Re: Deathly slow performance on SMP red-hat system

In article <44F87893(dot)9050400(at)phlo(dot)org>, Florian G. Pflug <fgp(at)phlo(dot)org> wrote:

% You could try doing:
%
% begin;
% select 1;
% select 1;
% ...
% rollback;

Yes, I tried that, but each select is on the order of 100ms.

% off. If that doesn't reveal a possible cause, then I'd suggest that
% you strace the backend you're connected to, and try to see where it's
% spending it's time. Since 100ms for a simple "select 1;" is way out

I tried that, but I don't have the results handy, and the machine's not
responding now. I remember the first time I looked at the result, I
thought it was spending an inordinate amount of time in gettimeofday,
but I also had an strace which showed time in send, which is where
I think the problem is.

I'm pretty strongly convinced that the time is being spent sending data
to the client. Actually I measured the full round trip and found it was
taking on the order of 200ms, which is consistent with no time at all to
perform the select, but 100ms each to send the query to the db and the
result to the client.

I think I need to checkout the socket layer on this machine.
--

Patrick TJ McPhee
North York Canada
ptjm(at)interlog(dot)com