|
I know he already showed consistent gets, but he seems to ignore it and focus on the elapsed time as viewed on the client side. Two issues here. The two client-side elapsed times are too close to each other, prone to errors of various sources. They need to be magnified for performance measurement. If he doesn't want to increase table size (actually the index size here), he can use PL/SQL to loop many times. In addition, repeated tests in random order of the two are needed to check reproducibility.
Secondly, if elapsed time is really important, server-side v$sql.elapsed_time should be checked, preferably combined with reference to cpu_time, to judge efficiency.
Yong Huang |
|