Towards Optimal Data Replication Across Data Centers

There has been proliferation of data centers that provide both computation and storage resources at diverse geographic locations. In a variety of wide area applications, data can be replicated to serve users with lower latency. This paper presents a technique that can effectively reduce the overall data access delay through gradual migration of data replicas. In contrast to previous solutions that either randomly select replica locations or process a large log of past data accesses, this new technique maintains only a small, decentralized summary of recent data accesses while achieving near optimal performance. This paper also includes an evaluation study that substantiates the effectiveness of the developed technique, and plans for extending the current research outcomes.