ACI single homed topology support

[David Yang]

My customer has an ACI implementation which there are 2 of the all leaf switches single homed with one of the two spines because of the circuits constraint, say leaf-1 only connects with spine-1, and leaf-2 only connects with spine-2.

leaf-1 and leaf-2 are configured as vPC peers for L2 connection to the legacy network via static binding. There appeared significant packet loss between ACI and the legacy network when vPC is up.

We understand this setup did not follow ACI design best practices, but the questions was whether it is supported ACI model? Any caveat or issue with this vPC setup on a pair of single homed leaf from ACI architecture perspective?

Thanks much!

David

[RedNectar]

I'm trying to get my head around your question. Is this what you are describing?  If so, I don't see any reason why it should not pass traffic, but of course this would not be a supported design. Any link failure (orphans) South of the leaf would put a load on the pee-link function that would now have to be routed via a 4-hop route (via leaf3 or 4). 

[David Yang]

Thank you Chris for your message and the diagram :). I really appreciate your thoughts. These are border leafs and they are running double side vPC with a pair of Nx7k's on the legacy network. They did forward traffic but there was connectivity issue between ACI and the legacy network when vPC was up.

Thanks,

[6askorobogatov]

Correct me if I'm wrong, but in this topology Leaf 1 and 2 are not connected and there is no spine to provide a peer-link for VPC 

[RedNectar]

Look again. Leaf1 and Leaf2 are connected and have two possible paths:

1. Leaf1> Spine1 > Leaf 3 > Spine2 > Leaf2
2. Leaf1> Spine1 > Leaf 4 > Spine2 > Leaf2

Although not ideal, ISIS will handle it.

[6askorobogatov]

Thanks, Chris.