Long Distance Replication in a Highly Available Database Management System

Abstract

Over the last couple decades, we have experienced a huge growth in telecommunication services and mobile networking. This growth has introduced database management systems tailor-made for telecommunication purposes. They differ from traditional database management systems by having strict requirements regarding response time, throughput, and availability. Telecommunication companies may have formal requirements to locate operational sites far apart, thus arises the problem of how to perform long distance replication of DBMS content, while still maintaining strict requirements regarding availability of data.

This thesis will study two different approaches to a potential 1-safe long distance replication architecture. The two alternatives differ in how they execute and replicate transactions. The alternatives will be evaluated through a theoretical analysis, and in addition, a simulator will be implemented in order to simulate the proposed system applying both of the alternative approaches.

The results show that one of the two approaches studied, alternative B, achieves best throughput both in the analytical model and in the simulations. This model takes advantage of parallelizing the long distance log stream, and this proves to be essential if the replication distance is kept relatively short. We will show that when the replication distance increases, the difference between alternatives will diminish.