COMPUTE STATISTICSinstructs Oracle to compute exact statistics about the analyzed object and store them in the data dictionary. When computing statistics, an entire object is scanned to gather data about the object. This data is used by Oracle to compute exact statistics about the object. Slight variances throughout the object are accounted for in these computed statistics. Because an entire object is scanned to gather information for computed statistics, the larger the size of an object, the more work that is required to gather the necessary information.
To perform an exact computation, Oracle requires enough space to perform a scan and sort of the table. If there is not enough space in memory, then temporary space may be required. For estimations, Oracle requires enough space to perform a scan and sort of only the rows in the requested sample of the table. For indexes, computation does not take up as much time or space, so it is best to perform a full computation.
Some statistics are always computed exactly, such as the number of data blocks currently containing data in a table or the depth of an index from its root block to its leaf blocks.
Use estimation for tables and clusters rather than computation, unless you need exact values. Because estimation rarely sorts, it is often much faster than computation, especially for large tables.
ESTIMATE STATISTICSinstructs Oracle to estimate statistics about the analyzed object and stores them in the data dictionary. When estimating statistics, Oracle gathers representative information from portions of an object. This subset of information provides reasonable, estimated statistics about the object. The accuracy of estimated statistics depends upon how representative the sampling used by Oracle is. Only parts of an object are scanned to gather information for estimated statistics, so an object can be analyzed quickly. You can optionally specify the number or percentage of rows that Oracle should use in making the estimate.
To estimate statistics, Oracle selects a random sample of data. You can specify the sampling percentage and whether sampling should be based on rows or blocks.
- Row sampling reads rows without regard to their physical placement on disk. This provides the most random data for estimates, but it can result in reading more data than necessary. For example, in the worst case a row sample might select one row from each block, requiring a full scan of the table or index.
- Block sampling reads a random sample of blocks and uses all of the rows in those blocks for estimates. This reduces the amount of I/O activity for a given sample size, but it can reduce the randomness of the sample if rows are not randomly distributed on disk. Block sampling is not available for index statistics.
- The default estimate of the analyze command reads the first approx 1064 rows of the table so the results often leave a lot to be desired.
- The general consensus is that the default value of 1064 is not sufficient for accurate statistics when dealing with tables of any size. Many claims have shown that estimating statistics on 30 percent produces very accurate results. I personally have been running estimate 35 percent. This seems to produce very accurate numbers. It also saves a lot of time over full scans.
- Note that if an estimate does 50% or more of a table Oracle converts the estimate to a full compute statistics.