Oracle FAQ Your Portal to the Oracle Knowledge Grid

Home -> Community -> Usenet -> c.d.o.server -> Re: Oracle Problem on AIX

Re: Oracle Problem on AIX

From: Joel Garry <>
Date: 20 Apr 2004 14:25:54 -0700
Message-ID: <> (Pete's) wrote in message news:<>...
> >
> > AIX & Oracle Version numbers would be nice to know. I've experienced
> > the same issue with AIX 4.3.3 and Oracle, &
> > I'm fairly certain it has nothing to do with Oracle directly, but, to
> > do with AIX. I have not been able to pin it down. I can let my
> > servers run for several months and then paging fills up. Monitoring
> > paging space 'free space', I noticed no clear pattern. i.e. I'd see
> > 30% free on a Tuesday and then see 50% free the next. I'll be
> > upgrading to 5.2 fairly soon, hopefully it's worked out there. I've
> > also noticed that when it fills, you might be able to telnet to it,
> > but forget about running any kind of os command. I then have to "kick
> > the server" to get it back(yuk I don't like the power resets without a
> > shutdown).
> >
> > HTH,
> > Pete's
> Just had an idea on this problem. Obviously, you should try to check
> which process is holding so much paging. I just performed an upgrade
> on a production DB today that's on an HACMP system before performing
> the upgrade, I noticed I did not shutdown the listener. At the time I
> had 'topas' running and was monitoring the overall system performance,
> I took note of where paging free space was at. I then shutdown the
> listener and wallaa, my paging space went from approx. 30% used(with
> the database shutdown and listener running) to approx. 5% used with no
> oracle processes running. The node has been up for approx. 25 days
> with no listener shutdowns or restarts. I'll be investigating the
> listener process further as paging is filled up.

Watch that netstat, other unix platforms have issues with TCP or giving up sockets, I'm too long gone from AIX to know if it might. The general problem is apps or users don't let tcp do its final fin_wait negotiations by exiting improperly, and the server doesn't time them out in a timely manner. So their process workspaces just keep adding up.


-- is bogus.
Solaris 10 is rewriting the tcp stack.  Great, another round of finger
pointing like DISABLE_OOB.
Received on Tue Apr 20 2004 - 16:25:54 CDT

Original text of this message