My BGP in EVPN-Based Data Center Fabrics blog post generated numerous comments from engineers disagreeing with my views on using IBGP-over-EBGP.
As usual, there were three kinds of comments:
- People like Alexander Grigorenko arguing with the technical merits of what I wrote (Alex published a blog post describing his perspective). They were right – I should have spent more time on the details;
- People like Paul saying “while I understand your discussion, nothing else would work for my fabric”. Fair point – if you have to carry 1M prefixes around you cannot use generic design guidelines found on the Internet;
- Anonymous commenters being annoyed because I called their baby ugly (more about this in another blog post).
Anyway, Alex’ comments triggered a major rewrite of the BGP in EVPN-Based Data Center Fabrics section of Using BGP in Data Center Leaf-and-Spine Fabrics article. It took me a long time to figure out how to structure it, and then more than a week ironing out the kinks based on feedback from people way closer to actual BGP source code than I ever was.
The new version of the document is online. There are still a few missing bits and pieces, like route target implications of using numerous AS numbers in a single EVPN fabric, or performance differences between IBGP and EBGP-based EVPN designs.