[c-nsp] ibgp on 6509 with sup2?

Joe Pruett joey at spiretech.com
Wed Feb 11 13:45:44 EST 2015


On 02/11/2015 09:56 AM, Gert Doering wrote:
> Hi,
>
> On Tue, Feb 10, 2015 at 03:42:23PM -0800, Joe Pruett wrote:
>> with bgp filtering might i be able to install just routes of /20 or
>> shorter (hoping that is a small enough number of routes). or would bgp
>> still consume all the routes before it filters and thus run out of ram?
>> i'd don't think i want to experiment with this in production :-).
>
> The easiest way is to filter on *export* from the 7200s, so the 6509
> has no chance to see the refused routes at all :-)

yep, i realized that i should control export after i got some other off
list responses.

>
>
> If I remember right, Sup2 had 256k FIB, going to half that if you enable
> uRPF.  So if you set your iBGP-sessions to "max-prefix 230000" (or
115000),
> and then experiment with feeding it more and more routes, you should
> be fairly safe...

my reading of max-prefix doesn't make it very useful. either you reset
the session after hitting the max, or you just log a warning and then
continue accepting prefixes. i'll just be very stingy with my export to
begin with and see how things go.

>
> [..]
>> and from my reading it sounds like even going to a sup 2t wouldn't
>> really solve the problem? although most of that discussion seems to be
>> external bgp, maybe ibgp wouldn't run into issues?
>
> A sup2t(-xl) can handle 1 million routes, so that's plenty, but most
> serious overkill to get it only for the routing table memory.

and it sounds like my line cards might have to be replaced...

>
>
>> oh, and i run full ipv6 as well, just to make it interesting.
>
> Sup2 and IPv6 is software forwarding, so that might be some reason to
> upgrade eventually...

so far my cpu is not breathing hard, and i run a routed vlan (with
software shaping) for every port. but, overall routed traffic is under
100mb. even my layer 2 traffic rarely makes the little bar graph on the
sup2 card flicker. i'm hoping i have years left on this setup. if the
ibgp pushes things too hard, i'll just live with the inefficient traffic.



More information about the cisco-nsp mailing list