A join ordering approach for multicomputer relational databases with highly skewed data

Multicomputer machines are currently used for many scientific applications. Recently, attention has been focused on the development of multicomputer database applications. The implementation efforts of database applications on parallel machines confront two fundamental obstacles; database applications are not regular, therefore, there is difficulty to extract any inherent parallelism; the participating relations can be highly skewed, thus causing unbalanced processor utilization. We present the operation ordering segment of DOME, a query optimization environment developed on an Intel i860 hypercube system, that includes techniques to handle the above problems and provide efficient exploitation of the processing power. DOME uses a sampling approach, as part of the query execution process, to derive the statistical information of the relations at every level of a query execution tree. The operation ordering segment examines the statistical information and determines one efficient query execution tree, among many alternatives.<>