Re: opportunistic tuple freezing

On Tue, 2009-09-15 at 20:56 -0700, Jeff Janes wrote:
> Under what kind of circumstances/workload to you think this patch is
> most likely to show its full potential? I can try to test it out, but
> would like some guidance. I am guessing it is when the anti-wrap
> around vacuums come due, but that is such a rare event, it could both
> be hard to test for and also be of limited real-world applicability.

I would expect the benefit to come when tuples start to reach
vacuum_freeze_min_age, and the vacuums start freezing them. Without the
patch, I expect that, on average, vacuum will freeze the same page
multiple times even if the tuples are all quite old during the first
round of freezing.

So this would really only be a problem in a long steady state with a
high volume of transactions. The patch will hopefully reduce the write
volume going on in the background.

I expect the biggest benefit comes when the tuples on a given page may
have been inserted/updated over a few million transactions. Under normal
circumstances, it won't be a huge win, but it's not a huge patch,
either ;)

Regards,
Jeff Davis