Role of Cache Replacement Policies in High Performance Computing Systems: A Survey

Abstract

Cache replacement policies play important roles in efficiently processing the current big data applications. The performance of any high performance computing system is highly depending on the performance of its cache memory. A better replacement policy allows the important blocks to be placed nearer to the core. Hence reduces the overall execution latency and gives better computational efficiency. There are different replacement policies exits. The main difference among these policies is how to select the victim block from the cache such that it can be replaced with another newly fetched block. Non-optimal replacement policy may remove important blocks from the cache when some less important (dead) blocks also present in the cache. Proposing better replacement policy for cache memory is a major research area from last three decades. The most widely used replacement policies used for classical cache memories are Least Recently Used Policy (LRU), Random Replacement Policy or Pseudo-LRU. As the technology advances the technology of cache memory is also changing. For efficient processing of big data based applications today’s computer having high performance computing ability requires larger cache memory. Such larger cache memory makes the task of replacement policies more challenging. In this paper we have done a survey about the innovations done in cache replacement policies to support the efficient processing of big data based applications.