Adaptive Cache Replacement Policy (Acrp): A Dynamic Replacement Policy for Cache Management

Abstract

The discovery of memory access time as one of the major design issues in the processor has increased effort in the development of different cache replacement policies to manage what to be the content of caches in both uniprocessor and multiprocessor. This is to improve the overall performance of the processor. Most time the potential benefits of these policies are not instantaneous due to the varying nature of the workloads. Therefore, it is difficult to identify which particular policy to apply during cache's content replacement. In this paper, an adaptive replacement technique using the minimum threshold value of the improvement factor of cache, which was analytically obtained, is proposed in order to explore the combine merits of fundamental replacement policies and strikes a minimum balance with their demerits. ACRP approach uses the threshold values of improvement factor obtained from the analytic modeling to determine when the current replacement policy's performance becomes or getting worse, the ACRP switches to another replacement policy from the pool of replacement policy. Specifically, the ACRP approach solves the problem associated with wrong usage of replacement policy and reduces processor cache miss rate caused by wrong evictions.