[PAST EVENT] Colloquium: Program Locality Research - Past and Present

November 11, 2011
McGlothlin-Street Hall, Room 020
251 Jamestown Rd
Williamsburg, VA 23185Map this location
Program data and access may be reorganized to improve memory performance. In complex code, however, restructuring is difficult since it depends on the composition of all program modules as well as the input and the underlying hardware. Moreover, the restructuring needs to be re-applied when the program, the workload or the hardware changes. As multicore processors become commonplace, and improving the data usage in one program can benefit its co-running peers, memory optimization is gaining both importance and complexity.

A program's active memory usage is known as program locality. In this presentation, I will review reuse distance based models as a unify basis for understanding and synthesizing the many types of techniques and tools emerged in the last two decades. One remaining problem was the measurement of program memory footprint. Because it was computationally intractable, all past methods either approximated or took snapshots. I will show a complete solution and the advances it has enabled in theory especially the relation with reuse distance and in practice such as the modeling of cache sharing on multicore processors.

About the speaker. Chen Ding received Ph.D. from Rice University, M.S. from Michigan Tech, and B.S. from Beijing University. His research seeks to understand the composite and emergent behavior in computer systems especially its dynamic parallelism and active data usage and develop software techniques for locality optimization, data management, and recently suggestion-based program parallelization and optimization. His work received young investigator awards from NSF and DOE. He co-founded the ACM SIGPLAN Workshop on Memory System Performance and Correctness (MSPC) and was a visiting researcher at Microsoft Research and a visiting professor at MIT. He is currently an IBM Center for Advanced Studies Fellow.

Department of Computer Science