What routing protocol should you use in the underlay of a #VXLAN network? Is #OSPF or #BGP better? That's the wrong question to ask. A network should always be designed based on its requirements. In my latest post I go through the characteristics of OSPF and BGP and compare them. There are some misconceptions around OSPF such as scalability while this is not a main concern in most datacenter networks. In a topology with a spine with 32 ports and leaf with 48 ports, 1536 ports can be provided. How many networks require more than 1536 ports? OSPF can easily scale to this and beyond. I have seen numbers mentioned like 100-200 switches. In the post I go through characteristics of each protocol such as: - Establishing adjacencies and sending protocol packets - Numbering - Scalability - Convergence - Policy - Support - Ability to template The goal of this post is to be thought provoking. To get you to consider the characteristics of each protocol and your requirements instead of designing based on for example what hyper scalers do. I hope you enjoy reading it! Link in the comments below 👇

Speaking solely from my experience as an operator, running BGP as both the underlay and overlay complicates troubleshooting scenarios. Those pesky 2am outages where it’s “All Hands On Deck”, I have seen many misinterpret and get flustered when trying to decipher the output of “show xyz” routing commands. I love BGP more than most, but there is significant value in separation of church and state (underlay and overlay). My preference is to run a flat Level2 IS-IS underlay and have BGP handle the overlay signaling. Great write up Daniel, would love to see some heavy DC-focused blogs in the future!

Very good article Daniel! as a side note, at some stage i would like to attach some precise numbers on OSPF vs BGP in Clos underlay, especially with regards to convergence and scale. Basically running a comparison with something like Sybil( https://compunet.ing.uniroma3.it/assets/publications/Caiazzi-Scazzariello-Sibyl.pdf ) or similar. Good approach with the comprehensive comparisons including aspects not directly related to routing (e.g. templates)

Great question! In almost all our EVPN VXLAN projects we use ISIS as underlay routing protocol and so far without a single issue. The first implementation we did, was in 2016 and if I remember correctly, I read somewhere in 2016, that it was recommended to use a different routing protocol for underlay, that you would use for external routing, i.e. peering outside DC (probably OSPF), for potential routing process fault isolation.

Distinguished Engineer at Greyson Technologies CCIE #5820 (R/S ; Collaboration) Cisco Community VIP 2022-2025 Cisco Learning Network VIP 2023-2025 Cisco Insider Champion 2024-2025 Cisco CCIE Advisory Council 2025-2026

VXLAN seems to be mostly used in DC networking which isn't an area of strength for me. I do recall that ISIS is frequently mentioned in SASE deployments, but you didn't mention it above. Any reason for that? I would also say that a VXLAN deployment implies common administrative control to me which would seem to make that an IGP desirable as opposed an EGP for the underlay network.