Re: [HACKERS] Some notes on optimizer cost estimates

Mike Mascari <mascarm(at)mascari(dot)com> writes:
>> to be useful for cost estimating. We could make it a parameter like
>> we do for CPU_PAGE_WEIGHT ... but you know and I know that no one
>> ever bothers to adjust those numbers in the field ...

> Would it be possible to place those parameters as run-time
> settings and then write a utility that can ship with the
> distribution to determine those values? Kind of a self-tuning
> utility?

Maybe. I'm not sure the average user would want to run it ---
to get believable numbers, you have to be using a table considerably
bigger than the kernel's disk cache, which means it takes a while.
(I've been testing with a gigabyte-sized table ... one of the index
scan runs took thirty hours :-( ... fortunately I have this machine
to myself, or there would have been some howls about the load.)

But it'd be nice to have comparable numbers for different platforms.
What I was really hoping was that someone on the list would be aware
of existing research I could borrow from.

regards, tom lane