On Fri, 2007-12-14 at 15:22 -0500, Robert Treat wrote:
> Simon, have you tested async_commit performance with any of the the beta
> releases? I was involved in a conversation on irc the other day that had
> some anecdotale reports that async_commit was not giving significant speed
> benfits, perhaps as a result of the other tuning and work that has gone into
> the release later in the development cycle. I've not had chance to verify
> this info, so I was wondering if you had.
Few of the features in 8.3 apply to all workloads, so such anecdotes
could be applied to any of them.
I don't think other tuning would mask that.
I measured ~900% performance gain on a pure INSERT workload. Any
transaction whose major component is the commit time will benefit from
increased performance, i.e. short transactions get even faster. Many web
apps fall into this category, especially those currently running on
MySQL, Solid etc. But its also easy to come up with workloads that would
show no gain at all.
HOT has been similarly misunderstood, though I feel we now have a good
understanding that it prevents performance drops, rather than simply
being a boost in itself, as was originally thought.
Of course, if anybody has some negative performance test cases for Async
Commit, we'd be able to discuss those.
But from all the things that have happened in recent weeks, I'm now
wondering how much information is out there about all the performance
features that have been added, and when to make use of them.
--
Simon Riggs
2ndQuadrant http://www.2ndQuadrant.com