I've got some use cases where I'd like to use something like OTV, but on Nexus 9Ks. Say two datacenters where each datacenter / site is 2 spine and 4 to 6 leaf switches, and only a few VLANs. I see little point to running VXLAN/EVPN within such small sites, just VPC and VLANs with SVI's on the core / spine switches.
VXLAN/EVPN Multi-Site with pseudo-Border GW looks like just the approach I want. However, the use case focuses on migration, where's I'm more interested in a non-VXLAN new N9K based site operating "in legacy mode".
Question #1: The documentation I've seen is a bit unclear, it says something about SVI's and trunks are not supported for external entities. I'd maybe add the word "**routing**" to that -- external entities that would want dynamic routing? I'd hope trunks and SVI's would be supported for internal clients, i.e. servers and other devices that are connected within a VLAN. Seeing as that's how the "legacy" network would be operating, and the description about putting the anycast GW on the border GW. What exactly IS the caveat here?
Question #2: With OTV, one only transports selected VLANs between datacenters. With the Multi-Site writeup, there is no discussion of VLANs that are *not* associated with VNI's, etc., mixed into the same VRF as the VNI-backed VLANs. Questions come to mind, like does routing between the two types of VLAN work, etc.? I built a small VIRL model (not multi-site per se, since the available Nexus image is not quite new enough) and routing between non-VNI and VNI-backed VLANs works fine, locally or over the overlay. I did have to put network statements into the VRF section of BGP for the non-VNI-backed VLANs. What is / is not supported in basic VXLAN/EVPN or Multi-Site along these lines?
I should note that associating non-extended VLANs with unique VNI's appears to be one way of dodging this whole discussion. Might that be a Best Practice? (Makes the configurations more uniform...).
