// networking fundamentals · intermediate

EtherChannel Explained: Link Aggregation, LACP, PAgP and Port-Channels

14 min read· Updated 2 June 2026 · By TechDirectory Editorial Team
In a nutshell: EtherChannel is Cisco's name for bundling multiple physical Ethernet links into one logical interface called a Port-Channel. The standards-based name is usually link aggregation, and the common negotiation protocol is LACP from IEEE 802.1AX. The point is simple: use several parallel links for more aggregate bandwidth and redundancy while making the switch, router, firewall or server treat them as one logical connection.

Quick definitions

TermWhat it means
EtherChannelCisco terminology for bundling multiple Ethernet links into one logical channel.
Port-ChannelThe logical interface created by the bundle, often written as Port-channel1, Po1 or ae0 depending on vendor.
LAGLink Aggregation Group. Vendor-neutral term for a bundle of links that behave like one logical link.
LACPLink Aggregation Control Protocol. Standards-based negotiation protocol defined in IEEE link aggregation standards.
PAgPPort Aggregation Protocol. Cisco-proprietary negotiation protocol for forming EtherChannels.
Static on modeA forced EtherChannel with no negotiation protocol. Works only when both sides are configured correctly.
HashingThe algorithm that maps a traffic flow to one physical member link.
Member linkOne physical Ethernet interface that participates in the bundle.

Watch the walkthrough

Video walkthrough:

This article expands on the linked EtherChannel lesson, including the before-and-after STP example, the hash-based load-balancing model, LACP and PAgP modes, and the commands used to verify a working bundle.

Why EtherChannel exists

Imagine two switches connected by one 1 Gbps cable. If that cable fails, the switches lose the inter-switch path. The obvious fix is to add a second cable. But in a normal Layer 2 network, two parallel links create a loop, and Spanning Tree Protocol blocks one of them to protect the network. You gain redundancy, but not active bandwidth.

EtherChannel solves that by making multiple physical links appear as one logical link. Two 1 Gbps links can become one 2 Gbps aggregate bundle. Four 10 Gbps links can become a 40 Gbps aggregate bundle. One flow does not become faster than a single member link, but many flows can share the bundle.

Without EtherChannelWith EtherChannel
Multiple physical links can create a Layer 2 loop.The bundle appears as one logical interface.
STP may block one or more redundant links.STP sees one port-channel and can leave all member links forwarding.
Bandwidth is often limited to one active link.Aggregate capacity can use multiple links.
A failed link can trigger topology changes.Traffic is redistributed across remaining members with less disruption.
Configuration is repeated per physical interface.Most settings live on the Port-Channel interface.

How it changes STP behaviour

Spanning Tree does not see the individual links inside a healthy Layer 2 EtherChannel as separate topology choices. It sees the Port-Channel. That is the magic from an operations point of view: you can connect switches with several cables without asking STP to choose one and block the rest.

If one member link fails, the Port-Channel can remain up as long as enough active members remain. Traffic that hashed to the failed link is remapped to surviving links. There may be brief packet loss during physical failure and hardware/software reconvergence, but it is usually far less disruptive than a full STP topology change.

Important distinction: EtherChannel does not remove the need for STP in Layer 2 networks. It makes parallel links safer by presenting them as one logical link. You still need correct STP root placement, VLAN design and loop prevention.

How load balancing really works

EtherChannel load balancing is usually per-flow, not per-packet. The switch examines fields such as source MAC, destination MAC, source IP, destination IP or Layer 4 ports, then runs those fields through a hash algorithm. The result chooses one physical member link for that flow.

This is why a single file transfer between two hosts normally cannot exceed the speed of one member link. If the flow hashes to GigabitEthernet1/0/1, all packets in that flow use that member. Sending packets from one flow across multiple links would risk out-of-order delivery, which can damage TCP performance.

Load-balancing choiceBest fitRisk
src-macMany different source devices behind the local switch.Poor distribution if one server sends most traffic.
dst-macMany destinations behind the far switch.Poor distribution if most traffic targets one MAC.
src-dst-macLayer 2 campus links with many source and destination hosts.Still blind to IP and application port patterns.
src-ipRouted or multi-subnet traffic with many source IPs.Can polarise traffic from a few busy hosts.
dst-ipTraffic to many destination IPs.Can polarise if traffic mostly targets one server.
src-dst-ipCommon default choice for routed or mixed traffic patterns.May still not distinguish many sessions between the same host pair.
src-dst-port or mixed IP/portMany TCP/UDP sessions between the same hosts, where supported.Availability and command syntax vary by platform.

The right hash depends on traffic. A campus uplink, a server uplink, a wireless controller link and a router-to-router link may need different choices. Always verify the platform's supported options with show etherchannel load-balance or the equivalent vendor command.

Static, PAgP and LACP modes

A Port-Channel can be forced on statically or negotiated dynamically. Dynamic negotiation is safer because each side exchanges control messages and confirms that the partner agrees to form the bundle.

MethodModesForms whenNotes
Static EtherChannelonBoth sides are forced onNo negotiation. Fast to configure but dangerous if one side is wrong.
PAgPdesirable / autoAt least one side is desirableCisco proprietary. Desirable actively negotiates; auto waits.
LACPactive / passiveAt least one side is activeStandards-based and widely preferred across mixed-vendor networks.
Mode pairExpected result
LACP active + activeForms.
LACP active + passiveForms.
LACP passive + passiveDoes not form because neither side starts negotiation.
PAgP desirable + desirableForms.
PAgP desirable + autoForms.
PAgP auto + autoDoes not form because neither side starts negotiation.
on + onForms if physical and logical settings match, but no protocol protects you from mistakes.
on + active/passive/desirable/autoUsually fails or behaves dangerously. Do not mix static and negotiated modes.

