RE: "Slow System" Problem
Hi, can you tell me how can I detect the "looping connections" from te
onstat -ntu?
thank's
Gustavo Tobares
Administrador de Sistemas y DBA
Centro de Computos - Red Megatone
TE: 0342-4500972 - Fax: 0342-4500940
-----Mensaje original-----
De: Greg Mitchell [mailto:greg.mitchell@tecsys.com]
Enviado el: Mi'rcoles, 14 de Mayo de 2003 02:53 p.m.
Para: informix-list@iiug.org
Asunto: Re: "Slow System" Problem
Zev,
That's a tough one. First off, try using something like Peter
Schmidt's most_active1.sh script at www.peterschmidt.com to determine
which engine thread is doing the most I/O at the time.
After that you may want to use the onlog utility to review the log
entries to determine who/what is putting out a high transaction
volume.
Use the pid from the onstat -g ses <sessid> output from the script to
get the percentage physical cpu usage.
ps -o pcpu,comm -p <pid>
As to isolating the percent virtual cpu. I'm not certain how that can
be done. It must be a metric you have to activate like the wait
statistics.
Another thing to check is the onstat -g ntu and see if any of your
connections are looping. I've had some programs burn up virtual CPU
by not disconnecting properly and it showed up in the onstat -g ntu.
regards,
Greg Mitchell
Tecsys Inc.
On Wed, 14 May 2003 11:20:12 -0400, "Zev Berezin"
<zevb@bhphotovideo.com> wrote:
>
>Hi All,
>
> Tech Specs:
>IDS: 7.31.UD5W5 - O.S: Solaris 6 - 10 CPU 248 MHZ
>
> I have a situation where the OLTP system usually hums along
>great but at times slows down drastically.
>This can happen at any time of the day (so forget the crons).
>It would be great to be able to pinpoint the sessions causing the
>problem or to be able to track CPU usage by session.
>At times I would kill the sessions where onstat -u showed the most
>reads and/or writes but that wouldn't help.
>Any help is greatly appreciated.
>
>TIA,
>Zev Berezin