Chapter 11. Additional network configuration

Individual interfaces might require modification. The example below shows modifications required to use the second NIC to connect to an infrastructure network with DHCP addresses, and to use the third and fourth NICs for the bond:

network_config:
  # Add a DHCP infrastructure network to nic2
  - type: interface
    name: nic2
    use_dhcp: true
  - type: ovs_bridge
    name: br-bond
    members:
      - type: ovs_bond
        name: bond1
        ovs_options:
          get_param: BondInterfaceOvsOptions
        members:
          # Modify bond NICs to use nic3 and nic4
          - type: interface
            name: nic3
            primary: true
          - type: interface
            name: nic4

The network interface template uses either the actual interface name (eth0, eth1, enp0s25) or a set of numbered interfaces (nic1, nic2, nic3). The network interfaces of hosts within a role do not have to be exactly the same when using numbered interfaces (nic1, nic2, etc.) instead of named interfaces (eth0, eno2, etc.). For example, one host might have interfaces em1 and em2, while another has eno1 and eno2, but you can refer to the NICs of both hosts as nic1 and nic2.

The order of numbered interfaces corresponds to the order of named network interface types:

The numbered NIC scheme only takes into account the interfaces that are live, for example, if they have a cable attached to the switch. If you have some hosts with four interfaces and some with six interfaces, you should use nic1 to nic4 and only plug four cables on each host.

You can hardcode physical interfaces to specific aliases. This allows you to be pre-determine which physical NIC will map as nic1 or nic2 and so on. You can also map a MAC address to a specified alias.

Interfaces are mapped to aliases using an environment file. In this example, each node has predefined entries for nic1 and nic2:

parameter_defaults:
  NetConfigDataLookup:
    node1:
      nic1: "em1"
      nic2: "em2"
    node2:
      nic1: "00:50:56:2F:9F:2E"
      nic2: "em2"

The resulting configuration is applied by os-net-config. On each node, you can see the applied configuration under interface_mapping in /etc/os-net-config/mapping.yaml.

There are two ways to set the default route of a host. If the interface is using DHCP and the DHCP server offers a gateway address, the system uses a default route for that gateway. Otherwise, you can set a default route on an interface with a static IP.

Although the Linux kernel supports multiple default gateways, it only uses the one with the lowest metric. If there are multiple DHCP interfaces, this can result in an unpredictable default gateway. In this case, it is recommended to set defroute: false for interfaces other than the one using the default route.

For example, you might want a DHCP interface (nic3) to be the default route. Use the following YAML to disable the default route on another DHCP interface (nic2):

# No default route on this DHCP interface
- type: interface
  name: nic2
  use_dhcp: true
  defroute: false
# Instead use this DHCP interface as the default route
- type: interface
  name: nic3
  use_dhcp: true

To set a static route on an interface with a static IP, specify a route to the subnet. For example, you can set a route to the 10.1.2.0/24 subnet through the gateway at 172.17.0.1 on the Internal API network:

    - type: vlan
      device: bond1
      vlan_id:
        get_param: InternalApiNetworkVlanID
      addresses:
      - ip_netmask:
          get_param: InternalApiIpSubnet
      routes:
      - ip_netmask: 10.1.2.0/24
        next_hop: 172.17.0.1

The Maximum Transmission Unit (MTU) setting determines the maximum amount of data transmitted with a single Ethernet frame. Using a larger value results in less overhead since each frame adds data in the form of a header. The default value is 1500 and using a higher value requires the configuration of the switch port to support jumbo frames. Most switches support an MTU of at least 9000, but many are configured for 1500 by default.

The MTU of a VLAN cannot exceed the MTU of the physical interface. Ensure that you include the MTU value on the bond and/or interface.

The Storage, Storage Management, Internal API, and Tenant networks all benefit from jumbo frames. In testing, a project’s networking throughput demonstrated substantial improvement when using jumbo frames in conjunction with VXLAN tunnels.

- type: ovs_bond
  name: bond1
  mtu: 9000
  ovs_options: {get_param: BondInterfaceOvsOptions}
  members:
    - type: interface
      name: nic3
      mtu: 9000
      primary: true
    - type: interface
      name: nic4
      mtu: 9000

# The external interface should stay at default
- type: vlan
  device: bond1
  vlan_id:
    get_param: ExternalNetworkVlanID
  addresses:
    - ip_netmask:
        get_param: ExternalIpSubnet
  routes:
    - ip_netmask: 0.0.0.0/0
      next_hop:
        get_param: ExternalInterfaceDefaultRoute

# MTU 9000 for Internal API, Storage, and Storage Management
- type: vlan
  device: bond1
  mtu: 9000
  vlan_id:
    get_param: InternalApiNetworkVlanID
  addresses:
  - ip_netmask:
      get_param: InternalApiIpSubnet

Neutron uses a default empty string for its external bridge mapping. This maps the physical interface to the br-int instead of using br-ex directly. This model allows multiple Floating IP networks using either VLANs or multiple physical connections.

Use the NeutronExternalNetworkBridge parameter in the parameter_defaults section of your network isolation environment file:

  parameter_defaults:
    # Set to "br-ex" when using floating IPs on the native VLAN
    NeutronExternalNetworkBridge: "''"

If you use only one Floating IP network on the native VLAN of a bridge, you can optionally set the neutron external bridge. This results in the packets only having to traverse one bridge instead of two, which might result in slightly lower CPU usage when passing traffic over the Floating IP network.

If a trunked interface or bond has a network on the native VLAN, the IP addresses are assigned directly to the bridge and is no VLAN interface.

For example, if the External network is on the native VLAN, a bonded configuration looks like this:

network_config:
  - type: ovs_bridge
    name: bridge_name
    dns_servers:
      get_param: DnsServers
    addresses:
      - ip_netmask:
          get_param: ExternalIpSubnet
    routes:
      - ip_netmask: 0.0.0.0/0
        next_hop:
          get_param: ExternalInterfaceDefaultRoute
    members:
      - type: ovs_bond
        name: bond1
        ovs_options:
          get_param: BondInterfaceOvsOptions
        members:
          - type: interface
            name: nic3
            primary: true
          - type: interface
            name: nic4