ml2/ovs: veth_mtu has no effect

[Emilien Macchi]

This bug could be invalid, but anyway.

Environment: OpenStack Havana 2013.2 / Debian Wheezy / Linux kernel 3.12 / OVS 2.0.1 / Neutron ML2 Plugin with OVS Agent

ml2_conf.ini: http://paste.openstack.org/show/jcbw7UmsIYjrCpSLaWi3/

When I spawn a VM, both the TAP (hosted on compute) and the eth (in the VM) continue to have 1500 as MTU, while I'm trying to enforce 1495.

The only one way I found was to use DEFAULT/dnsmasq_config_file=/etc/neutron/dnsmasq-neutron.conf parameter in DHCP agent configuration file and create /etc/neutron/dnsmasq-neutron.conf whith "dhcp-options-force=26,1495".

Am I missing something?

[Cedric Brandily (cbrandily)]

veth_mtu enforces mtu on veths (between br-int and physnets only).

IMHO, you can not specify on compute side the mtu on vm side.
The dhcp option seems the correct way to configure the mtu inside the vm.

[Emilien Macchi (emilienm)]

You're right, on my compute:

10: br-int: <BROADCAST,UP,LOWER_UP> mtu 1495 qdisc noqueue state UNKNOWN
12: br-1217: <BROADCAST,UP,LOWER_UP> mtu 1495 qdisc noqueue state UNKNOWN
97: phy-br-1217: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1495 qdisc pfifo_fast master ovs-system state UP qlen 1000
98: int-br-1217: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1495 qdisc pfifo_fast master ovs-system state UP qlen 1000

In my case, veth_mtu seems useless since I've been enforcing the MTU directly inside the VM using DHCP options.