• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

J4 ›› 2013, Vol. 35 ›› Issue (6): 15-23.

• 论文 • 上一篇    下一篇

一种针对iSCSI存储的缓存替换算法

赵建明1, 姚念民2,韩永2,蔡绍滨2   

  1. (1.福建师范大学福清分校,福建 福清 350300;
    2.哈尔滨工程大学计算机科学与技术学院,黑龙江 哈尔滨 150001)
  • 收稿日期:2012-06-15 修回日期:2012-10-19 出版日期:2013-06-25 发布日期:2013-06-25
  • 基金资助:

    国家自然科学基金资助项目(61073047);中央高校基本科研业务费专项资金资助(HEUCFT1202);哈尔滨市科技创新人才专项基金(2012RFLXG023)

A novel cache replacement algorithm for iSCSI storage    

ZHAO Jianming1,YAO Nianmin2,HAN Yong2,CAI Shaobin2   

  1. (1.Fuqing Branch of Fujian Normal University,Fuqing 350300;
    2.College of Computer Science and Technology,Harbin Engineering University,Harbin 150001,China)
  • Received:2012-06-15 Revised:2012-10-19 Online:2013-06-25 Published:2013-06-25

摘要:

由于IP网络的带宽限制和较长的传输延迟,为了完成数据的请求,iSCSI initiator和iSCSI target在确认对方发出的状态或控制命令前将等待更多的时间。iSCSI协议中通信开销所占的比重较大,因此iSCSI存储的缓存数据载入开销不同于传统的直连存储。现有的替换算法研究主要关注缓存命中率的提高,忽略了通过降低失效数据的载入开销来提高缓存系统性能。提出了一种针对iSCSI存储的缓存替换算法CFLLRU,该算法不仅考虑访问时间和访问频度,还将考虑缓存页面的逻辑块地址连续度,从而在兼顾命中率的同时降低失效数据的缓存载入开销。实验结果表明,在各种不同的典型负载和缓存大小下,CFLLRU的效果好于其它替换算法。

关键词: iSCSI, 网络存储, 替换算法, 命中率, 失效开销

Abstract:

Due to the bandwidth restriction and the long delay of IP network, iSCSI initiator and iSCSI target have to wait for a longer time until they acknowledge the status or the control command sent by the other party. The communication of iSCSI protocol takes a large part of the total cost. Therefore, the cache data's loading cost of iSCSI storage is different from the traditional direct attached storage. The existing researches focused on achieving high cache hit ratios, but rarely paid attention to reducing the loading cost of miss data for improving the storage system performance. In the paper, a cache replacement algorithm called CFL-LRU (LRU Combined with Frequency and data Length) was proposed for iSCSI storage. The algorithm not only considers the time and the frequency but also takes into account the logical block address continuity of pages, thereby compromising the hit rate and the miss penalty. Our tracedriven simulation results show that, for different kinds of workloads and cache size, the CFL-LRU outperforms the others.

Key words: iSCSI;network storage;replacement algorithm;hit ratio;miss penalty