Think of the following topology.
Here, R1,R2,R3, and R4 are eBGP routers. R3 and R4 also iBGP routers in order to communicate. But in some sample diagrams I noticed, for example, R9 and R8 could be an iBGP router but not eBGP router also. What may be the point in defining a router as only iBGP? Because iBGP uses underlying IGP, I see no use in adding a iBGP router.
The Internet is full of reasons to use iBGP. For instance, iBGP reasons? on the NIL Forum. BGP also uses a standard TCP connection to exchange route information. A big reason is that you want to redistribute your IGP routes into BGP, but you don't want to redistribute the BGP routes into your IGP; this can cause routing loops. Yes, it is easier to provide a default route to the eBGP routers, but that is not always appropriate, either.
iBGP inside of an AS is only used when:
- you need to configure border-routers (R3+R4)
- the network is too big and the OSPF limitation is reached
You normally split a huge network in smaller areas.
IMHO there is no need for iBGP as an IGP like in the 2nd case.
You normally split big networks in smaller areas which are easier to maintain.
iBGP is for internal use on an AS, is entablished between routers on the same AS. By other hand, eBGP is between routers on different AS, is on the way that internet works. On your question, you said that why routers on the same AS some time need to learn routes learned by BG, i guess, because BGP works with the major posibilities of some routing protocols, it can be used to control the way that traffic flow on your AS: maybe you are a transit AS connected to different ISP, Providers an so on, then you can control the path that traffic get to reach some networks, and with that, saving money(maybe your L2 providers have different prices), providing the less congested link to some ISP, use the fastest link to leave your AS, and some other scenarios.
In a transit network (ie one that forwards traffic to other networks), routers which provide paths between border routers (such as Rs 5,6,7,8,9 in your drawing) must speak iBGP in order to distribute the routes the border routers receive to the other border routers.
eBGP makes your border routers learn about best routes to reach each prefix on Internet.
iBGP will let all your internal routers know about those external prefixes.
R4 for instance should know that some prefixes are preferred through R1, even if it's directly connected to R2.
iBGP does usually not use internal routing protocol directly. If R1 anounces a prefix with its local interface as a source, either:
In both cases an internal routing protocol is needed to propagate internal routes (in one case including external links, in one case not).
Routers don't just need to exchange routes, they also need to actually forward the traffic.
The results of routing decisions are not stored in the packet in any way, so ALL the routers along the path need to know a route that will successfully deliver the packet to it's destination without it getting caught in a routing loop.
iBGP is the normal way to distribute information about external destinations to routers within a AS.
In your diagram R10 and R11 don't really need to know about individual external routes, they only have a single route to the outside world, so they can simply be configured with a default gateway.
R3, R4, R5 and R6 definitely need to know about external routes, so that when they receive a packet destined outside the network, they can send it towards the correct exit.
R7, R8 and R9 could benefit from knowing about external routing information but it's less critical. They could simply be configured with default routes pointing towards R5 and R6 but it may result in less efficient routing and less resiliance against failed links.
