Gossip Membership Management with Social Graphs for Byzantine Fault Tolerance in Clouds

Abstract : As computer systems have become more complex and dynamic, unstructured and decentralized techniques serve as basic building blocks in large-scale systems such as cloud computing systems. In particular, we consider a gossip-based algorithm, one of the unstructured overlay construction techniques. In this paper, we propose a membership management mechanism using the gossip-based algorithm with social graphs for the Byzantine fault tolerance problem. Experimental results show that our membership management mechanism copes with Byzantine nodes effectively in a scalable way without a bottleneck in dynamic computing environments, requiring only n ≥ 2f + 1 nodes.