From cwhite's wiki
Jump to: navigation, search


Etherchannel has various names, including but not limited to, LAG, NIC teaming, channeling, bonds and port channels. LAGs goal is to hide the member interfaces from the upper layer protocols. For example, spanning-tree will only see one 2Gbps link and not 2 x 1Gbps links. This results in active/active forwarding rather than active/standby with STP.

  • An individual flow cannot exceed the speed of a single member interface (4 x 1Gbps, single flow cannot exceed 1Gbps).
  • Link Aggregation Control Protocol (LACP) and Port Aggregation Protocol (PAgP) are two protocols you can use to setup LAG. You can also setup static LAG but that is not recommended as it can cause a STP loop upon LAG failure.
  • LAG driver is responsible for load distribution among members. Higher level protocols see LAG as the outgoing interface, not the member ports.

MultiChassis EtherChannel

This is also called MLAG which makes two physical switches appear as a single logical switch which offers better redundancy. Downstream device's thinks upstream boxes are one which allows active/active forwarding. There are a number of different technologies that give us the ability to perform MLAG such as Stackwise, VSS, vPC, MLAG, etc; MLAG technologies are generally proprietary.

The reason these technologies are generally proprietary is because the two devices rely on synchronizing their control plane information. This allows data to flow out one of the member ports in the MLAG.

Cisco MCEC Implementations

  1. StackWise Cross-Stack EtherChannel
    • Access platforms such as the 3750 and the 3850. All members in the stack need to be the same model switch (3750 with 3750).
    • Control plane sync over dedicated stacking cables.
    • Stack cable creates a bi-directional closed loop.
    • One control plane is shared among stack members (eg. OSPF adjacency is formed with the stack and not the individual members).
    • One management plane is shared among stack members.
    • Allows you to have a port-channel with member interfaces on multiple physical switches. This way, if one of the stack members dies, your port-channel does not go down.
  2. Virtual Switching System (VSS)
    • Aggregation platforms such as the Catalyst 4500/6500/6800.
    • Similar to StackWise as it syncs the control plane between the two physical devices. There is also only one management plane like with StackWise.
    • Control plane syncs over the Virtual Switch Link (VSL) which is typically 2 x 10G LAG.
    • If both devices have redundant supervisors, you will usually have 1 active supervisor and 3 standby supervisors.
  3. vPC
    • Used on the Nexus platforms
    • Control plane synchronization over a vPC peer link (typically 2x10G LAG).
    • Two independent control planes in the vPC
    • Two independent management planes in the vPC
    • Usually 2 active supervisors and 2 standby supervisors

StackWise can have more than 2 members but is limited depending on your platform (3750-X can support up to 9 in a stack). vPC and VSS are always a pair of switches. The logic result of all three is the same; turns a physical triangle into a logical P2P link.

VSS Documents:
Campus 3.0 Virtual Switching System Design Guide
Cisco Catalyst 6500 Virtual Switching System Deployment Best Practices

vPC Documents:
vPC Best Practices

LAG protocols

Cisco supports LACP and PAgP. PAgP is Cisco proprietary and is essentially pre-standard LACP. LACP is an open standard per IEEE 802.3ad.

PAgP configuration options
auto - Places an interface into a passive negotiation state. Will respond to PAgP packets but will not attempt to initiate a negotiation.
desirable - Places an interface into a active negotiation state. This interface will start to send PAgP packets in an attempt to find a PAgP port on the remote end.
LACP configuration options
passive - Places an interface into a passive negotiation state. It will respond to LACP packets but does not start LACP packet negotiation.
active - Places an interface into a active negotiation state. This port will actively send LACP packets in an attempt to negotiate with other interfaces.

Load balancing

The available load balancing methods varies per platform. Some examples are:

  • source & destination mac address
  • source & destination IP address
  • source & destination layer 4

The load balancing method is locally significant and does not need to match on each end. Adjustments should be made on each end to match your traffic patterns else you may not load balance across the links evenly (avoid polarization).

L2 vs L3 LAG

The LAG can be access, trunk, tunnel, layer 3, etc. The LAG process is independent of the port mode. LAG suffers from order of operations issues; the members and LAG interface must agree on parameters.

Useful Commands


show port-channel traffic
View traffic load balance between member ports. This will help you determine whether or not you need to change the load balancing method: