InetDaemon.Com

Route advertisements are not sent out on regular intervals as in other protocols. A full table exchange is sent out when BGP is first started, and then only incremental updates are sent when changes occur in topology.

BGP is a PATH VECTOR protocol, which means that it does not keep track of internal routing within the AS, but rather keeps track of paths through other autonomous systems to reach destination networks. Any network that is not being advertised, cannot be reached.

BGP uses the UPDATE message to advertise routes. It is important to remember that BGP routers listen for routes. A BGP router cannot forward a packet if it has not heard a route. By the same token, if a route isn't being advertised, it is also not possible to forward packets.

The UPDATE message contains:

1. Network Layer Reachability Information (NLRI)
   1. Prefix
   2. Length
2. AS_Path
3. AS_Path Attributes

When a BGP peer becomes unreachable, BGP sends UPDATE messages that contain 'withdrawal' information, requesting that other BGP routers remove those routes from their tables.

eBGP peers will advertise all known eBGP routes to all other eBGP peers. iBGP peers will only advertise their own internal routes to other iBGP peers. A BGP speaking router will never advertise another iBGP peer's routes to any other iBGP peer.

To clarify, I will use an example. Picture a string of three routers:

A = B = C

1. Routers A and C are iBGP peers of B.
2. Router A tells Router B about it's directly connected networks.
3. Router C tells router B about it's directly connected networks.
4. Router B will never tell A about C's routes, and will never tell C about A's routes.

If it is desirable for A and C to know each other's routes, then either use an Interior Gateway Protocol of some kind (RIP, OSPF, IS-IS, IGRP) to pass this information, or directly connect A and C so that you will have built a fully meshed BGP network like so:

    A
   / \
  B---C

1. Routers A and C are iBGP peers of B ( A <-> B <-> C )
2. Router A tells Router B about it's directly connected networks.
3. Router C tells router B about it's directly connected networks.
4. Router B will never tell A about C's routes, and will never tell C about A's routes.

This is a loop-prevention algorithm. The path between A and C should be managed by an interior gateway protocol.