Inter-domain Internet Routing

This lecture offers a very clear description of how the inter-domain routing works in the Internet. The Internet is divided into Autonomous Systems (ASes), each identified by a 16-bit number, that are owned and administered by a single commercial entity. Border Gateway Protocol (BGP) is a way for ASes to exchange reachability information among each other. ASes could be universities, enterprises or Internet Service Providers (ISPs) of different magnitudes. Big ISPs provide connectivity to smaller regional ISPs, and other ASes get their connectivity through the one of them. Transit relationships are between ISPs where they provide access to each other’s routing tables and involves financial agreements (provider-customer). Peering relationships is about providing mutual access to a subset of each other’s routing tables. This does not involve any direct financial settlement and is mostly for mutual benefits. Tier-1 (global) ISPs are in peering relationships with each other for global routing information. BGP is a protocol that goes beyond simple constraints of routing and reachability, and incorporates economic and other seemingly fringe considerations in routing decisions.

Exporting routes to neighboring ISPs is a decision involving careful consideration as no ISP wants to transit packets it isn’t making money on. Providers obviously export all of its customers’ information. But clearly an ISP does not want to be exporting routes exported to it by its provider as it would be paying money for it. In peering relationships, an ISP exports only its transit customers’ routes, and some mutual customers. Also, customer routes are given higher priority over routes advertised by other providers or peers. Likewise, peer routes are more preferable than provider routes, for economic reasons. BGP routers exchange route information over BGP sessions. Internally, an AS has an internal protocol like OSPF or RIP connecting its routers. Externally learned routes are propagated to other external routers through route reflectors, thereby ensuring scalability.

Questions:

1. Does it happen that ISPs prefer provider routes over peering routes for performance reasons? For example, it might be beneficial to reach an AS quicker through a major ISP than through peer links…

2. Typically the topology of routers in the AS is expected to be static. Why not build a multicast tree sourced at the border router (one each for every router) and send it down this tree? Wouldn’t this be the most efficient?

3. In the route selection process (Table 4-2) why is a route learnt via eBGP preferred over one learnt over iBGP?

Comments:

1. It would be good to see how the ASes are distributed in the geographical area. Is it possible that a bunch of regional (Tier-2) ISPs gang up and form a p2p AS-network among them, totally bypassing the Tier-1 ISPs? Or is that impossible?

2. Crazy question: Assuming that an AS spans a large geographical area (say from CA to WA). Will it be possible that routers within an AS is now connected by an external ISP? J Theoretically, it shouldn’t matter, I think…

3. This paper should definitely be on the reading list – it is neat and clear in explaining BGP.