An integrated method for estimating selectivities in a multidatabase system

A multidatabase system (MDBS) integrates information from autonomous local databases managed by diierent database management systems (MDBS) in a distributed environment. A number of challenges are raised for query optimization in such an MDBS. One of the major challenges is that some local optimization information may not be available at the global level. We recently proposed a query sampling method to drive cost estimation formulas for local databases in an MDBS 22]. To use the derived formulas to estimate the costs of queries, we need to know the selectivities of the qualii-cations of the queries. Unfortunately, existing methods for estimating selectivities cannot be used eeciently in an MDBS environment. This paper discusses diiculties of estimating selec-tivities in an MDBS. Based on the discussion, this paper presents an integrated method to estimate selectivities in an MDBS. The method integrates and extends several existing methods so that they can be used in an MDBS eeciently. It extends Christodoulakis' para-metric method so that estimation accuracy is improved and more types of queries can be handled. It extends Lipton and Naughton's adaptive sampling method so that both performance and accuracy are improved. Theoretical and experimental results show that the extended Lipton and Naughton's method described in this paper can be many times faster than the original one. In addition, the integrated method uses a new piggyback approach to collect and maintain statistics, which can reduce the statistic maintenance cost. The integrated method is designed for the MDBS in the CORDS project (CORDS-MDBS). Implementation considerations are also given in the paper.