quick bit of context so my question makes sense:

I've been asked to plan out some subnets and routing for a new IP range we've just been allocated. The range will be used to customer's equipment (on IPs allocated out of this range) to peering providers. </context>

Every routing example I've seen appears to revolve around one central "Core" router, which connects to all the other networks and routers. This would leave me with a setup looking something like this

Network diagram number 1

Full diagram at http://www.gliffy.com/go/publish/6360724

To me, this would seem to have a few disadvantages, namely a single point of failure, and wasting 2 IPs for each additional connection. That lead me to come up with this:

Network diagram number 2

Full diagram at http://www.gliffy.com/go/publish/6360825

In theory, this makes better use of the IP ranges, and has fewer single points of failure. Can anybody suggest why the first set-up seems preferable, even with it's seemingly obvious faults?



2 Answers 2


With your "one big /24" solution, each client router will need to make a decision of which upstream peering router to send traffic to, so they would either need to carry full internet routes or use some other (e.g. policy-based) method to make a decision.

With your "one big core router" solution, each client router simply needs to have a default route pointing to the one big core router, which in turn will make the upstream routing decision.

Of course, I wouldn't recommend having just "one big core router", but would prefer to see "two big core routers" instead. Each client router can then be dual-homed to those two big core routers, providing the necessary redundancy (both for connectivity and routing) when there is a core router failure.

  • Again, I was under the impression that BGP would handle all the routing. Will it not?
    – Foxocube
    Oct 24, 2014 at 9:29
  • BGP will handle things, but the point is that if you have one or two core routers then you won't need to run BGP on your client routers -- they just need a static default route pointing to your core layer. Do you really want to run BGP on all your client routers? Given that you have multiple upstream egress points, you would likely run with full internet routes on each client router, which would have memory and CPU implications given that there are currently >500,000 prefixes in a full internet table. Much better to have a core layer that's sized to handle all that.
    – Rob
    Oct 25, 2014 at 1:23
  • Given the size that routing table would be, the first layout suddenly makes a lot more sense!
    – Foxocube
    Oct 25, 2014 at 22:23

it is control. You client router don't know how to get to what. That is what your core router does. It is the middleman that connects your clients to their upstream.

Also their would be 9 interfaces if the B site is included.

  • I was under the impression that BGP would handle the routes between them? And yes, I miscounted the number of interfaces :(
    – Foxocube
    Oct 23, 2014 at 8:11
  • Also, what would the advantages be of this over a single large router doing everything?
    – Foxocube
    Oct 23, 2014 at 20:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.