[Rdo-list] Slow network performance on Kilo?

[Arkady_Kanevsky at dell.com]

And you will need switch settings to match that MTU.

-----Original Message-----
From: rdo-list-bounces at redhat.com [mailto:rdo-list-bounces at redhat.com] On Behalf Of Dan Sneddon
Sent: Friday, January 29, 2016 4:17 PM
To: rdo-list at redhat.com
Subject: Re: [Rdo-list] Slow network performance on Kilo?

On 01/27/2016 12:30 PM, Erich Weiler wrote:
> Hi Y'all,
>
> I've seen several folks on the net with this problem, but I'm still
> flailing a bit as to what is really going on.
>
> We are running RHEL 7 with RDO OpenStack Kilo.
>
> We are setting this environment up still, not quite done yet. But in
> our testing, we are experiencing very slow network performance when
> downloading or uploading to and from VMs. We get like 300Kb/s or so.
>
> We are using Neutron, MTU 9000 everywhere. I've tried disabling GSO,
> LRO, TSO, GRO on the neutron interfaces, as well as the VM server
> interfaces, still no improvement. I've tried lowing the VM MTU to
> 1500, still no improvement. It's really strange. We do get
> connectivity, I can ssh to the instances, but the network performance
> is just really, really slow. It appears the instances can talk to
> each other very quickly however. They just get slow network to the
> internet (i.e. when packets go through the network node).
>
> We are using VLAN tenant network isolation.
>
> Can anyone point me in the right direction? I've been beating my head
> against a wall and googling without avail for a week...
>
> Many thanks,
> erich
>
> _______________________________________________
> Rdo-list mailing list
> Rdo-list at redhat.com
> https://www.redhat.com/mailman/listinfo/rdo-list
>
> To unsubscribe: rdo-list-unsubscribe at redhat.com

To use jumbo frames with Neutron, you first need to make sure that all layers have the same MTU: bridge, interface, VLAN, and bond (if applicable).

Check these settings on both Controllers and Computes:

In /etc/nova/nova.conf, edit or create this line:
network_device_mtu=9000

In /etc/neutron/plugins/openvswitch/ovs_neutron_plugin.ini:
veth_mtu = 9000

That will increase the MTU on the virtual Ethernet adapter to. You can get away with an MTU here that matches the wire in Neutron VLAN mode.
In VXLAN you'd want this value to be 50 bytes less than the MTU on the wire.

You also want to check the DHCP options in the file /etc/neutron/dnsmasq-neutron.conf on all controllers:
dhcp-option-force=26,9000

You can also try setting that last option to 1400, that will limit the VM MTU, but if things speed up then you know you've got a bottleneck somewhere (like a less-than 1500-byte MTU somewhere in the path of VLAN traffic).

--
Dan Sneddon | Principal OpenStack Engineer
dsneddon at redhat.com | redhat.com/openstack
650.254.4025 | dsneddon:irc @dxs:twitter

_______________________________________________
Rdo-list mailing list
Rdo-list at redhat.com
https://www.redhat.com/mailman/listinfo/rdo-list

To unsubscribe: rdo-list-unsubscribe at redhat.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rdoproject.org/pipermail/dev/attachments/20160129/85f89e40/attachment.html>