Re: Client and Shadow Process stuck communicating in Network Layer

From: Stefan Koehler <>
Date: Fri, 26 Jun 2015 11:04:17 +0200 (CEST)
Message-ID: <>

Hi Patrick,

> Host OS is, and which believe contains fixes for all three.

The main problem with these network issues is that many releases (also the newest AIX 7.1 levels) were affected and sometimes it was re-introduced as well. Don't ask me about the code upstream of IBM ;-) However you can easily find out with "instfix -i | grep -i <APAR_NUM>"

The main issue with the AIX network stack (in this bug case) is, that it do not show the error clearly. For example the last few bytes are just not sent and it is very unpredictable. The last time we needed to request an Oracle client debug patch (which dumped some AIX kernel stuff when the issue was hit) to identify the issue :-(

Just from my experience and your problem description, i am pretty sure that the host without VIO also suffers one of these bugs.  

Best Regards
Stefan Koehler

Freelance Oracle performance consultant and researcher Homepage:
Twitter: _at_OracleSK

> Patrick Jolliffe <> hat am 26. Juni 2015 um 08:07 geschrieben:
> Thanks for the pointers Stefan. I have checked, and this particular host is not using VIO.
> Host OS is, and which believe contains fixes for all three.
> IV31011 is particularly interesting as connection IS loopback.
> We are actually have slightly similar, but different problem right now with a DB Link from Linux to AIX.
> I can see AIX side is stuck in write, Linux is stuck in read. (Note this AIX host DOES use VIO)
> Interesting to note from "netstat -an" command mentioned in one of the links you sent, I can see the data is sitting in "Send Q" buffer.
> To me proves that this is OS Level/Networking issue, rather than application/DB level.
> At least that is some evidence I can use to direct our efforts,
> Regards
> Patrick

Received on Fri Jun 26 2015 - 11:04:17 CEST

Original text of this message