Managing Mtu With Neutron
As of the OpenStack Mitaka release, new configuration options were introduced for Neutron related to MTU. Along those lines, some considerations were outlined with regards to MTU and Jumbo Frames, specifically. Related to the cases which are outlined and the related configuration options, below are our recommendations – Platform9 – which we've validated to be most suitable.
global_physnet_mtu
1500
MTU for all networks flat, VLAN, and overlay. This becomes the MTU set on internal tenant network ports such as qr, qvo, qvb, etc... and what is advertised to instances to auto-configure via DHCP. Leave untouched unless they want to use custom MTU such as jumbo frames. For overlay networks, this should be value of physical overlay layer. If left at 1500, VMs and internal ports on tunnel networks get MTU of 1450 automatically, to account for the VXLAN header added by br-tun before egress.
path_mtu
0
Override for overlay networks. Defaults to 0, meaning not used and overlay networks use global_physnet_ mtu. *For overlay networks such as VXLAN/GRE, neutron automatically subtracts the overlay protocol overhead from this value. Leave untouched (at 0) unless their overlay network is on a different physical network than for flat/VLAN traffic and uses different MTU
[DEFAULT]
global_physnet_mtu = 9000Info
The above parameter would need to be set on the Platform9 Deployment Unit (DU) which hosts the control plane components – including Neutron. Please submit a ticket and our Support team will be able to facilitate this request.
If you should wish to override the global physical network MTU – the following would need to be explicitly set on each hypervisor, according to the networks defined, e.g.
physical_network_mtus = provider:9000,provider2:9000Warning
For any existing networks, this will not affect their current MTU value. It will need to be set via the CLI/API, as outlined in our related Support article: Change MTU Size For Existing Network Using OpenStack CLI/API.
Last updated
Was this helpful?