DAROS: Distributed user-server assignment and replication for online social networking applications

In this paper we study the problem of assigning users to servers and data replication in a distributed manner for online social networking (OSN) applications. Typical OSN applications such as Facebook and Twitter are built on top of an infrastructure of servers, which handle the users data storage and communications. Thus, for a given user´s communication pattern, the loads of the servers depend critically on the assignment of users to servers. A good assignment will reduce the overall load of the system. Furthermore, by replicating data across the servers judiciously, the overall load can also be further reduced. Unfortunately, this optimal assignment and data replication problem is NP-hard. Therefore, we introduce a distributed heuristic algorithm in which the servers perform local computations and exchange information among each other iteratively in such a way that the algorithm converges to a good assignment and replication in terms of reducing the overall system load as well as balancing the loads among the servers. In contrast with a centralized algorithm, a distributed algorithm offers the advantage of balancing the computations among all the servers as well as the ability to naturally adapt to time-varying user´s communication patterns. Simulations results show promising performance for the proposed algorithm.