Research On Data Partition And Replication For Online Social Network Storage Systems

Recent years, Online social networks (OSN) have been increasingly popular andattracted hundreds of millions of users. The explosive growth of social networks andvast amounts of data generated by the users, raise high demand on the scalability ofback-end system. It has been an urgent need to solve the problem of user data manage-ment. However, the inherently complexity between user data poses great challenges tothe mechanism of data allocation and replication.Analyzing two real dataset from OSNs, we observe that the majority of the inter-actions of a user are attributed to a small subset of the user’s friends and that the largersocial degree of a user, the greater probability of the user’s data is accessed. Inspiredby this observation, we frst build a dynamic weighted social graph which diferenti-ates the importance of social interactions between a user and the user’s friends. Usingthis graph, we design WEPAR, an online partitioning and replication algorithm tak-ing into account read operations, write operations and replication overhead. Extensiveevaluations based on real datasets show that our approach signifcantly reduces storagecost and improves write response time with read response time comparable to that ofexisting algorithms. Besides, WEPAR also achieves excellent stability and scalability.The contributions of this paper are as follows:We introduce a dynamic weighted social graph, which diferentiates the impor-tance of friends by weighting the social relations between users. The nodes,edges and weights in the graph are dynamically changed over time.We propose WEPAR, an online partitioning and replication algorithm for datamanagement in OSN systems. To the best of our knowledge, this is the frst workthat considers both read and write operations and replication overhead in OSNs. All observations and evaluations are based on real OSN datasets, from SinaWeibo,RenRenandFacebook. SinaWeiboandRenRenarethetwomostpopularOSNs in China. The solution of this paper is practical and can guide the designof systems to run the services of OSNs.