[c-nsp] Dreaded FIB Exception on Sup2

[Adrian Chadd]

On Sun, Sep 14, 2008, matthew zeier wrote:
> I would be interested in the results of such an experiment (I was about 
> to research this this week myself).
> 
> Church, Charles wrote:
> >I got curious last week when I saw this thread.  From my (AS 26296)
> >point of view, there aren't a whole lot of routes in the /25 to /29
> >range, maybe a couple hundred total when I looked at it last week.   The
> >/24s were huge, about 142,000.  I'm curious how many of those /24s are
> >covered by larger aggregates.  Might be a fun experiment figuring out a
> >safe way to filter those.  Probably involve AS path length, maybe RIR
> >allocations.  

If your trust your upstream to get your outbound packets to their eventual
destination, then you don't -need- a full table; it just makes outbound TE
easier.

If you have a lot of peering and unreliable transit (!) then sure, full table
is a win.

You -can- do differential outbound TE without needing a full BGP table.
You just need to find a way to get traffic metrics for flows -to- destinations
(netflow? on a sup2? surely you jest Adrian..) and then you can craft temporary
injected routes into your Sup2 to do traffic balancing without anywhere near
a full table.

(Of course, this all falls apart the moment your upstreams start doing dumb
stuff like feeding gigabits of traffic through a hacky Sup2 + route injection
+ filtering, but hey, you're paying -transit- so you don't necessarily need
to do all of those wonderful BGP tricks with massively high end kit being
pushed to its performance/bug limit, right? :)



Adrian