No, you cannot compare them. You do not know the composition of the averages. What is the distribution of the values? Example - A sql statement is run. Compare the "sql trace" and "statspack" results. 3 of the execs in the Statspack are the ones in the SQL trace.
SQL trace - 3 executions, 1 second average (3 execs of 1 second each) (3 * 1) = 3 / 3 = 1 second average Statspack - 30 executions, 3 second average (29 execs of 1 second each, 1 exec of 61 seconds for a total of 90 seconds) (29 * 1) + (1 * 61) = 90 / 30 = 3 second average
OrSQL trace - 3 executions, 21 second average (2 execs of 1 second each, 1 exec of 61 seconds). (2 * 1) + (1 * 61) = 63 / 3 = 21 second average Statspack - 30 executions, 3 second average (29 execs of 1 second each, 1 exec of 61 seconds for a total of 90 seconds) (29 * 1) + (1 * 61) = 90 / 30 = 3 second average
Granted, this is an extreme case (well, not all that extreme considering some performance I have seen), but it illustrates that you cannot absolutely compare a subset with a whole.
-- Daniel Fink Oracle Performance, Diagnosis and Training OptimalDBA http://www.optimaldba.com Oracle Blog http://optimaldba.blogspot.com Zhu,Chao wrote:
I am comparing per Execution cost/elapsed time. Total elapsed time / total executions = per execution cost. they are comparable. On 8/22/07, Allen, Brandon <Brandon.Allen@xxxxxxxxxxx> wrote:Assuming your stats below are for the same query, your tkprof is only accounting for 32945 executions while the statspack is accounting for almost 1 million executions so you're not comparing apples to apples. Regards, Brandon
-- //www.freelists.org/webpage/oracle-l
