Performance Testing And Analysis Of Redundant Arrays Of Inexpensive Databases

The Internet revolution has focused scaling and redundancy research on systems, such as application servers using clusters, redundant Internet connections, and other redundant hardware. Conversely, back-end services, such as databases, have largely remained undistributed, unclustered, and housed in large, dedicated SMP machines. However, a new turn in database management recently was introduced with the proposal of a RAIDb (Redundant Array of Inexpensive Databases) standard. Various studies have indicated that redundant disk arrays and computer node clustering have greatly improved machine performance per dollar spent. Clustering provides immense processing capability and redundancy has proved to be a valuable resource in up time and availability. Based on these notions, RAIDb utilizes increased processing power through database computer clustering, while providing increased availability through the use of redundant databases and database controllers. This paper details the performance of the C-JDBC (Clustered JDBC), a Java-based implementation of RAIDb. Performance is measured using a variable-node RAIDb cluster against a standard database backend. Each database backend is flooded with SQL requests by a benchmark client, which keeps track of the number of requests per minute successfully served by the database engine. Results of the testing are compiled and interpreted, showing performance trends and comparisons of the database implementations.