home · mobile · calendar · colloquia · 2002-2003 · 

Colloquium - Burtscher

Trace-File Compression Using Value Predictors
Cornell University

Trace files record the execution behavior of programs. Unfortunately, nontrivial program traces tend to be very large and have to be compressed. While good compression schemes exist for traces that capture only the PCs of the executed instructions, these schemes are quite ineffective on traces that include additional information such as register values or effective addresses. Our novel value-predictor-based approach compresses this type of trace up to six times better. Moreover, our approach includes several other desirable features such as a fixed memory requirement and a single-pass algorithm. This talk follows the design steps of our compression utility and illustrates how value predictors can be used to effectively compress program traces.

Martin Burtscher received the BS/MS degree in computer science from the Swiss Federal Institute of Technology (ETH) Zurich in 1996 and the PhD degree in computer science from the University of Colorado Boulder in 2000. He is an assistant professor in the School of Electrical and Computer Engineering at Cornell University. His research focuses on high-performance microprocessor architecture, instruction-level parallelism, and compiler optimizations. His current work includes hardware and software-based value prediction, data compression, and memory latency reduction.

Department of Computer Science
University of Colorado Boulder
Boulder, CO 80309-0430 USA
May 5, 2012 (14:13)