CAJ | 학술논문

提出了一种数据块关系指导的缓存替换方法BDP(block correlations directed replacement policy).数据块关系所表示的空间局部性可以用来指导存储系统的缓存替换策略.BDP利用历史访问信息和实时访问信息,预测数据块未来的空间局部性特征,并设计数据块驻留时间的控制策略,减少现有替换算法对空间局部性预测失败带来的失效惩罚.同时给出了BDP的有效缓存模型.模拟结果显示,对大多数实际系统工作负载,BDP的失效率相对LRU算法下降11%～38%,优于现有的考虑空间局部性的替换算法.
제출료일충수거괴관계지도적완존체환방법BDP(block correlations directed replacement policy).수거괴관계소표시적공간국부성가이용래지도존저계통적완존체환책략.BDP이용역사방문신식화실시방문신식,예측수거괴미래적공간국부성특정,병설계수거괴주류시간적공제책략,감소현유체환산법대공간국부성예측실패대래적실효징벌.동시급출료BDP적유효완존모형.모의결과현시,대대다수실제계통공작부재,BDP적실효솔상대LRU산법하강11%～38%,우우현유적고필공간국부성적체환산법.
A block correlations directed cache replacement policy,BDP,is proposed.The spatial locality presented by block correlations can be used to direct cache replacement algorithm used in storage systems.BDP utilizes both history and real-time access information to accurately predict block's future access characteristics.A policy is also designed to control block's resident time so as to reduce the penalty incurred by the failure of predication.In addition,a validated cache model is given.Simulation results show that,for most real system workloads,BDP can reduce the cache miss ratio of classic cache replacement algorithms by 11%～38% and is better than other spatial locality based cache replacement algorithms.