On Mon, Sep 14, 2009 at 2:07 AM, Jeff Davis
<pgsql@j-davis.com> wrote:
On Mon, 2009-08-17 at 10:22 -0400, Tom Lane wrote:
> As always with patches that are meant to improve performance,
> some experimental evidence would be a good thing.
I haven't had time to performance test this patch yet, and it looks like
it will take a significant amount of effort to do so. I'm focusing on my
other work, so I don't know if this one is going to be in shape for the
September commitfest.
If someone is interested in doing some performance testing for this
patch, let me know. I still think it has potential.
Does the community think that experimental performance testing is a must-have in order for this patch to be acceptable? If so, it sounds like this should be marked as rejected or RwF, and no longer considered for this commit fest, and I should move on to a different patch.
I'll do some work on benchmarking it, but since it takes hundreds of millions of transactions to make a plausible scenario, this will not be done any time soon.
Also, I see that the number of frozen tuples is not logged. I'd like to add that to the info reported under Log_autovacuum_min_duration, both to help evaluate this patch and because it seems like something that should be reported.
Cheers,
Jeff Janes