[Users] IPOIB issue

Wed Oct 14 08:16:36 PDT 2015

That does seem to be the symptoms... grrr.

I am using kernel 2.6.32-504.8.1.el6.x86_64
I keep the kernel locked, but do update all other packages as we are running lustre here and rebuilding the modules is quite a pain.

Brian Andrus
ITACS/Research Computing
Naval Postgraduate School
Monterey, California
voice: 831-656-6238

From: users-bounces at lists.openfabrics.org [mailto:users-bounces at lists.openfabrics.org] On Behalf Of Mehmet Soysal
Sent: Wednesday, October 14, 2015 7:53 AM
To: users at lists.openfabrics.org
Subject: Re: [Users] IPOIB issue

Hi,
maybe this is similar to the ipoib issue discussed here around march this year.
"IPoIB on CentOS 6.5".

In the recent RHEL 6 Version it was fixed.
BZ#1187666<https://bugzilla.redhat.com/show_bug.cgi?id=1187666> solved it for our site.
Hope ther newest Redhat/CentOS 6.7 kernel is not faulty again.

Which kernel version you are using?

best regards
M.Soysal

On 14.10.2015 16:31, Andrus, Brian Contractor wrote:
Not sure if this is where I can get help with this, but...

Running standard CentOS 6.7
Mellanox infiniband adapters
Running OpenSM on one node

If that node reboots, suddenly it can no longer ping/be pinged across the ipoib network.
It can ping itself and it can ping any boxes that get rebooted.
Until they are rebooted, those boxes can ping each other but NOT any node that has been rebooted.

Almost like there is a new network and you have to reboot to join it...

Grrr...
ibhosts still shows all hosts. I can also ibping any node just fine. It is just the IPOIB bit that seems to choke.

Brian Andrus
ITACS/Research Computing
Naval Postgraduate School
Monterey, California
voice: 831-656-6238

_______________________________________________

Users mailing list

Users at lists.openfabrics.org<mailto:Users at lists.openfabrics.org>

http://lists.openfabrics.org/mailman/listinfo/users

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/users/attachments/20151014/ed347d9b/attachment.html>