A Method of Improving the Performance of Continuously Reading Discrete Small Blocks in the Large Storage System
一种改进大型存储系统离散小数据块读取性能的方法

In the large storage system,the operation of continuously reading discrete small blocks severely impacts the I/O performance. To solve this problem,this paper designs and implements a system,CBSS(correlative blocks speedup system),which implements precise prefetch and regulates the data distribution according the small blocks correlations,mined by a novel heuristic algorithm between the file system and block device. The system performance can be improved evenly and continuously without interruption and sudden state transitions. Furthermore,compared with other algorithms,this heuristic algorithm thinks about both the locality and the globality of the correlations. Through the experiments,it has been proved that CBSS and the algorithm are effective and the system I/O performance can be enhanced distinctly. Furthermore,the prototype can be used universally without modifying the file system and the storage devices.