* Gordan Bobic <gordan@freeuk.com> [001229 07:39] wrote:
> > * Adam Lang <aalang@rutgersinsurance.com> [001229 05:47] wrote:
> > > But isn't it recommended to run the server with fsync? If so, you
> shouldn't
> > > disable it on a benchmark then.
> >
> > Actually, if he ran Postgresql with WAL enabled, fsync shouldn't
> > make much of a difference.
>
> What's WAL? Are you referring to autocommit? I will admit that autocomit
> already improves performance so much that fsync() isn't going to matter
> that much, but it will still make a noticeable improvement. It certainly
> did on my system (~20%). FWIW, disabling autocommint, and introducing the
> overhead of doing a select for checking before EVERY INSERT and UPDATE,
> made an improvement of about 2-3 times on my application... But, different
> applications require different things, so... YMMV...
WAL is apparently something that orders writes in such a way that you
may loose data, but ordering is maintained such that if you have
transactions A, B and C (in that order) and you crash, you'll see
one of these:
1) A B and C
2) A and B
3) just C
With fsync on you should see A B and C, but WAL makes data recovery
a lot better.
(I think. :) )
--
-Alfred Perlstein - [bright@wintelcom.net|alfred@freebsd.org]
"I have the heart of a child; I keep it in a jar on my desk."