Title: Faster transactions in geo-distributed storage

Abstract: Distributed storage systems run transactions acrossmachines to ensure serializability. Traditional approaches for distributedtransactions are based on two-phase locking or optimistic concurrencycontrol.  However, these protocols suffer from performance degradationbecause of aborting and/or blocking. In addition, to provide fault tolerance,traditional approaches replicate data relying on an extra layer of consensusprotocols such as Paxos, which incurs extra cost. This talk focuses on onequestion: how can we improve the system performance without giving up theserializability guarantee? It will cover a new concurrency control protocolbased on dependency tracking, and how to extend it to support geo-replicationin a merged-layer style with lower overhead.

Bio: Shuai Mu is a post-doctoral researcher/lecturer at New York University. He earned his PhD from Tsinghua University (Beijing) in 2015.