Two cables, one logical link

What you'll build

• SW-Core — a switch, the core switch, and the root bridge
• SW-Access — a switch, the floor's access switch
• SRV-Files — a server, the file server hanging off the core
• PC-Floor — a pc, a workstation on the floor

Step by step

1. 1. Two switches, one uplink, and a root you chose

   Drag two switches onto the canvas, name them, and cable SW-Core's Gi0/1 to SW-Access's Gi0/1. Then make SW-Core the root bridge with `spanning-tree vlan 1 root primary`. That line decides in advance which end of any redundant link will block later — always the end farther from the root.

   • Cable SW-Core Gi0/1 ↔ SW-Access Gi0/1

   On SW-Core — Name the core switch and make it the root bridge

   enable
   configure terminal
   hostname SW-Core
   spanning-tree vlan 1 root primary
   end

   On SW-Access — Name the access switch

   enable
   configure terminal
   hostname SW-Access
   end

   Check: run show spanning-tree on SW-Access and look for Port Gi0/1.

   Why: With default priorities, the root bridge is whichever switch has the lowest MAC address — an accident of manufacturing. Pinning it on the core makes the rest of the spanning tree predictable, and the core is where the traffic converges anyway.

2. 2. A server on the core, a PC on the floor

   Plug a server into SW-Core's Fa0/1 and a PC into SW-Access's Fa0/1, and put both in 192.168.1.0/24. Every byte between them has to cross the uplink, which is what makes the uplink the link worth doubling.

   • Cable SRV-Files Eth0 ↔ SW-Core Fa0/1
   • Cable PC-Floor Eth0 ↔ SW-Access Fa0/1

   On SRV-Files — Name the file server and address it

   hostname SRV-Files
   ipconfig Eth0 192.168.1.100 255.255.255.0

   On PC-Floor — Name the floor PC, address it, and reach the server

   hostname PC-Floor
   ipconfig Eth0 192.168.1.20 255.255.255.0
   ping 192.168.1.100

   Check: run show interfaces status on SW-Access and look for Gi0/1 connected 1 auto 1000 gigabit.

   Why: Hosts on many access ports all funnel into one uplink, so the uplink saturates first. That is why the link between switches is the one engineers want to double — and why the next step is so disappointing.

3. 3. Add a second uplink — and watch spanning tree block it

   Cable SW-Core's Gi0/2 to SW-Access's Gi0/2, hoping for twice the bandwidth. Spanning tree sees two paths between the same two switches — a loop — and blocks one. Run `show spanning-tree` on SW-Access: Gi0/1 is still the root port, and the new Gi0/2 reads Altn BLK. All traffic still rides one cable.

   • Cable SW-Core Gi0/2 ↔ SW-Access Gi0/2

   Check: run show spanning-tree on SW-Access and look for Gi0/2 Altn BLK 4 128.Gi0/2 P2p.

   Why: Spanning tree cannot tell a deliberate second cable from an accidental loop. Two parallel links are a loop, so one of them is held in reserve: it takes over if Gi0/1 fails, but carries nothing until then.

4. 4. Bundle one end — and see that half a bundle is nothing

   On SW-Core, put both uplinks into channel-group 1 with LACP in active mode. Active means SW-Core now asks the far end, over both cables, to form a bundle. SW-Access is not configured, so nobody answers: `show etherchannel summary` shows both ports stand-alone (I), Po1 down (SD), and SW-Access's Gi0/2 is still blocked.

   On SW-Core — Put both uplinks into channel-group 1, LACP active

   enable
   configure terminal
   interface range Gi0/1 - 2
   channel-group 1 mode active
   end

   Check: run show etherchannel summary on SW-Core and look for 1 Po1(SD) LACP Gi0/1(I) Gi0/2(I).

   Why: LACP is a negotiation: a channel exists only when both ends agree on it. Until the far end answers, each member behaves as an ordinary port — the safe way to fail, because a half-configured bundle can never create a loop of its own.

5. 5. Finish the bundle on the other end

   Give SW-Access's Gi0/1 and Gi0/2 the same channel-group 1, LACP active. Now both ends agree: Po1 comes up (SU), both members read P, and spanning tree stops seeing two cables at all — `show spanning-tree` lists a single logical port, Po1, forwarding. The loop is gone because, as far as spanning tree knows, there is only one link.

   On SW-Access — Put the matching uplinks into channel-group 1, LACP active

   enable
   configure terminal
   interface range Gi0/1 - 2
   channel-group 1 mode active
   end

   On PC-Floor — Confirm the server is still reachable over the bundle

   ping 192.168.1.100

   Check: run show etherchannel summary on SW-Access and look for 1 Po1(SU) LACP Gi0/1(P) Gi0/2(P).

   Why: EtherChannel turns several physical links into one logical link. Spanning tree cannot block half of a single link, so every member carries traffic, and losing one member is not even a topology change — the channel just runs on the rest.

The theory behind it

• EtherChannel — link aggregation
• Spanning-Tree Protocol (STP & RSTP)

More in Switching & wireless

• Map the network with CDP and LLDP — Let two switches and a router discover each other, find the router's address from a switch that was never told it, add LLDP, then stop the router announcing itself toward the internet.
• Two departments, one switch — Split a single switch into Sales and Ops with VLANs, and watch the pings between them stop.
• One VLAN across two switches — Split four PCs into two VLANs across two switches, then carry both VLANs between the switches over a single tagged trunk.
• Router on a stick — Split one switch into two VLANs and route between them over a single router port.
• Route between VLANs on the switch — Give one switch an SVI in each VLAN, find out why the VLANs still can't talk, then turn on ip routing — inter-VLAN routing with no router at all.
• Choose your root bridge — Close a loop of three switches, find the port spanning tree blocks on its own, then decide which switch is root — and which one takes over when it fails.