A static data placement strategy towards perfect load-balancing for distributed storage clusters

Applications like cluster-based video-on-demand (VOD) systems are inherently data-intensive because clients frequently retrieve data stored on a distributed storage subsystem interconnected by a high-speed local network. To meet the quality-of-service (QoS) imposed by the clients, quick responses to access requests are fundamental for these applications. Among the numerous ways to reduce response times, data placement, has attracted much attention from researchers due to its effectiveness and low cost. In this paper, we propose a novel load-balancing and performance oriented static data placement strategy, called perfect balancing (PB), which can be applied to distributed storage subsystems in clusters to noticeably improve system responsiveness. The basic idea of PB is to balance the load across local disks and to minimize the discrepancy of service times of data on each disk simultaneously. A comprehensive experimental study shows that PB reduces mean response time up to 19.04% and 8.67% over the two well-known data placement algorithms Greedy and SP respectively.