link aggregation
Link aggregation bundles several physical cables between two devices so they behave as one fat, faster link. Think of widening a one-lane road into four lanes: more cars get through at once, and if one lane is closed the others keep traffic flowing. Bundling four 10 Gbps links gives roughly 40 Gbps of aggregate capacity and survives the loss of any single cable.
The bundle is treated as a single logical link with a single MAC identity, so the Spanning Tree Protocol sees one link, not several, and does not block the extras as loops. Traffic is spread across the member links by hashing each flow's headers — for example, source and destination MAC or IP and port — so that all the packets of one conversation take the same physical wire (keeping them in order) while different conversations land on different wires. The two ends usually agree on the bundle automatically using LACP (the Link Aggregation Control Protocol, 802.3ad), which also detects a dead member and stops sending over it. If a cable fails, its flows are rehashed onto the survivors with no manual intervention.
Link aggregation is the everyday way to add bandwidth and redundancy between a server and a switch, or between two switches, without buying faster interfaces. The honest caveat: it is not magic per-packet load balancing. Because each flow sticks to one member link, a single huge transfer is capped at the speed of one member, not the whole bundle — aggregation raises total throughput across many flows, not the speed of any one conversation.
Two switches joined by four 10 Gbps cables in one LACP bundle. Hundreds of user flows are hashed across the four links for ~40 Gbps total. But one user copying a single giant file rides one member link, so that copy tops out near 10 Gbps, not 40.
Aggregate bandwidth grows with flows; one flow is still capped at one member link's speed.
Both ends must agree on the bundle (ideally via LACP); plugging one end's links into separate, unconfigured ports creates a loop instead of an aggregate.