Performance Evaluations of a Parallel and Expandable Database Computer - The Multi-Backend Database Computer

This thesis is the actual application of a performance evaluation technique known as benchmarking to an experimental database management system DBMS. The specific DBMS evaluated is the Multi-Backend Database System MBDS which is a software multiple-backend database system. The unconventional nature of a multiple-backend computer system required the development of a special performance evaluation methodology which was the topic of several related thesis. A previously developed performance evaluation methodology had only been applied to MBDS on a very small scale and had not been used with the current set of modern MBDS hardware. The focus of this thesis is the verification of the performance claims made by the implementor of MBDS. These performance claims were, in fact, validated by conducting a series of relatively large-scale benchmarking experiments in which MBDS performed, generally, as predicted by its implementor. While the results are encouraging, future benchmarking experiments need to be conducted on an even larger database to examine MBDS performance under an extreme load. This will require the developement of a high-speed database loading utility which is not the focus of this thesis. Here, we report on the test databases, test transactions and test results which constitute the benchmarks used to verify the MBDS implementors claims of response-time reduction and response-time invariance.