[c-nsp] Setting "weight 255" as default for customer BGP with uRPF strict

[Brian Feeny]

Pete,

weight is the sledge hammer that will cause the FIB to have an 
adjacency with the customers interface, iregardless that this router 
sees a better localpref to another iBGP peer.

Customer Router C connects to your Router A and  your Router B.
Customer sets localpref to 100 on network A to router A, and 70 on 
network B to router A
Customer sets localpref to 100 on network B to router B, and 70 on 
network A to router B

Router A and  Router B are iBGP peers.

Router A has uRPF strict turned on.  It will have an entry in its FIB 
for network A to Router C, and network B to Router B.
If it receives packets from network B, from Router C (very possible), 
it will drop them, because uRPF thinks the reverse path is Router B, 
and so traffic for network B should only come from router B.

weight fixes this.

brian



On Nov 21, 2004, at 1:31 PM, Pete Templin wrote:

> Brian Feeny wrote:
>
>> Is anyone setting "weight 255" for all customer neighbor sessions?  I 
>> was looking at doing this for multi-homed
>> customers, so that asymmetrical routing, split prefix announcements, 
>> etc don't break uRPF (strict).  Then I just thought,
>> well, why make a special case for multi-homed customers, what harm is 
>> there on setting weight to all edge neighbor
>> sessions.  I mean its only going to effect packets on that local 
>> router, and if the packet has gotten so far as to actually
>> end up on the router directly connected to the peer, it just seems 
>> the best thing to do to hand it to the peer.
>
> Why "255" as the value?  Are you simply looking for a non-zero value? 
> Would a value larger than 32768 (default for locally originated 
> routes) make any difference in any scenarios (hint: it did for me 
> yesterday)?
>
> I don't think weight will actually override any more specific 
> announcements on the router, and therefore might not have the effect I 
> think you're after.
>
> pt