Reputation based cache management policy for performance improvement

Abstract

As the performance gap between CPU and main memory speed increases, memory subsystem design more critical. Caches are important part of modern memory hierarchies. Cache performance is important in computer system. As the cache size is limited it should be properly utilized. After the cache block is used by an application it should be moved from cache to main memory so that another block required by other application can use its space [2]. Now a day's multi-level cache are used everywhere to provide better cache performance. As the block which is moved from cache memory to main memory may be required in future, so it is kept in lower level cache in the cache hierarchy. Therefore the cost of accessing the block from main memory is minimized. Many approaches like LRU-K, LFU, LRFU, PROMOTE, DEMOTE are used to increase cache performance by promoting or demoting a memory block into the cache depending upon its latest history [1] [21]. Some are used for single level cache and some for hierarchical cache. The drawback of existing multi-level cache management techniques is that it uses hints of level L1 only. It does not provide sufficient information about the history of the cache block at all previous cache levels. Therefore, a hybrid cache management policy is proposed in this paper which takes its replacement decision based on the multiple level cache level. In this paper, a hybrid cache management policy is proposed which constitute the feature of two existing algorithms PEOMOTE and DEMTE that gives better hit ratio than other existing policies.