Plan Selection Based on Query Clustering

Query optimization is a computationally inten- sive process, especially for complex queries. We present here a tool, called PLASTIC, that can be used by query optimizers to amortize the opti- mization cost. Our scheme groups similar queries into clusters and uses the optimizer-generated plan for the cluster representative to execute all future queries assigned to the cluster. Query simi- larity is evaluated based on a comparison of query structures and the associated table schemas and statistics, and a classifier is employed for efficient cluster assignments. Experiments with a variety of queries on a commercial optimizer show that PLASTIC predicts the correct plan choice in most cases, thereby providing significantly improved query optimization times. Further, when errors are made, the additional execution cost incurred due to the sub-optimal plan choices is marginal.