可配置及历史信息感知的多级缓存策略

发布时间：2018-03-07 05:29

  本文选题：多级缓存　切入点：动态随机存取存储器　出处：《计算机研究与发展》2015年S2期 　论文类型：期刊论文

【摘要】：传统磁盘存储设备因其固有的机械特性,已不能满足当前的数据密集型应用程序的需求。基于闪存的固态存储设备(solid state drive,SSD)的出现改善了这种情况,并被广泛用作缓存以降低内存与磁盘之间的性能差距。针对由DRAM和SSD构成的多级缓存,提出了一种可配置的历史信息感知的多级缓存替换策略Charm.Charm允许用户配置应用的访问模式、读写模式等多项内容,并且还可以根据应用对文件的历史访问信息来判断访问模式,从而能够适应访问模式的变化.此外,Charm过滤掉那些只访问一次的数据,将多次访问的热数据缓存至SSD,减少对SSD的写入次数,提升SSD寿命.使用MCsim对Charm与现有的多级缓存替换算法进行了对比测试,在实际的工作负载下,Charm优于其它多级缓存算法.
[Abstract]:Traditional disk storage devices, due to their inherent mechanical properties, can no longer meet the needs of current data-intensive applications. The advent of flash based solid-state storage devices, solid state driveed SSDs, has improved the situation. And it is widely used as cache to reduce the performance gap between memory and disk. For the multilevel cache composed of DRAM and SSD, a configurable historical information aware multilevel cache replacement policy, Charm.Charm, is proposed, which allows users to configure the access mode of applications. Read / write mode, etc., and can judge the access pattern according to the historical access information to the file, so that it can adapt to the change of the access mode. In addition, charge filters out the data that accesses the data only once, In order to reduce the number of writes to SSD and improve the lifetime of SSD, the multi-access hot data is cached to SSD.The Charm is compared with the existing multi-level cache replacement algorithm by using MCsim. Under the actual workload, charge is superior to other multi-level cache algorithms.
【作者单位】： 武汉光电国家实验室(华中科技大学);信息存储系统教育部重点实验室(华中科技大学);华中科技大学计算机科学与技术学院;
【基金】：国家“八六三”高技术研究发展计划基金项目(2013AA013203) 华中科技大学自主创新研究基金项目(HUST:2014QN010)
【分类号】：TP333
，

本文编号：1578143