Abstract

Caches are commonly employed to hide the latency gap between memory and the CPU by exploiting locality in memory accesses. On today’s architectures a cache miss may cost several hundred CPU cycles. In order to fulfill stringent performance requirements, caches are also used in hard real-time systems. In such systems, upper and lower bounds on the execution time of tasks have to be computed. Different methods have been proposed for timing analysis, including measurement and static analysis. The sensitivity of a cache replacement policy expresses to what extent the initial state of the cache may influence the number of cache hits and misses during program execution. We have developed a tool to precisely compute sensitivity properties for a large class of replacement policies including LRU, FIFO, PLRU, and MRU. Analysis results demonstrate that the initial state can have a strong impact on the cache performance 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 underestimating the number of misses as strongly as is possible for FIFO, PLRU, and MRU may yield worstcase-execution-time estimates that are far wrong.

Citations