An Effective Cache Replacement Policy For Multicore Processors

Abstract

Cache management is increasingly important on multicore systems since the available cache space is shared by an increasing number of cores. Optimal caching is generally impossible at the system or hardware. The goal of cache management is to maximize data reuse. In this paper, a collaborative caching system is implemented that allows a program to choose different caching methods (generally LRU or MRU) for its data. We have developed a LRU MRU caching replacement policy for multicore processor which is simulated using Multi2sim simulator. We have compared the value of the following parameters (Instruction committed per cycle and branch predication Accuracy) of the multicore processor when different caching algorithm is used for cache management: replacement policies are LRU,MRU,FIFO and LRU-MRU. We use 2 benchmarks:Spec2006 and Splash-2 to see how or LRU-MRU replacement is behaving and what is the value of IPC, is it less than LRU,MRU,FIFO and Random or vice versa. We get Random replacement policy performance is worst when use for multicore processor and LRU and MRU performance is almost same. Here Performance refers to the value of IPC, More the value of IPC better the performance of the replacement policy used.