A Cache Partition System In Virtualized Environment Based On Reuse Distance Analysis

With popularization of multi-core technology, parallel computing model is used toimprove processing ability of the whole computer system. However, multi-coretechnology also brings serious shared cache contention problem, which leads to cache hitsdecrease and performance degradation of the co-scheduled tasks. Especially in the highperformance computing virtualized environment, cache resource contention breaks downthe isolation mechanism of virtualization, and causes computing performance degradationand the loss of safety and reliability.A cache partition system in virtualized environment based on reuse distance analysisprovides memory access addresses sampling and cache miss rate sampling of highperformance computing applications. Via the file segmentation and cyclic compressalgorithm, it computes reuse distance histogram of each application. Based on reusedistance analysis, this system classifies all applications on cache behavior characteristics.Cache partition mechanism receives virtual resource requests of multiple applicationsfrom users, and matches them with the applications’ classification result. With thecooperation of physical nodes, the system completes applications’ resource allocation andcache partition. This system provides cache partition in virtual machine granularity basedon reuse distance analysis. It minimizes the influence of cache interference, and enhancesapplications’ computing performance and quality of service.The experiment result shows that this system significantly improves the performanceof cache sensitive applications. It reduces execution time by10%~21%as well as cachemiss rate by10%~25%, increases throughput by10%~23%.