From the preceding sections on full mesh iBGP cores, and the scaling implications of iBGP cores, it's obvious that we should look for some way to eliminate the need for a full mesh among iBGP speakers in the same autonomous system. How could this be accomplished given the possibility of routing loops or reachability problems if we don't maintain a full mesh of connectivity? One possibility is route reflection, which reorganizes the iBGP core into a hierarchy and relaxes the iBGP route advertisement rules. Route reflection introduces the notion of reflected routes, which allows BGP speakers to break the rule of never advertising a route learned through iBGP to another iBGP peer in specific circumstances.
Route Reflection is described ...