Whenever a table is read in random order, additional disk accesses are required to read the rows in the required order. Disk costs are higher when the rows of a table are read in a sequence unrelated to physical order on disk. Because the pages are not read sequentially from the disk, both seek and rotational delays occur before each page can be read. As a result, the disk-access time is much higher when a disk device reads table pages nonsequentially than when it reads that same table sequentially.
Nonsequential access often occurs when you use an index to locate rows. Although index entries are sequential, there is no guarantee that rows with adjacent index entries must reside on the same (or adjacent) data pages. In many cases, a separate disk access must be made to fetch the page for each row located through an index. If a table is larger than the page buffers, a page that contained a row previously read might be cleaned (removed from the buffer and written back to the disk) before a subsequent request for another row on that page can be processed. That page might have to be read in again.
Depending on the relative ordering of the table with respect to the index, you can sometimes retrieve pages that contain several needed rows. The degree to which the physical ordering of rows on disk corresponds to the order of entries in the index is called clustering. A highly clustered table is one in which the physical ordering on disk corresponds closely to the index.
Enterprise Edition Home | Express Edition Home | [ Top of Page | Previous Page | Next Page | Contents | Index ]