[c-nsp] iBGP over IGP with ECMP

[Andrew Clark]

On Tue, Nov 29, 2011 at 12:40 PM, Peter Rathlev <peter at rathlev.dk> wrote:

> On Tue, 2011-11-29 at 11:16 -0600, Andrew Clark wrote:
> > I'm looking at using iBGP, peered to loopbacks, with EIGRP as my
> > "bootstrap" IGP.  I'm assuming BGP will just follow the IGP paths,
> > which will be ECMP and the load-balancing will be transparent to iBGP.
> > Is this a recipe for disaster?
>
> It'll work fine. Beware though that the hardware based L3 switches use
> per-flow load-sharing, not per-packet. That's probably what you want
> anyway to avoid out-of-order, but maybe also different from what a
> software router (or Dynamips) would do.
>
> > The total size of the routing table isn't too big, around 3K routes.
>
> You might want to play with "sdm prefer ..." on those switches then. The
> default ("desktop access IPv4") only allows for 2k routes. Of exceeded
> the least specific prefixes will be software switched. That would
> probably be a default route, and that one often carries most of the
> traffic in a setup with few routes.
>
> Use "sdm prefer routing" to have up to 8k routes.
>
> --
> Peter
>
>
Thanks!  I had already set the sdm template for routing.  I'll see how it
all plays out once it is turned up.

Andrew