Sensitivity of cache replacement policies

Abstract

The sensitivity of a cache replacement policy expresses to what extent the execution history may influence the number of cache hits and misses during program execution. We present an algorithm to compute the sensitivity of a replacement policy. We have implemented this algorithm in a tool called R elacs that can handle a large class of replacement policies including LRU, FIFO, PLRU, and MRU. Sensitivity properties obtained with R elacs demonstrate that the execution history can have a strong impact on the number of cache hits and misses if FIFO, PLRU, or MRU is used. A simple model of execution time is used to evaluate the impact of cache sensitivity on measured execution times. The model shows that measured execution times may strongly underestimate the worst-case execution time for FIFO, PLRU, and MRU.