What Is Link Aggregation?

Link aggregation joins several physical Ethernet links between two devices into one logical link, known as a link aggregation group or LAG. The switch, router or computer at each end treats the bundle as a single port: one MAC address, one entry in the forwarding table and one port in the spanning tree. Traffic is shared across the member links, and if one fails, its traffic moves to the others.

The standard is IEEE 802.1AX. It began in 2000 as IEEE 802.3ad, which is why that number still appears on datasheets and in Linux bonding mode names, and moved to 802.1AX in 2008. Vendors use their own names for the same idea: port trunking, EtherChannel, bonding on Linux and NIC teaming on Windows Server.

All member links must run at the same speed and in full duplex, and in the basic standard they must all connect the same two devices. Most switches allow up to eight active links in one group, with the exact limit set by the switch.

A LAG can be set up two ways. A static LAG is configured by hand on both ends and simply starts sending over every link that is up. LACP, the Link Aggregation Control Protocol, has the two ends exchange LACPDU messages to confirm they agree on which links belong to the bundle before using them, and keeps checking afterwards.

LACP is the safer choice. If someone patches one cable of a static LAG to the wrong port, the far end does not know and traffic can loop or vanish. LACP spots the mismatch and keeps that link out of the bundle. Each end runs in active mode, sending LACPDUs, or passive mode, answering them; at least one end must be active. A fast rate sends one LACPDU a second, so a link that stays up but stops passing traffic is removed after about three seconds; at the slow rate this can take around 90 seconds.

A switch does not split a single stream across the links, because the frames could then arrive out of order. Instead it runs a hash over each frame’s addresses, typically the source and destination MAC addresses, IP addresses or TCP and UDP ports, and sends every frame of the same conversation down the same member link. Two 1 Gb/s links therefore give 2 Gb/s in total, but any one flow is still limited to 1 Gb/s.

This matters when sizing a network. A LAG works well where many flows share an uplink, such as dozens of IP cameras feeding a recorder or many machines reporting to a server, because the hash spreads them across the links. It helps little for one large file transfer between two computers; for that, a faster single link such as 10 Gigabit Ethernet is the better fit.

A LAG protects the links between two devices but not the devices themselves. Industrial networks usually combine it with a redundancy protocol that covers switch failures as well.

A common industrial design runs a ring topology between cabinet switches for fast recovery, with a LAG on the uplinks from the ring to the server room, where traffic from the whole site comes together. Some vendors also offer multi-chassis LAG, which spreads one bundle across two switches so a whole switch can fail, but this is vendor-specific rather than part of the basic standard.

Most of our managed industrial switches support LACP, alongside VLANs, QoS and ring redundancy, so aggregated uplinks can be built from standard ports or fibre SFP slots.

We supply managed industrial switches and multi-port edge computers with LACP groups, VLANs and ring settings configured to your network design as standard, and confirm each aggregated link before the system ships.