One thing to be aware of when using overlay networking technologies is that the additional headers added to the encapsulated packets may cause them to exceed the maximum transmission unit (MTU) of the switchport or interface. The MTU is the largest size of packet or frame that can be sent over the network. Encapsulating a packet with VXLAN headers may cause the packet size to exceed the default maximum 1500-byte MTU. Connection issues caused by exceeding the MTU manifest themselves in strange ways, including partial failures in connecting to instances over SSH or a failure to transfer large payloads between instances, and more. To avoid this, consider lowering the MTU of interfaces within virtual machine instances from 1500 bytes to 1450 bytes to account for the overhead of VXLAN encapsulation to avoid connectivity issues.
An alternative to dropping the MTU is to increase the MTU of the interfaces used for the VTEPs. It is common to set a jumbo MTU of 9000 on VTEP interfaces and corresponding switchports to avoid having to drop the MTU inside instances. Increasing the MTU of the VTEP interfaces has also been shown to provide increases in network throughput when using overlay networks.
The DHCP agent can be configured to push a non-standard MTU to instances within the DHCP lease offer by modifying DHCP option 26. To configure a lower MTU, complete the following steps:
- On the controller node, modify the DHCP configuration file at /etc/neutron/dhcp_agent.ini and specify a custom dnsmasq configuration file:
[DEFAULT]
dnsmasq_config_file = /etc/neutron/dnsmasq-neutron.conf
- Next, create the custom dnsmasq configuration file at /etc/neutron/dnsmasq-neutron.conf and add the following contents:
dhcp-option-force=26,1450
- Save and close the file. Restart the Neutron DHCP agent with the following command:
# systemctl restart neutron-dhcp-agent
Inside an instance running Linux, the MTU can be observed within the instance using the ip link show <interface> command.