Cisco Nexus vPC Routing

[vsurresh]

Hello. I would be grateful for any input/help.

Please see the attached diagram. Bottom two Nexus are in Site 1 and top 2 nexus are in site 2. (Please ignore the L3 links between N1-N2 and N3-N4)

10.17.13.10 is dual connected to both Nexus. We run HSRP on both Nexus and N1 is HSRP active from the control plane point of view. From the data plane point of view, HSRP is Active/Active. (peer-gateway is configured)

Similar set up in Site 1. 10.13.13.10 is dual connected to both Nexus. We run HSRP on both Nexus and N3 is HSRP active from the control plane point of view. From the data plane point of view, HSRP is Active/Active. (peer-gateway is configured)

We run OSPF between N1-N3 and N2-N4. The traffic is load-balanced on both links via ECMP. Is this a valid design? From my understanding, this can result in asymmetric routing.

For example, traffic from 10.13.13.10 can use any of the uplinks as per EtherChannel hashing. If the traffic lands on N3 the path would be 10.13.13.10 >> N3 >> N1 >> 10.17.13.10. The return traffic may come via the other link 10.17.13.10 > N2 > N4 > 10.13.13.10

Please let me know if should you require more information.

Thanks

[balaji.bandi]

It all depends on you configure your OSPF with Traffic engineering. 

here is the design guide and best practice  for reference :

https://www.cisco.com/c/dam/en/us/td/docs/switches/datacenter/sw/design/vpc_design/vpc_best_practices_design_guide.pdf

[paul driver]

Hello

Can you confirm that the L3 interconnects between switches in the same site are separate l3 links and not your peer-keealive link?

With your design the physcal switches at L3 should see its connected OSPF link as a single ospf peering, vPC should not come into it here, so each switch would have a two ospf peerings. ( eg: N1-N2, N1-N3 etc) correct?

[vsurresh]

Hi, Paul. Thanks for your response.

Answering your first question, yes, L3 interconnects between switches in the same site are separate L3 links.
Regarding the second question, yes, there would be two OSPF peerings on each switch. From N1 point of view, the active route to reach 10.13.13.10 would be via N3. If the link between N1-N3 goes down, N1 will use the path via N2. (OSPF will re-calculate the path and add N1-N2 path to the routing table)

Where I get a bit confused is, for example, traffic from 10.13.13.10 can take a path of N3 > N1 > 10.17.13.10. The return path may come via 10.17.13.10 > N2 > N4 rather than 10.17.13.10 > N1 > N3 because of the vPC port-channel hashing. Is this okay?

Regards
Suresh