Applies ToOracle Server - Enterprise Edition - Version: 8.1.7 to 188.8.131.52
Information in this document applies to any platform.
GoalThis article explains how a nested loop join works.
FixA nested loop join is the simplest join technique.
Let R and S be the two relations (tables) to be joined.
And R is designated as outer table and S is inner table.
For each row 'r' of the outer table, all rows 's' of inner table are read and compared with the row from the outer table. Whenever the join condition is satisfied, the two rows are concatenated to form. a row which is placed in an output relation.
In simplest words, each row of the cartesian product R x S is tested on the join condition. If it it satisfied then it is included into the join result.
Hence, for a nested loop join to occur, a where clause is not always needed.
The algorithm for the same is as follows:
for each row r belongs to R
for each row s belongs to S
if r and s satisfy C then
put (r concatenate s) in the output table
Instead of retrieving one row each time, an entire block of tuples (rows) is read from disk and cached in main memory. This is more efficient because this way, several random accesses are replaced by sequential rows accesses which is faster. A further performance improving measure is to use the relation with the lowest cardinality as the outer relation in order to reduce the I/O costs which will be given by :
size of outer table + (size of outer table)*(size of the inner relation)
And this can be minimized if :
size of outer table <= size of inner table.
Furthermore, we can switch the direction in which the inner relation is read each time. This has the advantage that the last block read by the previous inner loop is the first block of the oncoming inner loop. It still is in a memory buffer and therefore doesn't need to read from disk.
来自 “ ITPUB博客 ” ，链接：http://blog.itpub.net/38267/viewspace-743198/，如需转载，请注明出处，否则将追究法律责任。