Routing flaps, was Re: Ping flooding
avg at ncube.com
Fri Jul 12 05:25:56 UTC 1996
>I doubt that a large provider such as Sprint or MCI or Agis or any other
>major internet backbone is going to be able to hide much of their
>internal route flapping, as their internal routes ARE their external routes.
Well, not. BGP (exterior) table is cleanly separated from IGP
table in modern backbones, so bouncing interior circuits do not
generally produce routing flap visible outside (if it is not severe
enough to cause partitioning). The links to exchange points
and customer-access routers is another kettle of fish altogether...
There are more than few ways to induce interesting instabilities in
the iBGP-hack based backbones, some going as far as to produce
self-feeding route oscillations. However, fascistic filtering on
both IGP and BGP routing information (so they won't interact in
unexpected ways) helps.
>My original note was intended to reply to the specific case of using
>dynamic routing to distribute routes which could just as easily be done
>with static routes - I.E. where there is one and only one path to the
>destination. In this case, it is sometimes desirable and necessary to
>run some sort of interior routing protocol, but it is not desirable that
>flaps caused by a break along the one and only one path be propogated to
>the outside world, but instead the packets should be null0'd at the point
>of path convergence.
Yes, you can eliminate global impact of interior flaps this way;
but believe me, to design a truely reliable and booletproof interior
routing takes no less efforts than just place all the static routes.
If you don't think so you probably never had people starting gated
on their workstations just for the heck of it :)
More information about the NANOG