In content networks, a replication group refers to a set of nodes that cooperate with each other to retrieve information objects from a distant server. Each node locally replicates a subset of the server objects, and can access objects stored by other nodes at a smaller cost. In a network with autonomous nodes, the problem is to construct efficient distributed algorithms for content replication that decrease the access cost for all nodes. Such a network also has to deal with churn, i.e. random "join" and "leave" events of nodes in the group. Churn induces instability and has a major impact on cooperation efficiency. Given a probability estimate of each node being active that is common knowledge between all nodes, we propose in this paper a distributed churn-aware object placement algorithm. We show that in most cases it has better performance than its churn unaware counterpart, increases fairness and incites all nodes to cooperate. Key words: content networks, repli...
Eva Jaho, Ioannis Z. Koukoutsidis, Ioannis Stavrak