LACP and Link Aggregation: Why Four Links Are Not Four Times the Speed

Link aggregation combines several physical links into one logical link, adding bandwidth and redundancy at the same time. LACP is the standard protocol that negotiates it. The single most important thing to understand is what aggregation does not give you: a single conversation does not get faster.

Why a flow cannot be split

TCP requires packets to arrive in order, or close enough that reordering does not trigger retransmission. If a switch sprayed one session's packets across four links with different queue depths, they would arrive out of order and performance would collapse.

So the switch hashes some combination of header fields — source and destination MAC addresses, IP addresses, and often port numbers — and uses the result to pick one link. All packets in a given flow take the same link, arriving in order.

The consequence: four bundled 1 Gbps links give 4 Gbps of aggregate capacity across many flows, and any single transfer is still capped at 1 Gbps. A question describing a single large file transfer that did not speed up after aggregation is describing this, and the answer is not a misconfiguration.

It also means distribution is only as even as the hash inputs are varied. Two switches exchanging traffic between a handful of servers may hash almost everything onto one link while the others idle. Hashing on Layer 4 ports rather than MAC addresses usually fixes it, since port numbers vary per session.

LACP and the alternatives

LACP is the open standard, originally 802.3ad and now part of 802.1AX. Both ends exchange control frames to agree that a bundle exists, which is what makes it safe: a link is only added to the bundle once the far end has confirmed it, so a miscabled port does not silently join.

Modes matter. Active sends LACP frames and actively seeks to form a bundle; passive responds but does not initiate. Active-to-active works, active-to-passive works, and passive-to-passive does not form a bundle at all because neither side speaks first. That pairing is a standard exam question.

Static aggregation forces the bundle with no protocol and no verification. It works when both ends are configured correctly and creates a loop when they are not, because one side starts forwarding across links the other treats as independent. Prefer LACP wherever both ends support it; use static only for devices that cannot negotiate.

Vendor-specific predecessors exist and behave similarly; the standard is what to know.

Configuration requirements

Every member link must match on speed, duplex, and media type, and the switchport configuration — access or trunk, VLAN membership, allowed VLANs — must be identical on every member. A mismatch prevents the bundle forming, or worse, forms a partial bundle that behaves unpredictably.

Apply configuration to the logical interface rather than to members individually, which is how most platforms prevent drift by design.

By default all links must terminate on the same physical switch at each end, which makes that switch a single point of failure. Multi-chassis aggregation — where two switches present themselves as one aggregation peer — removes it and is the standard data centre approach, though it adds its own complexity and its own failure modes.

Interaction with spanning tree

Worth understanding because it explains the benefit. Spanning tree blocks redundant paths to prevent loops, so four parallel links without aggregation means one forwarding and three blocked — redundancy with no added bandwidth.

An aggregated bundle appears to spanning tree as a single logical link, so all members forward. That is the real gain: converting blocked capacity into usable capacity. Path cost is calculated for the bundle as a whole, which also affects topology decisions, so review root bridge placement after aggregating — the priority settings covered under spanning tree priority.

Losing a member link does not bring the bundle down; capacity drops and flows are rehashed onto surviving links, which briefly disrupts sessions that move. Recovery is faster than spanning tree reconvergence, which is the availability argument — complementary to gateway redundancy through FHRP, which handles the router rather than the link.

Troubleshooting

A bundle that will not form is almost always a mismatch: speed, duplex, VLAN configuration, or both ends set to passive. Check those four before anything else.

Uneven utilization is the hashing algorithm meeting insufficiently varied traffic. A member flapping in and out points at a physical problem — rising error counters on that link, diagnosed as described under CRC and frame check sequence errors.

Exam angles

Know LACP as the 802.3ad/802.1AX standard for combining links into one logical link, providing aggregate bandwidth and redundancy. Know that per-flow hashing means a single session cannot exceed one member link's speed. Remember that passive-to-passive never forms a bundle, that all members must match in speed, duplex and VLAN configuration, and that the bundle appears to spanning tree as one link so all members forward.