Accuracy of filtered traces

Reducing the size of large address traces by "filtering" them through a small direct-mapped cache is a useful technique for making more efficient use of both secondary storage and processors used for trace-driven simulation. However, when filtered traces are used to drive cache simulators, the distortion introduced by such filtering can produce substantial errors in the miss ratios obtained, despite earlier reports to the contrary. We present the results of a systematic study of such errors, including a model for compensating for some errors.<
>