First of all, BGP has no "split horizon" rule. IBGP neighbors will not
propagate routes to each other as a matter of loop protection (since we have
no way to add AS numbers w/in the AS--don't worry about confederations yet).
This really has nothing to do w/split horizon. Try this:
R1----R2---R3
R1 will forward it's routes to R2. R2 will forward it's routes to R3. R3
will forward it's routes to R2. Problem: R3 has no routes for R1. Why?
R1 and R3 need to have a neighbor connection between them. Once this is
done, R3 will have routes from R1 and R1 will have routes from R3. The
reason for this can be seen if you really think about it:
R1---R2---R3
\ /
\ /
R4
R1 forwards routes to R2 and R4. R2 forwards routes to R3. R4 forwards
routes to R3. R3 now has a loop. The only loop protection BGP has is AS
numbers. We are in the same AS here so that won't help. If neither R2 or
R4 forwards routes from R1, we have no loop. This is fundamental to
IBGP---all routers must be fully meshed to see all routes.
On to route reflectors. Route reflectors relax the rule of full mesh
connectivity. Basically, all routers internal to an AS peer with the route
reflector only. The route reflector "reflects" routes from each peer to all
the other peers and to all non-clients ONLY if the routes came from a client
or a different non-client. This makes sure that all non-clients have the
routes from the route reflector peers and other peers within the AS.
We do not need any form of "split horizon" here because the route reflector
serves as the means for a full mesh. The route reflector knows (you can
call it magic, or you can do a lookup on CCO for cluster IDs) not to forward
routes back in the direction they came. Do NOT confuse this w/ split
horizon. Split horizon really has no place here.
Good sources for further study:
Internet Routing Architectures by Halabi--this is considered the bible for
BGP. You ever want to be a CCIE, you need this book.
Routing TCP/IP Vol. II by Doyle--somewhat new, but still a beauty.
HTH
Scott
""Joep Hoet"" wrote in message
[EMAIL PROTECTED]">news:[EMAIL PROTECTED]...
> The BGP split horizon rule is confusing me.
> It states that routes learned by a IBGP router are not propagated to other
> IBGP routers.
> I guess this only regards IGP learned routes and trying to avoid that an
> IBGP router learns a IGP route before other parts of the AS which are
needed
> for this route do. Is this right?
> .....
>
> Secondly, when a route reflector is configured and it receives an update
> from a client, this route is propagated to all clients (except the
> originating client), and, which confuses me, to all nonclients.
>
> This is like an IBGP router (in this case a cluster) propagating routes to
> other IBGP peers, isn't ?
>
> How come there is no need for a split horizon rule here, or rather, since
I
> don't seem to understand it very well, why is the effect that causes us to
> have a split horizon rule in the first place not there?
Message Posted at:
http://www.groupstudy.com/form/read.php?f=7&i=35680&t=35679
--------------------------------------------------
FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html
Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
