller is a Professor of Computer Sciences at the University of Wisconsin, Madison, USA. He is also Chief Scientist of the DHS-funded Software Assurance Marketplace (SWAMP) Research Facility.
He is known as "the father of fuzzing", and is a pioneer in the field of automated software robustness testing.
He received his B.A. degree from the University of California, San Diego in 1977, and M.S. and Ph.D. degrees in Computer Science from the University of California, Berkeley in 1980 and 1984.
Professor Miller is a Fellow of the Association for Computing Machinery (ACM).
Professor Barton P. Miller.