Reducing the Number of Transported Routes

One of my friends sent me this design challenge:

Assume you’re migrating from another WAN transport technology to MPLS. The existing network has 3000 routes but the MPLS carrier is limiting you to 1000 routes. How could you solve this with MPLS?

Personally, I think MPLS is a red herring.

A better question would be “how do you reduce the number of routes transported across your WAN network” or “how do you reduce the routing interaction with your MPLS service providers” (particularly intriguing if you use more than one of them).

As always, there are several options and it’s impossible to recommend the best one:

  • Readdressing is usually out of question (or at least too messy to try). It might also break numerous firewall rules and other hard-coded stuff… unless you automated everything, but then it wouldn’t be hard to readdress, would it?
  • The usual answer would be to summarize the routes. The usual challenge is that you might not be able to do it (because random addressing). Furthermore, summarization is a lossy compression, and loss of forwarding information might result in black holes.
  • RFC 1925 states that there’s nothing that cannot be solved with another layer of abstraction. In this case, we could use any one or more of a half-dozen overlay technologies (IPsec, GRE, VXLAN, DMVPN, LISP…), or use an overlay technology sprinkled with unicorn dust (aka SD-WAN). The beauty of CE-to-CE tunnels is that they totally eliminate the need for PE-CE routing, and (when combined with VRFs) create independent routing domains, so you can use multiple SPs without the associated hassle.
  • Finally, you could go for a really exotic solution like Carriers-Carrier (using additional MPLS labels as the data-plane abstraction mechanism).

3 comments:

  1. Hello mr Ivan Pepelnjiak,

    I would like you could make more clearer two sentences :

    >>> Personally, I think MPLS is a red herring.

    what do you mean ?
    I tried to look into www.wikipedia.org without finding a match.

    >> RFC 1925 states that there’s nothing that cannot be solved with another layer of abstraction. In this case, we could use any one or more of a half-dozen overlay technologies (IPsec, GRE, VXLAN, DMVPN, LISP…), or use an overlay technology sprinkled with unicorn dust (aka SD-WAN). The beauty of CE-to-CE tunnels is that they totally eliminate the need for PE-CE routing, and (when combined with VRFs) create independent routing domains, so you can use multiple SPs without the associated hassle.
    Finally, you could go for a really exotic solution like Carriers-Carrier (using additional MPLS labels as the data-plane abstraction mechanism).

    Not so exotic and not so new if you prepared for CCIE SP in 2005 inventing your own labs

    I tested DMVPN over Carrier Supporting Carrier

    Many thanks for sharing your thoughts on so many subjects.

    Sincerly Yours
    Giuseppe Larosa CCIE Service Provider # 14802, CCNA Voice, CCNP R&S, JNCIA, ITIL SMFv3
    Cisco Support Community Hall of Fame member




    Replies
    1. "Red herring": first hit on Google ==> https://en.wikipedia.org/wiki/Red_herring

      On CsC being exotic or not: just because it's on CCIE SP blueprint and you (and me) configured it doesn't mean it's not exotic ;)
    2. Thanks Sir Ivan,
      how kind of you to answer so quickly!
      I know you can do that too for sure.

      I met in person Luca Martini in Milan and I have seen one of the Kompella brothers' in Alicante Juniper EMEA Professional services meeting in October 2010.

      To be honest MPLS gives me a job since 1999, so for me MPLS and advanced Multicast are keys to distinguish from the others. Of course I can not compare with you and I do not want to. But at least I'm able to interact with you and this is something .

      I have saw a thread on CSC / MPLS about LFA and you have written clearly that is should be a NO NO in a good network design.

      Best Regards
      Sincerly Yours



      Giuseppe Larosa

Add comment
Sidebar