Why mac-pinning preferred over VPC-HM?

[Afilias Canada]

Hi,

For uplink port channels in 1000v, I am seeing mac-pinning chosen in many best practice guides when used in Cisco UCS.  However, I have not been able to find any documentation why mac-pinning is infact better choice than VPC-HM for deploying 1000v within VMware vSphere on UCS.

Can someone enlighten me on this?

Thanks.

Brian

[Robert Burns]

Mac-pinning is a form of VPC-HM.

Since UCS doesn't support host-based port channels (yet), the only other option is mac pinning.  Mac pinning is simple and solid.  I recommend MAC pinning to most people unless they severly requires the aggregate bandwidth of a port channel - assuming the Host can support it.

Regards,

Robert

[cutran]

Hello Brian.  vPC-HM (host mode) is how the Nexus 1000v does the port-channeling.  With this vPC-HM there are 3 options on how the port-channeling is done.  This is configured on the port-profile of type "Ethernet" that is used by the physical NICs on your ESX/ESXi servers.  The command used for this is:

     channel-group auto mode active | passive| on [mac-pinning or cdp]

1. mac-pinning
   • load-balances the available links (i.e. port-channel) based upon mac-address of the VM.  So VM1 will flow through say vmnic0 and VM2 will flow through vmnic1, where vmnic0 and vmnic1 is this vPC-HM port-channel.  If vmnic0 fails, then all VM traffic will start to flow on vmnic1 as normal HA.
2. cdp
   • This is load-balanced based up the Cisco Discovery Protocol (cdp) on the upstream switch.  This is typically not used and is not recommended.
3. lacp {active | passive}
   • Link Aggregation Control Protocol (lacp) provides a better hashing algorithm to load-balance on the available uplinks in that port-channel providing better utlization of the links and allows that VM to use all availalbe links.
   • lacp can be used if all interfaces are connected to the same physical upstream switch
   • if the uplink interfaces are connected to 2 separate physical upstream switches, then a Multi-Chassis EtherChannel technology needs to be available on those upstream switches for lacp to be used.  Example of Cisco switches that has this capabilities are all Nexus 7K, 5K/2K using vPC.  Cisco Catlayst 6K using VSS or Catalyst 3700 using stacking.

With the UCS blades interfaces pinned to a particular FI (i.e. vnic0 to Fabric-A and vnic1 to Fabric-B) and since the FI are not capable to have vPC for the blade servers, that leaves the option of either mac-pinning or cdp.  Since mac-pinning is the recommended between the 2 options, that is why you see all deployments of UCS with N1KV today using mac-pinning.  Hope this helps.

Cuong

[Afilias Canada]

Thank you guys.  I am trying to see why cdp approach is not recommended.  Is there anything that makes it undesirable?   I understand this would require a CDP enabled switch and thus not an option when you don't have cdp enabled upstream switch.  

Thanks again.

[cutran]

As Robert commented, it's quite simple and pretty solid.  CDP does the exact same thing as mac-pinning but again, based upon mac-address vs. upstream switch using cdp.  Since not everyone has CDP, mac-pinning is preferred to make things simpler and has a lot more solid track right now.

[Afilias Canada]

Thanks

Regards,

Brian

[fbcasero62]

hello,

can i configure mac pinning and static pinning together?

In lab we have configured vPC-HM with static pinning to force the fabric used for a service and it works ok, but static pinning does not work in lab with mac pinning..

Regards

Francisco