Date and Location: 4/20, Wed., 2:00PM. CS Conference Room, 206
Speaker: Wei Xie
Title: A Version Consistent Hashing Scheme for Better Manageability in Distributed Storage Systems

Abstract: Consistent hashing is widely used in very large scale parallel/distributed file systems due to the better scalability and adaptivity to node additions. However, it suffers performance degradation in case of addition of nodes, which is very common in data centers. In this study, we identify the problem of performance degradation of adding nodes to a cluster using the existing consistent hashing algorithm. A version consistent hashing scheme that targets on maintaining the performance in such cases is proposed. In this scheme, the history of node membership changes is maintained so that all the locations that a data item might have stayed could be calculated so that the data lookup is efficient without the data migration being completed. Comparing to the existing placement algorithms, the version consistent hashing does not incur much extra computation cost and memory footprint, but significantly reduces the overhead when storage nodes are added.