Query Optimization Using Scrambled Sample Tables

The patent pre-generates sample tables with stored statistical information about data distributions before query execution. These sample tables contain pre-computed selectivity data for various predicates, allowing the optimizer to quickly retrieve accurate estimates without performing complex calculations at query time. This preliminary preparation resolves the contradiction by providing both speed (pre-computed results) and accuracy (real data samples).

The patent creates simplified copy representations of the actual data through sample tables that contain representative subsets of the full dataset. These copies preserve the statistical properties and data distribution characteristics needed for accurate selectivity estimation, while being small enough to store and process efficiently. The copy approach maintains accuracy without requiring access to the entire large dataset during query optimization.

Instead of performing sampling at query time, the patent performs the sampling action in advance and stores the results in sample tables. The sample tables capture data distribution characteristics and statistical information that can be directly reused for multiple queries. This eliminates the time-consuming sampling operation from query execution while maintaining the accuracy benefits of using actual data samples.

The sample tables serve themselves by containing pre-computed statistical information that automatically answers selectivity estimation queries without requiring additional processing. The stored sample data acts as a self-contained reference that provides accurate estimates instantly, eliminating the need for repeated sampling operations and reducing processing time for each query.