Bonding - Link Aggregation
Linux bonding provides a method for aggregating multiple network interfaces (the slaves) into a single logical bonded interface (the bond). Cumulus RMP bonding supports the IEEE 802.3ad link aggregation mode. Link aggregation allows one or more links to be aggregated together to form a link aggregation group (LAG), such that a media access control (MAC) client can treat the link aggregation group as if it were a single link. The benefits of link aggregation are:
Linear scaling of bandwidth as links are added to LAG
The Cumulus RMP LAG control protocol is LACP version 1.
Example: Bonding 4 Slaves
In this example, front panel port interfaces swp1-swp4 are slaves in bond0 (swp5 and swp6 are not part of bond0).
The name of the bond is arbitrary as long as it follows Linux interface naming guidelines, and is unique within the switch. The only bonding mode supported in Cumulus RMP is 802.3ad. There are several 802.3ad settings that can be applied to each bond.
All of the following settings except for
bond-slaves are set to the
recommended defaults and should only be added to a configuration in
/etc/network/interfaces if you plan to use a different setting.
bond-slaves: The list of slaves in bond.
bond-mode: Is set to 802.3ad by default and must not be changed.
bond-miimon: How often the link state of each slave is inspected for link failures. It 100, the recommended value.
bond-use-carrier: How to determine link state. It defaults to 1.
bond-xmit-hash-policy: Hash method used to select the slave for a given packet; it defaults to layer3+4 and must not be changed.
bond-lacp-rate: Rate to ask link partner to transmit LACP control packets. It defaults to 1.
bond-min-links: Specifies the minimum number of links that must be active before asserting carrier on the bond. It defaults to 1, but a value greater than 1 is useful if higher level services need to ensure a minimum of aggregate bandwidth before putting the bond in service. Keeping
bond-min-linksset to 1 indicates the bond must have at least one active member for bond to assert carrier. If the number of active members drops below the
bond-min-linkssetting, the bond will appear to upper-level protocols as link-down. When the number of active links returns to greater than or equal to
bond-min-links, the bond will become link-up.
See Useful Links below for more details on settings.
To configure the bond, edit
/etc/network/interfaces and add a stanza
auto bond0 iface bond0 address 10.0.0.1/30 bond-slaves swp1 swp2 swp3 swp4
However, if you are intending that the bond become part of a bridge, you don’t need to specify an IP address. The configuration would look like this:
auto bond0 iface bond0 bond-slaves glob swp1-4
man interfaces for more information on
When networking is started on switch, bond0 is created as MASTER and
interfaces swp1-swp4 come up in SLAVE mode, as seen in the
ip link show command:
3: swp1: <BROADCAST,MULTICAST,SLAVE,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast master bond0 state UP mode DEFAULT qlen 500 link/ether 44:38:39:00:03:c1 brd ff:ff:ff:ff:ff:ff 4: swp2: <BROADCAST,MULTICAST,SLAVE,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast master bond0 state UP mode DEFAULT qlen 500 link/ether 44:38:39:00:03:c1 brd ff:ff:ff:ff:ff:ff 5: swp3: <BROADCAST,MULTICAST,SLAVE,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast master bond0 state UP mode DEFAULT qlen 500 link/ether 44:38:39:00:03:c1 brd ff:ff:ff:ff:ff:ff 6: swp4: <BROADCAST,MULTICAST,SLAVE,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast master bond0 state UP mode DEFAULT qlen 500 link/ether 44:38:39:00:03:c1 brd ff:ff:ff:ff:ff:ff
55: bond0: <BROADCAST,MULTICAST,MASTER,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP mode DEFAULT link/ether 44:38:39:00:03:c1 brd ff:ff:ff:ff:ff:ff
All slave interfaces within a bond will have the same MAC address as the bond. Typically, the first slave added to the bond donates its MAC address for the bond. The other slaves’ MAC addresses are set to the bond MAC address. The bond MAC address is used as source MAC address for all traffic leaving the bond, and provides a single destination MAC address to address traffic to the bond.
Egress traffic through a bond is distributed to a slave based on a packet hash calculation. This distribution provides load balancing over the slaves. The hash calculation uses packet header data to pick which slave to transmit the packet. For IP traffic, IP header source and destination fields are used in the calculation. For IP + TCP/UDP traffic, source and destination ports are included in the hash calculation. Traffic for a given conversation flow will always hash to the same slave. Many flows will be distributed over all the slaves to load balance the total traffic. In a failover event, the hash calculation is adjusted to steer traffic over available slaves.
Caveats and Errata
An interface cannot belong to multiple bonds.
Slave ports within a bond should all be set to the same speed/duplex, and should match the link partner’s slave ports.
A bond cannot enslave VLAN subinterfaces. A bond can have subinterfaces, but not the other way around.