The Optimization Management Strategy Based On Multi-processor Sharing Cache

With the rapid development of Multi-Processor,it becomes the mainstream architecture of high performance microprocessor.With the further development and popularity of multi-processor and semiconductor technology,more cores will be integrated on a chip,which not only brings more challenges and pressure to design cache structure,but also make the problem of "memory wall" become increasingly obvious and intensified.The current mainstream multi-processor uses a shared or private cache structure.However,a single shared or private cache structure cannot effectively balance the capacity and access delay.Shared cache structure can effectively use the resources,but the global wire delay resulting in slower access speed;private cache structure through data replication get faster access speed,but capacity constraints lead to more access failure.How to effecttively use the multi-processor shared resources,making the cache play to the maximum extent,has very important significance.In view of these problems,This thesis studies cache management strategy by L1 and L2 cache in this thesis.The main research contents including:1.Cache management method based on sharing instruction cache.The cost of technical difficulty and manufacturing of the first level cache are the highest in all levels of the cache,so make full use of the first level cache on the cache manage-ment is very important.This paper puts forward a management method,the structure with two cores shared the same L1 instruction cache,each core has private L1 data cache,and all cores shared the last level cache.This management method better use of cache space in general.Experiments show that,the cache hit rate of this management method and system throughput are higher.2.The optimization management strategy based on multi-processor sharing level two cache.This paper studies the organization and management mechanism of level two cache resources in multi-processor,proposes improved bubble cache replacement algorithm,distinguish shared and private data with the different cache management strategy.Cache replacement strategy is an important method for optimal management of the cache,this paper uses the advantages of bubble replacement algorithm which consider the frequency and recently used informationof the accessed data blocks,and improved the algorithm.Experiments show that the improved algorithm,although less increase the hardware overhead,but effectively improved the system throughput and the hit rate of cache;And in this part I distinguish shared and private data,improve the management method to optimize the shared cache lifetime,experiments show that the method can enhance the system throughput and the weighted speedup.