An efficient cache replacement policy with distinguishing write blocks from read blocks in disk arrays

  • Authors:

  • Yulin Wang;Guangjun Li;Xiaojun Wu;Shuisheng Lin

  • Affiliations:

  • Inst. of Commun./Info. Engineering, Univ. of Electronic Science and Technology of China, Chengdu, Sichuan, China;Inst. of Commun./Info. Engineering, Univ. of Electronic Science and Technology of China, Chengdu, Sichuan, China;Integrative Research Dept., Huawei Tech. Lmt. Corp, Shengzhen, Guangdong, China;Inst. of Commun./Info. Engineering, Univ. of Electronic Science and Technology of China, Chengdu, Sichuan, China

  • Venue:
  • ISPA'05 Proceedings of the 2005 international conference on Parallel and Distributed Processing and Applications
  • Year:
  • 2005

Quantified Score

Hi-index 0.00

Visualization

Abstract

The cache in disk array controller is critical to disk array performance. Many researches have been done to improve the hit-ratio in the cache, including the prefetch algorithms and the replacement algorithms, such as the LRU-K algorithm, the 2-Q algorithm and so on. All these algorithms assume that it takes the same cost to replace all the data blocks. But the cost of replacing write blocks is much higher than the cost of replacing read blocks. Based on the facts a new replacement algorithm named the write-prior least recently used (WP-LRU) algorithm is presented in this paper. The data blocks in the cache are divided into read blocks and write blocks according to the host access mode. The two types of data blocks are managed with different methods. The LRU algorithm is only used to read blocks and all victim blocks are read blocks. Some especial operations are performed for all write blocks in the cache. A simulation model is developed and the simulation results show that the WP-LRU algorithm can improve the performance of disk arrays by reducing the average service time remarkably while it keeps the same hit-ratio as the standard LRU.