On Fri, 16 Nov 2001, mlw wrote:
> Sequential and random read test, these are a no brainer.
>
> The cpu costs are not so easy. I don't have a very good idea about what they
> "really" mean. I have a guess, but not enough to make a benchmark routine.
>
> If someone who REALLY knows could detail a test routine for each of the cpu
> cost types. I could write a program that will spit out what the numbers should
> be.
>
> I envision:
>
> pgprofile /u01/postgres/test.file
>
> And that would output something like:
>
> random_page_cost = 2
> cpu_tuple_cost = 0.00344
> cpu_index_tuple_cost = 0.00234
> cpu_operator_cost = 0.00082
Actually, it could be done if the 'EXPLAIN EXACTLY' was implemented. Such
a command would give you same output as explain plus precise timings each
step took. Idea was floated in the list awhile ago. I think the problem
with it was properly separating borders of queries, but still, it'd cool
-alex
