Algorithmic Techniques for Independent Query Sampling

Abstract

Unlike a reporting query that returns all the elements satisfying a predicate, query sampling returns only a sample set of those elements and has long been recognized as an important method in database systems. PODS'14 saw the introduction of independent query sampling (IQS), which extends traditional query sampling with the requirement that the sample outputs of all the queries be mutually independent. The new requirement improves the precision of query estimation, facilitates the execution of randomized algorithms, and enhances the fairness and diversity of query answers. IQS calls for new index structures because conventional indexes are designed to report complete query answers and thus becomes too expensive for extracting only a few random samples. The phenomenon has created an exciting opportunity to revisit the structure for every reporting query known in computer science. There has been considerable progress since 2014 in this direction. This paper distills the existing solutions into several generic techniques that, when put together, can be utilized to solve a great variety of IQS problems with attractive performance guarantees.