Operating system impact on trace-driven simulation

Abstract

Trace-driven simulation is commonly used by the computer architecture research community to pursue answers to a wide variety of architectural design issues. Traces taken from benchmark execution have been extensively studied to optimize the design of pipelines, branch predictors, and especially cache memories. Today's computer designs have been optimized based on the characteristics of these benchmarks. One important aspect that has been ignored in a majority of these trace-driven studies is the effect of the operating system interacting with the benchmark. It has been acknowledged that operating system overhead can introduce a level of interference that can limit the benefits of new designs. The reason why the operating system has been, for the most part, ignored in these studies is the lack of readily available tools that can generate kernel-laden traces. In this paper we describe two tracing systems that allow the capture of operating system and application traces. We have captured traces of both benchmark and commercial applications using these tools. We show the effects of including operating system activity while studying various architectural design tradeoffs. We have found that the operating system can increase the number of instructions captured in the trace (for a given time sample) by as much as 100X, can significantly modify the instruction mix, and produce different ranges of reference locality. The results of this study show that the inherent characteristics of the applications will dictate the amount of overhead introduced by the operating system.