Lukas Miedema - Optimistic Concurrency Control in Geographically Distributed Databases for Concurrent OLTP Transactions

author:Lukas Miedema
title:Optimistic Concurrency Control in Geographically Distributed Databases for Concurrent OLTP Transactions
keywords:
topics:Other
committee:prof.dr. M. Huisman
graduation date:7 July 2017


Abstract

The need for distributed, scalable databases is larger than ever. Applications handling user-generated data like Twitter and Facebook have a global presence, but require performance characteristics from their database as if it was close to the user. Current database solutions either only offer low-latency reads by replicating the data, or skip ACID compliance by letting transactions do dirty writes with the possibility of overwriting the work of others. To solve this, we propose a way of using optimistic concur- rency control (OCC) between nodes in the network. Every node holds a full copy of the data, but takes ownership over only a specific partition of the data that is geographically close to the users who may need to write to it. To evaluate this approach, a distributed database testing system is implemented, simulating latency between the nodes representative of various real-world network sizes. On this database our solution is then compared to an implementa- tion of two-phase locking (2PL) using the TPC-C benchmark.