Data Declustering with Replications

Declustering is used to distribute blocks of data among multiple devices, thus enabling parallel I/O access and reducing query response times. Many data declustering schemes have been proposed in the literature. However, these schemes are designed for non-replication systems, and thus they will fail if any disk fails. Assume that a single disk would fail once every five years, a non-replication system with 100 disks would have failed every 18 days. Data replication is a technique commonly used in multidisk systems to enhance availability of data during disk failures and, often as a second goal, to improve I/O performance of read-intensive applications. In this paper, we propose a LOG data declustering scheme for systems with replication. Furthermore, we present a novel replication algorithm. Although the replication algorithm is designed for the LOG declustering scheme, it is also applicable to existing schemes such as DM, GFIB, and GRS. Finally, as demonstrated by our experimental results, the LOG scheme with the proposed replication algorithm provides a significant performance improvement compared to the state-of-the-art data declustering schemes.