Red Hat Training

A Red Hat training course is available for Red Hat OpenStack Platform

Chapter 13. Network Interface Bonding

This chapter defines some of the bonding options you can use in your custom network configuration.

13.1. Network Interface Bonding and Link Aggregation Control Protocol (LACP)

You can bundle multiple physical NICs together to form a single logical channel known as a bond. You can configure bonds to provide redundancy for high availability systems or increased throughput.

Important

Using single root I/O virtualization (SR-IOV) on bonded interfaces is not supported.

Red Hat OpenStack Platform supports Linux bonds, Open vSwitch (OVS) kernel bonds, and OVS-DPDK bonds.

The bonds can be used with the optional Link Aggregation Control Protocol (LACP). LACP is a negotiation protocol that creates a dynamic bond for load balancing and fault tolerance.

Red Hat recommends the use of Linux kernel bonds (bond type: linux_bond) over OvS kernel bonds (bond type: ovs_bond). User mode bonds (bond type: ovs_dpdk_bond) must be used with user mode bridges (type: ovs_user_bridge) as opposed to kernel mode bridges (type: ovs_bridge). However, do not combine ovs_bridge and ovs_user_bridge on the same node.

On control and storage networks, Red Hat recommends the use of Linux bonds with VLAN and LACP, because OVS bonds carry the potential for control plane disruption that can occur when OVS or the neutron agent is restarted for updates, hot fixes, and other events. The Linux bond/LACP/VLAN configuration provides NIC management without the OVS disruption potential.

Here is an example configuration of a Linux bond with one VLAN. 

params:
            $network_config:
              network_config:

              - type: linux_bond
                name: bond_api
                bonding_options: "mode=active-backup"
                use_dhcp: false
                dns_servers:
                  get_param: DnsServers
                members:
                - type: interface
                  name: nic3
                  primary: true
                - type: interface
                  name: nic4

              - type: vlan
                vlan_id:
                  get_param: InternalApiNetworkVlanID
                device: bond_api
                addresses:
                - ip_netmask:
                    get_param: InternalApiIpSubnet

The following example shows a Linux bond plugged into the OVS bridge 

params:
            $network_config:
              network_config:

            -  type: ovs_bridge
                name: br-tenant
                use_dhcp: false
                mtu: 9000
                members:
                  - type: linux_bond
                    name: bond_tenant
                    bonding_options: "mode=802.3ad updelay=1000 miimon=100"
                    use_dhcp: false
                    dns_servers:
                      get_param: DnsServers
                    members:
                    - type: interface
                      name: p1p1
                      primary: true
                    - type: interface
                      name: p1p2
                  - type: vlan
                    device: bond_tenant
                    vlan_id: {get_param: TenantNetworkVlanID}
                    addresses:
                      -
                        ip_netmask: {get_param: TenantIpSubnet}

The following example shows an OVS user space bridge: 

params:
            $network_config:
              network_config:

          -    type: ovs_user_bridge
                name: br-ex
                use_dhcp: false
                members:
                - type: ovs_dpdk_bond
                  name: dpdkbond0
                  mtu: 2140
                  ovs_options: {get_param: BondInterfaceOvsOptions}
                  #ovs_extra:
                  #- set interface dpdk0 mtu_request=$MTU
                  #- set interface dpdk1 mtu_request=$MTU
                  rx_queue:
                    get_param: NumDpdkInterfaceRxQueues
                  members:
                  - type: ovs_dpdk_port
                    name: dpdk0
                    mtu: 2140
                    members:
                    - type: interface
                      name: p1p1
                  - type: ovs_dpdk_port
                    name: dpdk1
                    mtu: 2140
                    members:
                    - type: interface
                      name: p1p2

13.2. Open vSwitch Bonding Options

The overcloud provides networking through Open vSwitch (OVS). Use the following table to understand support compatibility for OVS kernel and OVS-DPDK for bonded interfaces. The OVS/OVS-DPDK balance-tcp mode is available as a technology preview only.

Note

This support requires Open vSwitch 2.9 or later.

OVS Bond mode

Application

Notes

Compatible LACP options

active-backup

High availability (active-passive)

 

active, passive, or off

balance-slb

Increased throughput (active-active)

  • Performance is affected by extra parsing per packet.
  • There is a potential for vhost-user lock contention.

active, passive, or off

balance-tcp (tech preview only )

Not recommended (active-active)

  • Recirculation needed for L4 hashing has a performance impact.
  • As with balance-slb, performance is affected by extra parsing per packet and there is a potential for vhost-user lock contention.
  • LACP must be enabled.

active or passive

You can configure a bonded interface in the network environment file using the BondInterfaceOvsOptions parameter as shown in this example: 

parameter_defaults:
  BondInterfaceOvsOptions: "bond_mode=balance-slb"

13.3. Linux bonding options

You can use LACP with Linux bonding in your network interface templates: 

      - type: linux_bond
        name: bond1
        members:
        - type: interface
          name: nic2
        - type: interface
          name: nic3
        bonding_options: "mode=802.3ad lacp_rate=[fast|slow] updelay=1000 miimon=100"
  • mode - enables LACP.
  • lacp_rate - defines whether LACP packets are sent every 1 second, or every 30 seconds.
  • updelay - defines the minimum amount of time that an interface must be active before it is used for traffic (this helps mitigate port flapping outages).
  • miimon - the interval in milliseconds that is used for monitoring the port state using the driver’s MIIMON functionality.

For more information on Linux bonding options, see 7.7.1. Bonding Module Directives in the Red Hat Enterprise Linux 7 Networking Guide.

13.4. OVS bonding options

The following table provides some explanation of these options and some alternatives depending on your hardware.

Table 13.1. Bonding Options

bond_mode=balance-slb

Balances flows based on source MAC address and output VLAN, with periodic rebalancing as traffic patterns change. Bonding with balance-slb allows a limited form of load balancing without the remote switch’s knowledge or cooperation. SLB assigns each source MAC and VLAN pair to a link and transmits all packets from that MAC and VLAN through that link. This mode uses a simple hashing algorithm based on source MAC address and VLAN number, with periodic rebalancing as traffic patterns change. This mode is similar to mode 2 bonds used by the Linux bonding driver. This mode can be used to provide load balancing even when the switch is not configured to use LACP.

bond_mode=active-backup

This mode offers active/standby failover where the standby NIC resumes network operations when the active connection fails. Only one MAC address is presented to the physical switch. This mode does not require any special switch support or configuration, and works when the links are connected to separate switches. This mode does not provide load balancing.

lacp=[active|passive|off]

Controls the Link Aggregation Control Protocol (LACP) behavior. Only certain switches support LACP. If your switch does not support LACP, use bond_mode=balance-slb or bond_mode=active-backup.

other-config:lacp-fallback-ab=true

Sets the LACP behavior to switch to bond_mode=active-backup as a fallback.

other_config:lacp-time=[fast|slow]

Set the LACP heartbeat to 1 second (fast) or 30 seconds (slow). The default is slow.

other_config:bond-detect-mode=[miimon|carrier]

Set the link detection to use miimon heartbeats (miimon) or monitor carrier (carrier). The default is carrier.

other_config:bond-miimon-interval=100

If using miimon, set the heartbeat interval in milliseconds.

bond_updelay=1000

Number of milliseconds a link must be up to be activated to prevent flapping.

other_config:bond-rebalance-interval=10000

Milliseconds between rebalancing flows between bond members. Set this value to zero to disable rebalancing flows between bond members.