BGP Configuration Using Route Reflector

In iBGP, there is a rule to avoid routing loop by not advertise any route receive from other iBGP neighbor.  The rule have consequence that every router in the AS needs peer with all router so every router have routing from other router in the AS.  This is called full mesh configuration. 

Imagine if an AS have 100 routers.  It needs to create 99 peers in every router, and there will be 99x100 peers in the network.  There are two mechanism as alternative of full mesh iBGP, BGP Confederations, and BGP Route Reflectors.  This post will describe route reflector.

How It Works
BGP route reflector is allowed to break iBGP loop avoidance rule, by advertising any route that received from its clients to other clients, other non client peers.  Clients is a router that connect to a BGP route reflector.  This mechanism eliminate needs for full mesh iBGP.
There are two types of router in an AS with route reflector, namely clients router, and non client router.  Client router peers only with route reflector, on no need for full mesh iBGP.  Non client peers with other non client routers, and need to be fully meshed.

Operation
1.  Routes receive from clients is reflected to other clients, and non peer client (or other BGP route reflectors)
2.  Routes receive from another non client peers (or route reflectors) is reflected to clients only
All the routes relected is the best path routes only.

Redundant RR
Usually a cluester of client have single RR, and cluester will be identified with ROUTER_ID of the RR.  It is possible to use redundant RR in a cluster to avoid single point of failure.  All RR in a cluster can be configured with 4-byte CLUSTER_ID so that RR can discard routes from other RRs in the same cluster.


To configure route reflector in Juniper router see Configuring Route Reflector in Juniper Router.

Reference:
http://www.faqs.org/rfcs/rfc2796.html