Why does EtherChannel keep both links active when STP normally blocks one?
I built the classic setup in the lab: two switches with two cables between them. Before configuring anything, spanning tree put one of the ports into blocking — makes sense, it's preventing a loop.
Then I bundled the two ports into an EtherChannel (channel-group 1 mode active on both sides) and suddenly both links are forwarding and STP isn't blocking either one. Aggregate throughput doubled.
What I don't get: the physical loop is still there — two cables between the same two switches. So why doesn't STP block one member like it did before? How does bundling make a genuine parallel path suddenly loop-safe? And is LACP doing something special here, or is it just about how STP views the port-channel?
- Quick tip while you test this: bring up the two links one at a time in the lab and watch `show spanning-tree` — you'll see it stay as one Po1 the whole time rather than a port flipping to blocking. — Shira Sivroni,
2 Answers
The key insight: STP runs on the logical port-channel, not the physical members. Once two ports bundle into Port-channel1, spanning tree sees a single interface. There's no parallel path to block, so every member link forwards. You get the aggregate bandwidth AND STP loop-freedom at the same time.
Why it's safe: internally the port-channel never loops a frame back out the member it arrived on, and load balancing is per-flow, not per-packet. A hash (src/dst MAC by default, or IP/L4) pins each conversation to one member, so frames stay in order.
Bonus STP effect: the bundle's path cost drops because cost is derived from bandwidth. Two 1 Gbps links bundled look like ~2 Gbps, so the port-channel is more likely to become the root/designated path than a single link would be.
Standard negotiation is LACP (802.1AX/802.3ad):
interface range g0/1 - 2
channel-group 1 mode active
!
interface port-channel1
switchport mode trunk
Mode combos that actually form: active/active or active/passive. passive/passive never bundles (nobody initiates). All members must match speed/duplex, trunk/access mode, allowed VLANs, and native VLAN or the bundle won't come up. In the lab, run show etherchannel summary and show spanning-tree — you'll see one Po1 in the STP topology instead of two blocked ports.
- The "STP sees one interface" part finally made it click for me. So the redundancy is basically free until the *last* member fails and the whole Po goes down? Thanks! 🙌 — 22maya2n2,
- Worth stressing the per-flow point: a single big transfer (one src/dst pair) rides *one* member, so 2x1G bundle ≠ 2G for one flow. It's aggregate across many conversations. Pick a hash that matches your traffic (L4 if lots of flows share MAC/IP). — ariel.lahav22,
Adding the troubleshooting angle, since a bundle that won't form is the most common lab issue.
If both physical ports stay independent (STP then blocks one, and you're confused why), check for a config mismatch. LACP requires the member ports to be consistent on:
- speed & duplex
- switchport mode (both access or both trunk)
- allowed VLAN list + native VLAN on trunks
- STP settings on the interface
Any mismatch and the port drops out of the channel — sometimes err-disabled. show etherchannel summary shows a flag per port: P = bundled, s/I = suspended/independent, D = down. If you see I, LACP isn't negotiating (often passive/passive, or one side is a plain access port).
Also remember a channel is only as loop-safe as its config: never leave one member outside the channel-group, or you do create a real parallel path and STP will (correctly) block it while the bundle carries traffic — messy. Configure all members into the group first, then bring the interfaces up.