Validation of SPEC6(TM) CFP95 traces for accurate performance evaluation of computer systems

Performance evaluation of computer systems and processors is necessary during design, procurement, and capacity planning. The increasing complexity of realistic benchmarks combined with the relatively sluggish rate of detailed performance analysis has resulted in an ever-increasing gap between the size of the workload/traces and the speed of analysis. In this paper, we present a methodology for generating and validating representative traces from SPEC(TM)95 benchmark suite using R-metric and K-metric. Experimental results demonstrate the superiority of our proposed technique over the ubiquitous profile-driven methodology and the uniform sampling approach.