Compressing Very Large Database Workloads for Continuous Online Index Selection

The paper presents a novel method for compressing large database workloads for purpose of autonomic, continuous index selection. The compressed workload contains a small subset of representative queries from the original workload. A single pass clustering algorithm with a simple and elegant selectivity based query distance metric guarantees low memory and time complexity. Experiments on two real-world database workloads show the method achieves high compression ratio without decreasing the quality of the index selection problem solutions.