Show 134 – OSPF Design Part 1 – Debunking the Multiple Area Myth

Derick Winkworth, Paul Gear, and Darren O’ Connor chat with Ethan Banks about their experiences with OSPF. The major focus of the conversation is on single vs. multiple areas.

What We Discuss

  • What should the reference bandwidth be set to?
  • How big of an integer does OSPF use to store the cost metric?
  • When should your OSPF design assume one area?
  • Does that old guideline of “50 routers in an area” still hold true?
  • When is it wise to split an OSPF domain into multiple areas?
  • The differences between what you’re taught in a certification program and reality.
  • How does route summarization play into OSPF area design?
  • OSPFv3 & IPv6: how are real-world deployments going?
  • Timer and tweaks and oscillations – oh, my!
  • Why using Bidirectional Forwarding Detection (BFD) is better than tweaking OSPF timers.

Links

ospf-book-cover

Ethan Banks
Ethan Banks, CCIE #20655, has been managing networks for higher ed, government, financials and high tech since 1995. Ethan co-hosts the Packet Pushers Podcast, which has seen over 2M downloads and reaches over 10K listeners. With whatever time is left, Ethan writes for fun & profit, studies for certifications, and enjoys science fiction. @ecbanks
Ethan Banks
Ethan Banks
  • Filip

    re link metrics. is anyone using static link costs? like 10 for local paths, 100 for regional, 1000 international? it makes for easy reading which the route came from.

    • Alexandra Stanovska

      Yes, I’ve seen it being used. For example 10x average RTT between link endpoints and bandwidth inserted into some clever formula to accommodate latency between long distance links. Service provider with inter(national /continental) MPLS connectivity.

      • http://twitter.com/mellowdrifter Darren O’Connor

        I see no issue with that. OSPF simply uses cost. While cost could be based on link speed it doesn’t have to be. You could even cost links according to real cost if you paid per Gb on a link. The value itself is arbitrary

    • Oliver

      Yeah…personally I think it makes more sense to statically set the costs on links rather than using automatic costs based on the the link speed that the router sees. Simpler, easier to troubleshoot, and less things to go wrong.

      • http://twitter.com/mellowdrifter Darren O’Connor

        And if you have 1500 links?

        • Oliver

          Well if your entire network was configured without static costs then that’ll be a pain to change all of them. But, ultimately all of those links needed some manual configuration anyways. My thinking is that “ip ospf cost whatever” should just be part of the standard configuration on different types of interfaces…like the SVI on the primary HSRP router for a access network has 100, the backup one has 110, core router links are 20, etc.

          • http://twitter.com/mellowdrifter Darren O’Connor

            True that. With IS-IS you need to do this manually anyway. A bit of extra work, but it’s not like you’re upgrading all links at the same time…

  • http://twitter.com/Iamjeffvader Keith Humphreys

    Good shows guys, looking forward to more of the same on old school technologies that have a big place in my heart and in bigger place in large SP networks :)

    I think it was Derick who referenced the inter area limitations of type 10 LSAs a couple of times. I have designed a few networks with stub areas, more for administrative reasons rather than technical limitations. I am wondering what the panels thoughts on LDPoRSVP tunnels and if they have used them? Say you have 2 non-backbone areas (1 and 2) and you want to use FRR end to end you create RSVP tunnels as illustrated below. You run targeted LDP across this and you get the same effect and you should see SDH like failover timers bypassing the area scope limitation of the type 10’s.

    r1–area1–abr1–area0–abr2–area2–r2

    I have only done this in a lab and never had the requirement to run it in production. What are peoples thoughts on this kind of deployment? Maybe something that can be discussed in future OSPF design podcasts?

    On the lowering of timers mentioned (LDP and BGP having a dependency on OSPF) setting a hold time on an interface will mitigate against flapping links on protocol stability. I do agree lowering standardized timers isn’t a great idea unless you have a really good reason but in a service provider environment, particularly when carrying voice, waiting over half a second for failure detection is generally unacceptable to customers. Very fast BFD timers mixed with OSPF/LDP synchronization and interface-up hold timers can provide this sort of set up where FRR based RSVP-TE isn’t desirable due to its complexity.

    In response to Filip I do something similar to standardize a topology type, makes it easier for a NOC environment to understand a network layout they may not be overly used to. If your network is well designed hierarchically then it does have benefits. Of course there will be an errant link that gets planned in to the core that you need to fudge the metric to get traffic over it without impacting forwarding elsewhere. TE is best used there but if you are an LDP only domain it gets increasingly difficult to predict traffic patterns as the network size increases without something like Cariden Mate or Opnet capacity planning software.

  • Stephen skinner

    I still beleive in having multple areas in an ospf network…..
    I am currently working on a network which spans multple geographic areas which has grown organically over the past 10 years , as such i have over 200 routers in area 0 ranging from 2600 series routers all the way upto sup720,6509.
    Without an area design in place , we now have a core router which has routes that refresh every 30 minutes . This is very hard to troubleshoot and is taking me an age to organise. With areas i could easily isolate the troublsome router and stop all the route-flapping