Fri, Jun 26, 2015 at 09:23:34AM CEST, gslavic(a)gmail.com wrote:
Hello Jiri,
Are there any news regarding the teaming problem we at SOX had ?
Did you manage to find time to investigate it on RH7 or ask somebody
in RH for opinion ?
I am under some pressure to solve this problem quickly (even if it
means going back to classical bonding) :-(
Can you give me the advice what to do:
- Change the port aggregation from teaming to bonding.
- Try to upgrade/reinstall kernel team module (how ?!? - I know
Linux - but I don't know it that well to just-change-kernel-modules on the
fly :-) )
- Give you some other useful data that can help you find some other
solution.
I would highly appreciate any (and I really mean ANY) help, advice
or suggestion you can give us regarding this problem.
Would you be able to compile and test latest net-next kernel?
https://git.kernel.org/cgit/linux/kernel/git/davem/net-next.git
That will tell us more about if the issue still exists in upstream of if
it is just in rhel/centos.
Thanks
Best regards
G.Slavic
-----Original Message-----
From: Jiri Pirko [mailto:jiri@resnulli.us]
Sent: Monday, 08 June 2015 12:22
To: Goran Slavić
Subject: Re: Problems with teaming ...
Mon, Jun 08, 2015 at 12:10:03PM CEST, gslavic(a)sox.rs wrote:
>Not today (I am over my head is some other work) but I will contact you
>tomorrow/the day after tomorrow.
>
>What is your concern ? / What does this program do ? Problem is that any
>installation or testing on this server must be approved from "higher up"
>because this is very important server to us :-)
>
>If you think that the problem is with the switch that this server is
>connected to or some other upstream equipment - I can tell you that this is
>probably not the issue. If you see the graphs from my original post the
>switch has been balancing traffic to the server almost perfectly (5-10%
>plus/minus) and that same switch has 2 other LACP aggregated port groups
>where classical bonding works in communication with other servers.
I'm 99,9% sure that the problem is inside kernel team module. For me, on
upstream kernel the issue does not happen. Looks like this is limited to
rhel7/centos7.
I would test it on rhel7, unfortunatelly I'm no longer Red Hat employee
so I don't have access to rhel7 machine.
Will try ask someone from RH.