On the use of trace sampling for architectural studies of desktop applications

Abstract

Examines the feasibility of performing architectural studies with trace sampling for a suite of desktop application traces on Windows NT. This paper makes three contributions: we compare the accuracy of several sampling techniques to determine cache miss rates for these workloads, we present victim cache and branch prediction architecture studies that demonstrate that sampling can be used to drive such studies, and we show how sampling may be used to accurately and efficiently derive the parameters for A. Agarwal et al.'s (1988) analytical cache model. Of the sampling techniques used for the cache miss ratio determinations, the stitch technique, which assumes that the state of the cache at the beginning of a sample is the same as the state at the end of the previous sample, narrowly outperforms the more complex INITMR technique of D.A. Wood et al. (1991) for these workloads. These two techniques are more accurate than the others and are reliable for caches up to 64 KB in size.