Bruce Momjian <maillist@candle.pha.pa.us> writes:
>> I am currently trying to investigate the poor performance reported by
>> Edmund Mergl --- since gprof doesn't really work on my Linux box, I
>> am reduced to running a profilable postmaster on my HPUX box with the
>> database area NFS-mounted from the Linux box, where there is enough disk
>> space for the benchmark. This setup gives new meaning to the term
>> "slow", but I should be able to get a useful profile out of it.
> I can run it here. What do you want me to do?
What I'm after currently is a profile of the no-indexes case. Build a
one-million-row test database using the script Edmund provided (see
his message of 22 May 1999 06:39:25 +0200), but stop the script before
it invokes "make_idx". Then, with a backend compiled -pg, run this
query:update bench set k500k = k500k + 1 where k100 = 30
and send me the gprof results.
If you have time, it'd also be useful to have a profile of the same
update with indexes in place (run Edmund's make_idx script and then do
the update again).
I'm pretty close to having a profile from my NFS lashup, but it would
be nice to have profiles of the same thing from other machines as
a cross-check that no artifacts are getting introduced...
thanks, tom lane