MP- BGP EVPN VXLAN - External Routing nodes integration at Spine

[vishaw jasrotia]

Hello All,

I am running a Cisco 9k spine (2 No.) Leaf(20 No.) architecture. Its completely a MP- BGP EVPN VXLAN design. My concern is with integration of External Routing nodes(router and firewall) for WAN. I want this connectivity at spine instead of Border leaf.

Is this architecture is acceptable ?

What are the concern if I go with this design?

Thanks in advance

[Patrick Colbeck]

Yes you can its called Border Gateway on Spine (BGW on spine). Its supported in DCNM as well (latest couple of versions at least).

[melchib]

Border Gateways are used for L2 extension or extending your fabric to another fabric.  Doesn't seem this is what there looking for.

[melchib]

I'd heavily advise against this.  You should keep the spines as simple as possible.  If you try and connect anything besides your leafs to the spines, now you have to create the VRFs on your Spines and additional BGP config, etc, etc.  Yuck!

I'd recommend if you have a core that your border leafs are connected to, connect there or better yet - a wan services L3 switch pair that are connected to the core.

[blandrum]

Is that a concern though with the easy fabric automation provided by DCNM?

[melchib]

If your asking is it a concern to muddy your Spine with other tasks such as terminating firewalls and route peering with WAN edge devices, because you manage them with DCNM...Yes - It's a major concern.  Keep the spines as simple as possible - that's the beauty of a spine/leaf fabric.  Spines dumb, Leafs smart.  With our without DCNM does not change this.

[vishaw jasrotia]

Thanks All for your reply.

Because of some design constrains and some ports limitation, it was not possible for me to connect L3 nodes to leaf. Though this is a ideal design. But We have completed the connectivity through spine using head end replication by statically defining the peer IPs and not through MP-BGP. Through MP-BGP this is not possible.

Thanks

[vishaw jasrotia]

Thanks for your response.
Please see my last reply.