Awake: decentralized and availability aware replication for P2P cloud storage

Abstract

The traditional decentralized availability-based replication algorithms suffer from high dependence on the underlying system's churn behavior, randomness in replica selection, and the inability of maximizing the replicas availability. These drawbacks result in poor data availability especially in low available systems as well as where the churn behavior is mispredicted. In this paper, we propose dynamic, fully decentralized availability aware algorithm named Awake, with the goal of maximizing the availability of replicas. Compared to the existing solutions, Awake always provides the maximum availability of replicas regardless of the underlying system's churn behavior. By employing Awake, a data owner can select its replicas only based on the aggregated availability information of nodes obtained in a fully decentralized manner with asymptotically the same message overhead as the communication complexity of the underlying system. Awake has linear space complexity in the number of registered users to the system. Our extensive simulation results show that in comparison to the best existing decentralized solutions, regardless of the underlying churn model of the system, Awake improves the availability of replicas with a gain of about 21%. Likewise, Awake is scalable by showing the same performance independent of the system size.