Merlin Moncure wrote:
> On 10/14/06, Chris Mair <chrisnospam@1006.org> wrote:
> >The interesting part is the graph that shows updates / sec real time
> >vs. running total of updates:
> >http://www.1006.org/misc/20061014_pgupdates_bench/results.png
>
> one small thing: the variances inside the trendline are caused by
> using integer timestamps...each slanted line is one second. The blue
> line has a very slight wobble which is the effects of the vacuum..its
> very slight. Actually in this test it would probably be good to
> vacuum extremely often, like every 100 records or so.
I was thinking what would happen if you used 8.2 for this test and had a
process continuously vacuuming the table, i.e. start a new vacuum as
soon as the previous one finished, with a reasonable vacuum_delay
setting (not sure what would qualify as reasonable; probably needs its
own set of tests to determine the sweet spot).
--
Alvaro Herrera http://www.CommandPrompt.com/
The PostgreSQL Company - Command Prompt, Inc.