Data placement cost optimization and load balancing for online social networks

With the rapid development of broadband wireless technology and popularity of the intelligence devices, the number of online social networks (OSNs) users is growing every day. The huge data of users need to be replicated and placed over multiple geographically distributed clouds to be accessible by the users in a reasonable time. Therefore, reducing the storage cost and keeping a reasonable performance of the storage system becomes more and more important. Storing data items in the same cloud may minimize cost but incurs the worst imbalance. In addition, the data transfer cost in OSNs with millions of connections is significant and is not negligible. Therefore, our goal is to optimize the total cost of data storage and transfer while guaranteeing users' latency requirements and keeping a reasonable load balancing. A novel graph-partitioning based algorithm is proposed to achieve our goal. Experiments on two different Facebook datasets demonstrate that our strategy can significantly reduce the total cost and keep a reasonable load balancing in comparison with other representative placement strategies.