Socially-Aware Data Replication

Online social networking has become one of the most important forms of today\u27s communication. While an online social network can be attractive for many socially interesting features, its competitive edge will diminish if it is not able to keep pace with increasing user activities. Deploying more servers is an intuitive way to make the system scale, but for the best performance one needs to determine where best to put the data, whether replication is needed, and, if so, how. This dissertation is focused on replication, specifically, solving an original problem where social data needs to be replicated in a distributed storage system such that server performance in terms of efficiency and load balancing is optimized and equal data availability is guaranteed for all users, regardless of how active they read and write data and how frequently they socially interact with one another. A socially-aware replication solution, called S-CLONE, is proposed for this problem, which is shown in both simulation and experiment to outperform random replication, a state-of-the-art method used in most OSNs today, and be comparable with an evolutionary algorithm approach that provides the best Pareto optimality