基于IPC与公平性的共享Cache划分

提出一种兼顾高速缓冲存储器(Cache)公平性及系统吞吐率的划分方法, 使用Cache访问监控器记录各应用访问Cache的命中及失效次数, 通过动态划分算法决定每个应用占用的Cache数量, 解决了共享Cache访问冲突导致的Cache污染. 实验表明: 在吞吐率方面, 该方法较传统的LRU替换策略可获得最高37.90%, 平均15.71%的提升, 比公平性最优的划分算法可获得最大47.37%, 平均14.11%的吞吐率提升； 在公平性方面, 较传统的LRU替换策略可获得最大4倍, 平均77%的提升; 比失效率最优的划分算法可获得最大9倍, 平均229倍的公平性提升.

Shared Cache Partitioning Based on IPC and Fairness

This paper proposes a shared Cache partitioning algorithm, based on throughput (IPC) and fairness, which in order to find a balance between fairness and throughput. We used the Cache access monitor to collect the missing rate information, and then decide the amount of Cache resources allocatedto each application by the dynamic partitioning algorithm to resolve the problemof Cache pollution. Experiments show that shared Cache partitioning based on IPC and fairness improves throughput by 15.71% on average （up to 37.90%） over least recently used and by 14.11% on average （up to 47.37%） over fairness based Cache partitioning, and improves fairness by 77% on average （up to 4 times） over least recently used and by a factor of 2.29 on average （up to 9 times） over utility based Cache partitioning.