| 摘要 |
A database system with improved query execution performance is described. In order to achieve as much accuracy as possible by optimizing large block I/O related query plans, the database system implements a methodology employing two new statistical measures for use in a database server. The first statistical measure, called "Data Page Cluster Ratio" (DPCR), indicates the "packing" of data pages by measuring the "clustering" property of the data pages, i.e., how well clustered the consecutive data pages of a given relational table are in the extents. The ratio is measured for data pages accesses either via a scan of an index or via a scan of the table. The second statistical measure, called Index Page Cluster Ratio (IPCR), indicates the "packing" of index leaf pages by measuring the "clustering" property of index pages, i.e., how well clustered the consecutive index leaf pages of a given relational table's index are in the extents. The ratio is measured for index leaf pages accesses via the scan of a non-clustered/clustered index. Additionally, another set of formulae is derived to use these statistical measures for the estimation of the number of I/Os in the optimizer's cost model. By employing these measures, the database server's optimizer can provide an intelligent way of considering and selecting useful large block I/O access strategies, thereby leading to increased performance gains during query execution.
|
