Network-Aided Concurrency Control in Distributed Databases
Srinivasa, Rashmi, Department of Computer Science, University of Virginia
Reynolds, Paul, Department of Computer Science, University of Virginia
Concurrency control is an integral part of a database system. Devising a concurrency control technique that has a low lost opportunity cost and a low restart cost is a hard problem. The interconnection network in a distributed database system can act as a powerful coordination mechanism by providing certain useful properties. We identify several such useful network properties, and present a new family of concurrency control techniques that are built on top of these properties. Network-aided concurrency control techniques use network properties to keep the lost opportunity cost and restart cost low. Our thesis is that network properties can be exploited to achieve efficient concurrency control of transactions.
We also contribute to the evaluation of concurrency control techniques, both analytically and through simulation. Traditional analytical models fail when trying to model certain kinds of workloads. We present a new analytical modelling technique that overcomes the limitations of traditional models. We also show that current perception of the relative merits of concurrency control techniques is flawed. Timestamp ordering concurrency control techniques have been perceived as poor performers until now. We show that two timestamp ordering techniques — a traditional technique and a network-aided technique — perform better than the most popular concurrency control technique for a wide range of conditions.
PHD (Doctor of Philosophy)
All rights reserved (no additional license for public reuse)