BGP on loopback: advantages
Compared to the BGP per overlay design, the newer BGP on loopback design offers the following main advantages:
Significantly improved scalability
-
Improved scalability is particularly true on the Hubs, since they must reflect a much smaller number of routes between a much smaller number of IBGP peers.
Note that the problem of BGP Route Reflection is a “square problem” (O(n2)): the number of routes to reflect equals to the number of routes multiplied by the number of peers (except for the peer that has originally advertised the route). Therefore, reducing both these numbers significantly improves the scalability of the BGP Route Reflector.
-
Since only a single BGP route is generated for each LAN prefix, BGP ADD-PATH functionality is no longer required. Thus, the Hubs do not inflate the number of routes, neither when reflecting them between the Spokes within the region nor when advertising them to other regions.
-
Last but not least, no BGP convergence needs to occur, when a single overlay flaps anywhere in the network. (There is always just a single BGP route in the network for a given LAN prefix no matter how many overlays are currently alive.)
Significantly simplified configuration
-
With a significantly simplified configuration, the difference can benefit even a single region with a simple topology, but it becomes more apparent in topologies with multiple Internet links, and it is nearly unmissable when the solution expands to multiple regions.
-
Fewer BGP neighbors to configure: there is always a single BGP session between a Spoke and a Hub no matter how many overlays exist.
-
There is no need to configure tunnel IPs (consequently, no need to plan tunnel subnets).
-
There is no need to implement duplicate route filtering, because there is no BGP ADD-PATH and hence no duplicate routes.
-
Last but not least, new simpler overlay configuration can be supported for the topologies with multiple Internet links. See Topologies with multiple Internet links.