Block Correlations Directed Multi-copies Data Layout Technology

Improving the random access performance of Disk-based storage system is a key issue in storage area. This paper proposes the block correlations directed multi-copies date layout technology. Through the I/O stream mining to obtain access related features of the disk data blocks (block correlations), and based on the block correlations to build the data block copies set, which will be placed to the appropriate disk location, so the same data block have multiple copies on disk. When there is application access, through identifying block correlations to select the appropriate copies set for prefetching, so while ensuring the prefetching accuracy, it could control the amount of data prefetching and the number of times of prefetching I/O. Experiments show that using multi-copies data layout technology can decrease the average response time by 4.6%-13%.