"Clay Luther" <claycle@cisco.com> writes:
> Performance follows an odd "peak and valley" pattern. It will start
> out with a high insertion rate (commits are performed after each
> "device set"), then after a few thousand device sets, performance will
> drop to 1 device/second for about 5 seconds. Then it will slowly ramp
> up over the next 10 seconds to /just below/ the previous high water
> mark. A few thousand inserts later, it will drop to 1 device/second
> again for 5 seconds, then slowly ramp up to just below the last high
> water mark.
My best guess is that the dropoffs occur because of background checkpoint
operations, but there's not enough info here to prove it. Four inserts
per second seems horrendously slow in any case.
What are the table schemas (in particular, are there any foreign-key
constraints to check)?
Are you doing any vacuuming in this sequence? If so where?
What's the disk hardware like? Do you have WAL on its own disk drive?
regards, tom lane
PS: pgsql-performance would be a better list for this sort of issue.