Workload metrics based on instruction grouping
The talk will discuss workload metrics based on instruction grouping. the focus is on the Java bytecode instruction set, but the principles are generally applicable across architectures, wherever a workload can be described by frequencies of a set of instructions or other basic operators. We discuss the motivation and justification for grouping frequencies of related instructions and propose an iterative grouping methodology. The effects of our approach are demonstrated using cluster analysis of SPEC JVM98 benchmarks.
Jozo Dujmovic was born in Dubrovnik and received his BSEE, MS, and doctorate from the University of Belgrade. He is a Professor of Computer Science at San Francisco State University, where he served for four years as Chairman of Computer Science Department. His teaching and research activities are in the areas of software metrics, decision analysis, and computer performance evaluation. He is the author of more than 100 refereed publications, recipient of three best paper awards, and a Senior Member of IEEE. He served as General Chair of IEE MASCOTS 2000 and is currently serving as a General Chair of ACM WOSP 2004.
Carl Herder earned his BS and MS degrees in Computer Science at SFSU. His experience in software engineering spans 18 years, and encompasses a wide range of roles and environments. Mr. Herder is involved in ongoing research in Java workload characterization. He is currently a Senior Software Engineer with Cyanea Systems Corporation in Oakland.