MTU considerations

MTU considerations

The Networking service uses the MTU of the underlying physical network to calculate the MTU for virtual network components including instance network interfaces. By default, it assumes a standard 1500-byte MTU for the underlying physical network.

The Networking service only references the underlying physical network MTU. Changing the underlying physical network device MTU requires configuration of physical network devices such as switches and routers.

Jumbo frames

The Networking service supports underlying physical networks using jumbo frames and also enables instances to use jumbo frames minus any overlay protocol overhead. For example, an underlying physical network with a 9000-byte MTU yields a 8950-byte MTU for instances using a VXLAN network with IPv4 endpoints. Using IPv6 endpoints for overlay networks adds 20 bytes of overhead for any protocol.

The Networking service supports the following underlying physical network architectures. Case 1 refers to the most common architecture. In general, architectures should avoid cases 2 and 3.

Note

You can trigger MTU recalculation for existing networks by changing the MTU configuration and restarting the neutron-server service. However, propagating MTU calculations to the data plane may require users to delete and recreate ports on the network.

When using the Open vSwitch or Linux bridge drivers, new MTU calculations will be propogated automatically after restarting the l3-agent service.

Case 1

For typical underlying physical network architectures that implement a single MTU value, you can leverage jumbo frames using two options, one in the neutron.conf file and the other in the ml2_conf.ini file. Most environments should use this configuration.

For example, referencing an underlying physical network with a 9000-byte MTU:

  1. In the neutron.conf file:

    [DEFAULT]
    global_physnet_mtu = 9000
    
  2. In the ml2_conf.ini file:

    [ml2]
    path_mtu = 9000
    

Case 2

Some underlying physical network architectures contain multiple layer-2 networks with different MTU values. You can configure each flat or VLAN provider network in the bridge or interface mapping options of the layer-2 agent to reference a unique MTU value.

For example, referencing a 4000-byte MTU for provider2, a 1500-byte MTU for provider3, and a 9000-byte MTU for other networks using the Open vSwitch agent:

  1. In the neutron.conf file:

    [DEFAULT]
    global_physnet_mtu = 9000
    
  2. In the openvswitch_agent.ini file:

    [ovs]
    bridge_mappings = provider1:eth1,provider2:eth2,provider3:eth3
    
  3. In the ml2_conf.ini file:

    [ml2]
    physical_network_mtus = provider2:4000,provider3:1500
    path_mtu = 9000
    

Case 3

Some underlying physical network architectures contain a unique layer-2 network for overlay networks using protocols such as VXLAN and GRE.

For example, referencing a 4000-byte MTU for overlay networks and a 9000-byte MTU for other networks:

  1. In the neutron.conf file:

    [DEFAULT]
    global_physnet_mtu = 9000
    
  2. In the ml2_conf.ini file:

    [ml2]
    path_mtu = 4000
    

    Note

    Other networks including provider networks and flat or VLAN self-service networks assume the value of the global_physnet_mtu option.

Instance network interfaces (VIFs)

The DHCP agent provides an appropriate MTU value to instances using IPv4, while the L3 agent provides an appropriate MTU value to instances using IPv6. IPv6 uses RA via the L3 agent because the DHCP agent only supports IPv4. Instances using IPv4 and IPv6 should obtain the same MTU value regardless of method.

Creative Commons Attribution 3.0 License

Except where otherwise noted, this document is licensed under Creative Commons Attribution 3.0 License. See all OpenStack Legal Documents.