Estimation error-aware query optimization: an overview
Résumé
The quality of a query execution plan is heavily dependent on the accuracy of estimated statistics. Unfortunately, estimation errors can occur due to the use of invalid assumptions like Attribute Value Independence, outdated statistics, etc. Such errors result in a query execution plan that is suboptimal. Motivated by this issue, a great deal of work on query optimization has been done over the last decades. Several researches were proposed to solve the problem of performance penalty caused by estimation errors. Some researchers suggested revising the cost model so as to improve the accuracy of estimates. Others proposed new optimization algorithms that manage estimation errors. This paper provides an overview of studies that were done as part of the second solution. We identify two main approaches (single-point and multi-point based optimization) and compare them with respect to the most important challenges of query optimization.