Open Stack

On Mon, Nov 14, 2022 at 8:26 AM Franck VEDEL <
franck.vedel at univ-grenoble-alpes.fr> wrote:

> Hello.
> Thanks a lot Erik my problem was the MTU.
> if I go back to a situation with MTU=1500 everywhere, all is working fine
> !!!
>
> Is the following configuration possible and if so, how to configure with
> kolla-ansible files ? :
>
> 3 networks:
>   - external (2 externals, VLAN 10 and VLAN 20): MTU = 1500
> - admin:MTU=1500
> - management : MTU = 9000 (a scsi bay stores volumes, with mtu 9000 ok).
>
> Like this:
> `
>

It is possible, but in some ways not advisable. Just from a general
networking standpoint, I wouldn't set any interface used for traffic coming
to / from the internet to use Jumbo Frames. Strange things happen when you
start fragmenting to fit through standard internet routers, particularly
when you run into something on the other end that is also using a large
MTU. It's fine for internal management networks, storage networks, and the
like. You could move your tenant / tunneling vlan over to a different
interface and let that other one serve your internal needs.

That being said, you need to account for VXLAN encapsulation overhead in
your MTU considerations. Whatever your physical interface config is set to,
your  tenant networks need to use 50 bytes less. I think this is fine by
default when using 1500, but can get weird when using Jumbo frames. If you
put an override config file in /etc/kolla/config/neutron/ml2_ini.conf with
something like:

[ml2]
path_mtu = 9000

it should tell Neutron to take that into account.

-Erik

Thanks a lot if you have a solution for this.
> If  impossible, I stay with 1500… it’s working, no problem.
>
> Franck
>
>
> Le 12 nov. 2022 à 21:00, Franck VEDEL <franck.vedel at univ-grenoble-alpes.fr>
> a écrit :
>
> 3) Networking issues like mismatched MTU
>
>
> My MTU (between nodes ) is 9000….
>
> I believe my problem is the  MTU.
>
> I modified /etc/kolla/config/neutron.conf and
> /etc/kolla/config/neutron/ml2_conf.ini.conf
> then kolla-ansible -i multinode reconfigures
>
> (case 1 here:
> https://docs.openstack.org/newton/networking-guide/config-mtu.html)
>
> I test again everything and functions that did not work work again but not
> all....
>
> For example, instances get an ip through dhcp but can't ping the router,
> but on some networks it works.
> However, before the reboot of the servers, I had not had a problem with
> the MTU of 9000.
>
> I'm going back to a 1500 MTU on Monday on site.
>
> Thank you Eric!!!
>
> Franck VEDEL
>
>
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.openstack.org/pipermail/openstack-discuss/attachments/20221116/a0d7e24f/attachment-0001.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Capture d?e?cran 2022-11-14 a? 14.25.19.png
Type: image/png
Size: 273266 bytes
Desc: not available
URL: <https://lists.openstack.org/pipermail/openstack-discuss/attachments/20221116/a0d7e24f/attachment-0001.png>