Ernest E Vogelsinger <ernest@vogelsinger.at> writes:
> But this wouldn't explain the huge differences I just posted (first query -
> 1500 msecs, follow-ups - 10 msec)
No, I was just responding to Arjen's wondering where 40 or so msec had
gone ... that's of the right order of magnitude to be cache load effects.
> Just now I had the server sit for approx 20 minutes, rerunning the same
> query resulted in 3155 msec, a followup again 10.85 msec.
It's really hard to believe that you could see that kind of ratio from
any sort of cache effects, even kernel disk buffer cache which is
normally pretty large. Are you sure you were getting the same plan each
time? I'd like to see EXPLAIN ANALYZE output from all three cases.
regards, tom lane