Vivek Khera wrote:
> >>>>> "TL" == Tom Lane <tgl@sss.pgh.pa.us> writes:
>
> TL> My best guess is that the dropoffs occur because of background checkpoint
> TL> operations, but there's not enough info here to prove it. Four inserts
> TL> per second seems horrendously slow in any case.
>
> I'll concur with this diagnosis. I've been doing a bunch of
> performance testing with various parameter settings, and the
> checkpoint frequency is a big influence. For me, by making the
> checkpoints occur as far apart as possible, the overall speed
> improvement was incredible. Try bumping the number of
> checkpoint_segments in your postgresql.conf file. For my tests I
> compared the default 3 with 50 segments.
>
> Check your logs to see if you are checkpointing too frequently.
That warning message is only in 7.4.
--
Bruce Momjian | http://candle.pha.pa.us
pgman@candle.pha.pa.us | (610) 359-1001
+ If your life is a hard drive, | 13 Roberts Road
+ Christ can be your backup. | Newtown Square, Pennsylvania 19073