A Performance Study of Epoch-based Commit Protocols in Distributed OLTP Databases

Waudby, J; Ezhilchelvan, P; Mitrani, I; Webber, J

doi:10.1109/SRDS55811.2022.00026

A Performance Study of Epoch-based Commit Protocols in Distributed OLTP Databases

Lookup NU author(s): Dr Jack Waudby, Dr Paul Ezhilchelvan ORCiD, Emeritus Professor Isi Mitrani, Professor Jim Webber

Downloads

Accepted version [.pdf]

Licence

This is the authors' accepted manuscript of a conference proceedings (inc. abstract) that has been published in its final definitive form by IEEE, 2022.

For re-use rights please refer to the publisher's terms and conditions.

Abstract

Distributed OLTP systems execute the highoverhead,two-phase commit (2PC) protocol at the end of everydistributed transaction. Epoch-based commit proposes that 2PCbe executed only once for all transactions processed within a timeinterval called an epoch. Increasing epoch duration allows moretransactions to be processed before the common 2PC. It thusreduces 2PC overhead per transaction, increases throughput butalso increases average transaction latency. Therefore, required isthe ability to choose the right epoch size that offers the desiredtrade-off between throughput and latency. To this end, we developtwo analytical models to estimate throughput and average latencyin terms of epoch size taking into account load and failureconditions. Simulations affirm their accuracy and effectiveness.We then present epoch-based multi-commit which, unlike epochbasedcommit, seeks to avoid all transactions being aborted whenfailures occur, and also performs identically when failures donot occur. Our performance study identifies workload factorsthat make it more effective in preventing transaction aborts andconcludes that the analytical models can be equally useful inpredicting its performance as well.