Member-port rules

The physical interfaces inside an EtherChannel must be compatible. If the switch detects mismatches, it may suspend a member, leave it standalone, or refuse to form the channel. The exact behaviour depends on platform and mode, but the design rule is simple: make the members identical before bundling them.

A common troubleshooting pattern is a bundle that exists, but one member shows suspended, independent or not-in-bundle. Start by comparing the physical interface configuration on both ends, then compare the Port-Channel configuration.

Layer 2 vs Layer 3 EtherChannel

A Layer 2 EtherChannel is a switchport bundle. It can be an access Port-Channel for one VLAN or a trunk Port-Channel carrying multiple VLANs. STP treats the Port-Channel as one logical bridge port.

A Layer 3 EtherChannel is a routed bundle. The Port-Channel is configured with no switchport and an IP address, and routing protocols or static routes use it as a normal routed interface. This is common between distribution and core layers, between routers and switches, or in data-centre designs where Layer 3 boundaries are preferred.

TypeWhere it fitsConfiguration focus
Layer 2 access Port-ChannelSwitch to server, switch to access device, one VLAN.Access VLAN, STP edge settings where appropriate.
Layer 2 trunk Port-ChannelSwitch to switch, switch to hypervisor, switch to firewall carrying VLANs.Native VLAN, allowed VLAN list, STP and trunk mode.
Layer 3 routed Port-ChannelCore/distribution, router interconnects, routed data-centre fabrics.no switchport, IP addressing, routing protocol adjacency.

Basic configuration examples

The exact syntax varies by platform, but this Cisco IOS-style example shows the basic pattern for a Layer 2 trunk using LACP. Configure the physical members with the same settings, assign them to a channel group, then configure the logical Port-Channel.

LACP trunk example:
conf t
interface range GigabitEthernet1/0/1 - 2
 description Uplink members to SW2
 switchport mode trunk
 switchport trunk allowed vlan 10,20,30
 channel-group 1 mode active
 no shutdown
!
interface Port-channel1
 description LACP trunk to SW2
 switchport mode trunk
 switchport trunk allowed vlan 10,20,30
end

For a Layer 3 Port-Channel, remove switching from the Port-Channel and the physical members, then place the IP address on the Port-Channel only.

Routed Port-Channel example:
conf t
interface range TenGigabitEthernet1/0/1 - 2
 description Routed members to Core-2
 no switchport
 channel-group 10 mode active
 no shutdown
!
interface Port-channel10
 description Routed LACP link to Core-2
 no switchport
 ip address 10.10.10.1 255.255.255.252
end
Configuration habit: Build the Port-Channel deliberately on both ends, verify it forms, then add production VLANs or routing adjacencies. For existing links, use a maintenance window: changing channel modes, trunk settings or load-balancing algorithms can briefly interrupt traffic.

Verification and troubleshooting

The first command most engineers reach for on Cisco gear is show etherchannel summary. It tells you which Port-Channels exist, which protocol is in use, and whether member ports are bundled. On many Cisco platforms, a member marked P is bundled in the Port-Channel; flags such as independent, suspended or down point to trouble.

CommandWhat it tells you
show etherchannel summaryFast overview: Port-Channel, protocol, logical state and member status.
show etherchannel portPer-member detail, partner information and operational state.
show etherchannel load-balanceCurrent load-balancing hash method.
show interfaces port-channel 1Interface status, counters, errors and bandwidth on the logical interface.
show spanning-tree interface port-channel 1How STP sees the logical link.
show interfaces trunkWhether the Port-Channel is trunking and which VLANs are allowed.
show lacp neighborLACP partner state on platforms that support the command.

Design patterns and pitfalls

EtherChannel is simple in diagrams and subtle in production. The biggest misconception is that a bundle creates one super-fast pipe for every application. It does not. It creates aggregate capacity across multiple flows. That is perfect for uplinks, trunks and busy server farms, but it will not make a single TCP session exceed one member link.

Planning switch uplinks, campus cores or data-centre trunks?

Browse Singapore system integrators, telecom providers and network specialists who design, configure and troubleshoot enterprise switching environments.

Browse system integrators

Frequently asked questions

What is EtherChannel?

EtherChannel is Cisco terminology for bundling multiple physical Ethernet links into one logical interface called a Port-Channel. It increases aggregate bandwidth, provides redundancy and lets STP treat the bundle as one logical link.

Is EtherChannel the same as LACP?

Not exactly. EtherChannel is Cisco's broader term for the bundle. LACP is the standards-based negotiation protocol used to form link aggregation groups. EtherChannel can also be static or use Cisco's proprietary PAgP protocol.

Does EtherChannel make one connection faster than a single link?

Usually no. EtherChannel load-balances per flow using a hash algorithm. One TCP session normally uses one member link, while many separate flows can spread across the bundle and use the aggregate capacity.

Which EtherChannel mode should I use?

For most new networks, use LACP active mode on at least one side. Active plus active and active plus passive form a bundle. Passive plus passive does not. Static on mode should be reserved for cases where negotiation is not supported and both sides are tightly controlled.

What causes EtherChannel ports to become suspended or standalone?

Common causes include mismatched speed, duplex, trunk mode, native VLAN, allowed VLAN list, MTU, channel mode or Port-Channel configuration. The physical members and the logical Port-Channel must be compatible on both ends.

Sources and further reading