Load balancing over WAN triangle

Discussion in 'Cisco' started by Will, Feb 22, 2005.

  1. Will

    Will Guest

    I am about to have 3 sites (A, B and C) linked by 2Mbs WAN links,
    where site A is head office. Apart from the redundancy I would like to
    load balance traffic back to head office:

    A <--> B
    A <--> C

    Eventually this triangle will grow into a ring with 4-6 nodes, and
    then multiple (5-7) branch sites will stem from each node on the

    I can change the routing protocol from the current RIPv2 to one
    necessary for load balancing as described above. Which routing
    protocol can achieve this? And is this suitable for a WAN with
    approximately 35 routers? I intended to use OSPF, however I saw in
    another thread that someone may have achieve load balancing over a
    triangle using EIGRP.

    Any advice would be greatly appreciated.
    Will, Feb 22, 2005
    1. Advertisements

  2. Will

    Merv Guest

    Please remember that EIGRP uses a compound routing metric . As teeh
    network topology becomes more complex it can become quite a challenge
    to properly manipulate a compund metric to achieve the desired routing

    OSPF and RPv2 use a simple metric.
    Merv, Feb 22, 2005
    1. Advertisements

  3. Think carefully about what you wish for... what traffic do you really
    want to load balance? Don't forget that while you are sending traffic
    using a "load balancing" route (e.g., from A->B->C), the other remote
    also has traffic. Unless your traffic is extremely unbalanced (i.e.,
    only B, then only C have significant traffic) any load balancing will
    have a strong tendency to degrade performance rather than enhance it.

    Once you can define what you REALLY want to achieve (under ALL
    loading conditions), you can look at how best to implement the load

    FWIW: If you do determine that triangular load balancing is
    appropriate, it is easier to do in a D-V protocol than in a link
    state protocol (i.e., you don't want to use OSPF). On the other
    hand, if you have 35 sites in your "backbone" with significant
    redundant link availability, the only protocol to consider is a
    link state protocol.

    Load sharing -- availability -- cost/complexity
    you only get to choose two out of three.

    Good luck and have fun.
    Vincent C Jones, Feb 22, 2005
  4. Will

    anybody43 Guest

    I am writing this hoping that someone shoots me down such that I might
    learn somthing. I have little experience of WANs.

    Load sharing - the last refuge of the desperate - me.

    I know that it is widely used however it does make troubleshooting more
    difficult and can lead to problems that are harder to reproduce than
    would otherwise be the case.

    I would not consider load sharing unless there was no alternative. I
    won't do it just bacause it's there.


    "Eventually this triangle will grow into a ring with 4-6 nodes"

    "this triangle" - OK
    "will grow into a ring with 4-6 nodes" - hmmmm!

    As far as I know the usual design has a simple core into which
    the other sites are connected.

    My first cut would be a core of 2 sites with the sites dual connected,
    once to each of the "core" sites. I would consider a core of 3 sites I
    guess if there was a requirement to lose a site and still have a
    resilient network.

    Depending on the applications and the link speed you could find
    that you lose performance significantly as the number of hops

    The ringy thingy (Oh no! It has reminded me of Full-duplex Token-less
    Ring-less Wire) is not something that I like the idea of much.

    It is a cheap solution though for a fully resilient network. 6 sites 6
    links. Very good value for money.
    anybody43, Feb 22, 2005
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.