On Tue, 26 Nov 2002, Andrew Sullivan wrote:
> On Tue, Nov 26, 2002 at 11:06:47AM -0700, scott.marlowe wrote:
> > So, the difference in performance was around 4% slower.
> > I'd hardly consider that a big hit against the database.
> > Note that in every test I've made up and run, the difference is at most 5%
> > with vacuumdb -z running continuously in the background. Big text fields,
> > lots of math, lots of fks, etc...
> Also, it's important to remember that you may see a considerable
> improvement in efficiency of some queries if you vacuum often, (it's
> partly dependent on the turnover in your database -- if it never
> changes, you don't need to vacuum often). So a 5% hit in regular
> performance may be worth it over the long haul, if certain queries
> are way cheaper to run. (That is, while you may get 4% slower
> performance overall, if the really slow queries are much faster, the
> fast queries running slower may well be worth it. In my case,
> certainly, I think it is.)
Agreed. We used to run vacuumdb at night only when we were running 7.1,
and we had a script top detect if it had hung or anything. I.e. vacuuming
was still a semi-dangerous activity. I now have it set to run every hour
(-z -a switches to vacuumdb). I'd run it more often but we just don't
have enough load to warrant it.
In response to
pgsql-performance by date
|Next:||From: Bruce Momjian||Date: 2002-11-26 19:09:38|
|Subject: Re: [HACKERS] Realtime VACUUM, was: performance of insert/delete/update|
|Previous:||From: Andrew Sullivan||Date: 2002-11-26 18:24:39|
|Subject: Re: performance of insert/delete/update|
pgsql-hackers by date
|Next:||From: Peter Eisentraut||Date: 2002-11-26 18:50:51|
|Subject: Re: elog(PANIC) should abort()?|
|Previous:||From: Peter Eisentraut||Date: 2002-11-26 18:42:42|
|Subject: Re: [HACKERS] Location of language .mo files or|