多核系统中共享Cache的冒泡替换算法  

Algorithm of Bubble Replacement in Multi-Core Shared Caches

作　　者：林晓敏[1] 桂婷[2] 乔福明[1] 胡同森[1] 

机构地区：[1]浙江工业大学计算机科学技术学院,浙江杭州310014 [2]浙江工业大学之江学院,浙江杭州310024

出　　处：《微电子学与计算机》2011年第4期118-121,共4页Microelectronics & Computer

摘　　要：在多核环境下,对共享L2 Cache的优化显得尤为重要,因为当被访问的数据块不在L2 Cache中时(发生L2缺失),CPU需要花费几百个周期访问主存的代价是相当大的.在设计Cache时,替换算法是考虑的一个重要因素,替换算法的好坏直接影响Cache的性能和计算机的整体性能.虽然LRU替换算法已经被广泛应用在片上Cache中,但是也存在着一些不足:当Cache容量小于程序工作集时,容易产生冲突缺失;且LRU替换算法不考虑数据块被访问的频率.文中把冒泡替换算法应用到多核共享Cache中,同时考虑数据块被访问的频率和最近访问的信息.通过分析实验数据,与LRU替换算法相比,采用冒泡替换算法可以使MPKI(Misses per Kilo instructions)和L2 Cache命中率均有所改善.The shared L2 Cache optimization is extremely important in multi-core architecture, because the cost that L2 miss led to takes hundreds of cycles to access main memory is quite large. The replacement algorithm is an im- portant factor considered for designing Cache and directly affects the performance of Cache and computers overall performance. Although the LRU replacement algorithm has been widely used in on-chip Cache, but there are also several deficiencies： When Cache size is smaller than program working set, it is easy to lead to conflict miss; LRU Replacement Algorithm not consider the access frequency. Bubble replacement algorithm in this article is applied to multi-core shared Cache; meanwhile take the recent access information and the frequency information of the data block into consideration. By analyzing the experimental data, compared with the LRU replacement algorithm, MPKI and the hit rate of L2 has been improved.

关 键 词：替换算法 CACHE LRU 冒泡替换 多核 

分 类 号：TP302[自动化与计算机技术—计算机系统结构]