In reviewing the various docs on CCO it seems to alway show the 6120 FI dual homing to a pair of upstream core or aggragation switches. I;m trying to understand the advantages of dual homing vs just running strain port channels from Fabric A to its north bound switch and Fabric B to its northbound switch (assuming the northbound switches are not capable of VSS/VPC).
Refer to the Attached diagram if you would.
Option 1 is how I always see it in the CCO diagrams
If these links are not port-channels (just trunks) how does the FI decide which one to forward on for a given VNIC?
I guess the main question is which is the better design and why?
Hopefully my diagrams come out alright.....
Cisco IT uses trunked port channels with full mesh design for Network from the 6120 to the gateways. There are dynamic(default) or static pinning options for determining which blade uses which outbound network uplinks.
I'm farmilar with the static pinning option where I can map a specific vNIC to a pin-group northbound. What I'm trying to figure out is how the dynamic pinning works.
If my L3 default gateway for a given vNIC is on my "left" northbound switch but the dynamic pinning forwards my traffic to my "right" northbound switch I would like to try and avoid that.
Just can seem to find details in the documents.
Sounds like you are connecting the 6120's to a L3 router, in which case you might/should run the 6120s in Switch Mode, not End Host Mode, so they can learn the northbound MAC addresses and do the "pinning" like a normal switch.
As long as both 6120s are connected to the "left-hand L3 gateway" then Switch Mode will work out the pinning like a normal switch.
Option 1 is the best design since this addresses the switch/trunk failure also.
FI-A is not going to forward its NB Network traffic through Right-NB Switch unless there is a failure in left-NB switch on the direct link from FI-A to left-NB-switch because of the network design on its own. Moreover that is what we are trying to achieve in the Option itself...!
Option 1 is ubiquitous but without Switch mode it won't do what is needed because the pinning is arbitrary and so the traffic could go to the wrong switch.
If either VNIC on a blade (ie. both fabrics) expects to send L2 data to a specific device northbound (ie. asymmetric network because not all northbound devices offer the same service) then use Switch Mode :-)
Just a note (caveat?) on the switch mode...
If the northbound switches provide different services on different vlans, you would normally prevent irrelevant vlans on the trunks going to each northbound switch (To prevent unnecessary flooding). This is NOT possible with UCS (yet).
So going for switch mode leaves you with a "all vlans everywhere" approach, with all the considerations that brings (STP, flooding etc.)...