Re: Multihome BGP / Confed issue (update)

From: Darren Pedley (darren@dnsl.com)
Date: Sun Feb 12 2006 - 14:12:51 GMT-3


I have found that the problem is related to confed's or route-reflectors, as if I fully mesh my ibgp (without confed or r/r) then it works (and the bgp tables look correct). So just to recap here is my current setup.

    |---- r1 ----- r2
    | |
bb1 | |
    | |
    |------------- r3

bb1 is AS254
r1,r2,r3 are AS3

route 1.1.1.0/24 is advertised from bb1

if r1,r2,r3 are fully messh, then bgp tables look fine on all routers (i.e. two paths to 1.1.1.0/24)
if r2 is a route-reflector, then r2 will only send the route to one of its peers (it appears to be random as to which one, although I suspect it has something to do with when the peering process completes)

I get similar results from a confed, but won't go into the config.

Config on routers is pretty simple, i.e.

bb1
neighbor 20.20.20.21 remote-as 3 // r1
neighbor 30.30.30.31 remote-as 3 // r3
network 1.1.1.0 mask 255.255.255.0
int loop1
  ip address 1.1.1.1 255.255.255.0

r1
neighbor 20.20.20.20 remote-as 254 // bb1
neighbor 2.2.2.2 remote-as 3
neighbor 2.2.2.2 next-hop-self
neighbor 2.2.2.2 update-source loop0

r2
neighbor 1.1.1.1 remote-as 3
neighbor 1.1.1.1 route-reflector-client
neighbor 1.1.1.1 next-hop-self
neighbor 3.3.3.3 remote-as 3
neighbor 3.3.3.3 route-reflector-client
neighbor 3.3.3.3 next-hop-self

r3
neighbor 30.30.30.30 remote-as 254 // bb1
neighbor 2.2.2.2 remote-as 3
neighbor 2.2.2.2 next-hop-self
neighbor 2.2.2.2 update-source loop0

"no sync" and "no auto-sum" are also in the configs

Can anybody explain why this is? i.e. why does it look right when fully messhed? If you look at the link below it kinda suggested this was an ibgp thing, but looks to me like ibgp works fine, its only when rr or confed's are in use.....

Thanks again,

Darren

On Fri, Feb 10, 2006 at 03:44:22PM +0800, Darren Pedley wrote:
> Hi,
>
> I've just found this:
>
> http://www.groupstudy.com/archives/cisco/200202/msg00516.html
>
> Which I missed before (sorry), appears to explain my problem. Anybody else clarify that this is "standard behavior".
>
> Thanks
>
> Darren



This archive was generated by hypermail 2.1.4 : Wed Mar 01 2006 - 11:28:17 GMT-3