Episode 9 – What Your Mamma Never Told You About OSPF

As a continuation of our deep dives into routing protocols, this episode of Network Collective takes a closer look at OSPF. What does graph theory, CAP theorem, and distributed databases have to do with routing? What exactly is that type 4 LSA for? What is OSPF really good for? These and other questions are discussed by our guests Russ White and Nick Russo.


Russ White
Guest
Nicholas Russo
Guest
Jordan Martin
Co-Host
Eyvonne Sharp
Co-Host

Outro Music:
Danger Storm Kevin MacLeod (incompetech.com)
Licensed under Creative Commons: By Attribution 3.0 License
http://creativecommons.org/licenses/by/3.0/

2 Comments

  1. August 9, 2017
    Reply

    Nick Russo here. I never had a chance to talk about my “high availability” design for my OSPF hub/spoke network. I will explain it briefly here.

    Scaling OSPF over hub/spoke is already hard enough, but what if you need some kind of backup hub? Adding 2 hubs seems to be the obvious answer, but I’ll propose something else, since scaling OSPF in this situation gets sloppy.

    Let’s assume we are using DMVPN or something like it. If the tunnel source (NBMA) at the hub is an anycast address and is advertised to the Internet or private WAN provider from multiple points, then outstations can connect to the closest one. You could also do a prioritycast (active/standby) design with longer matches (say a /23 and /24 over the Internet, or if over a private WAN, both /32s but with AS prepending out) to prefer one site over the other. Either way, I hope you see where I am going with this. The failure of one hub (or the PE-CE link to it) would cause the spoke tunnel to simple rebuild itself towards the other hub.

    For the record, YES I use this in production, and YES it works great. I also coupled it with IPsec. The failure time DOES take a little longer since the failover is stateless, so the IKE/IPsec process, DMVPN process, OSPF process all need to “rerun” before the network convergence. Assuming a few seconds of downtime is a tolerable trade-off for a high scaling OSPF hub/spoke network, this design is a strong option.

    Note that LinkedIn also uses stateless anycast in a similar way with TCP by injecting a /24 onto the Internet in multiple regions of the world. Expensive state-retaining middle boxes be gone!

Leave a Reply

Your email address will not be published. Required fields are marked *