[Bug: 21.5-b28] performance on aix, cpu hog, another session gprof output

[Jerry James]

On Tue, Mar 11, 2008 at 12:06 PM, Thomas Mittelstaedt
<T.Mittelstaedt at cadenas.de> wrote:
> Okay, I have attached the truss output with the start:

Thank you.  This time the mass of kioctl calls is on a socket with
file descriptor 5, which is created on line 6797 of the truss output.
It is a TCP/IP socket.  The kioctl calls appear to be mostly
1074030207 (== 0x4004667F), which I believe is FIONREAD.  However, I
do not have access to an AIX box to check the headers.  Thomas, can
you verify that?  If that is the case, it seems unlikely to be the
source of the performance problem.

We need to have some idea of where the CPU time is going.  Would you
mind running truss again, but this time passing it the -d option?  I
think that's the one that puts a time stamp on every event.  It's
probably also a good idea to tell truss to trace all children, which
is -f I think.  With that extra information, maybe we can see where
the big jumps in time are, which will help us narrow our focus.

In any case, I've been doing this on my lunch hour, which is now over.
 I think my employer would like me to get back to work. :-)  I won't
respond for several more hours, most likely.  Anybody who has a clue,
feel free to jump in!
-- 
Jerry James
http://loganjerry.googlepages.com/