Hari Balakrishnan, Interdomain Internet Routing, MIT lecture notes, January 2009.This paper discusses the fundamentals of how routing between different administrative domains works in the Internet and how is it influenced by various technical as well as administrative issues by introducing the Border Gateway Protocol (BGP) version 4 which is the current standard inter-domain routing protocol.
The paper starts off by highlighting that the fundamental reason for the scalability of the Internet is due to the topological addressing scheme which allows Internet addresses to be aggregated in the forwarding tables of the routers. The current structure of the Internet infrastructure comprises of many ISPs that 'cooperate' and 'compete' amongst themselves to provide global connectivity. Based on their size, they can be broadly divided into 3 tiers with many Tier 3 ISPs comprising of a small number of geographically localized customers and few Tier 1 ISPs operating on a global level. These ISPs are more like autonomous systems (AS) and for effective routing on the Internet, it is essential for them to exchange reachability information. For this, we basically have 2 type of inter-AS relationships:
- Transit: This is a provider-customer relationship wherein the smaller ISP pays the larger one to provide access to all destinations in its routing tables.
- Peering: This is more of a peer based relationship based on mutual consent wherein two similar sized ISPs exchange mutually beneficial routing information amongst themselves. The author mentions that these peering deals are almost always confidential and not disclosed publicly. (...which I fail to understand why?)
- Scalability: It is important that the entire system of BGP updates scale well with the increasing number of ASes as well as the churn in the network.
- Policy: It is important that ASes have the freedom to enforce various forms of routing policies such as route filtering or internal route rankings.
- Cooperation: This is also an important aspect of BGP design. It is extremely necessary to have cooperation among various ASes resulting in a mutual beneficial environment in spite of having internal route filters and ranking policies for its own profit.
Overall, I really liked the paper especially because of the ease and lucidity of its explanation of the BGP and its implications. However, I would really like if we can have a deeper discussion on the scalability and convergence issues of BGP in the class. Moreover, it is important to realize that BGP doesn't always give optimal routes and what can be done for it to perform better in this area is very much an open question. Security aspects of BGP (S-BGP) is also an interesting area which remain to be explored. As a side note, the author mentioned somewhere in the paper that some ISPs now have 'delay guarantees' for certain kinds of traffic. Given the nature of the Internet, this statement was kind of intriguing for me. Another point that was a matter of curiosity was that the author mentioned that there are exceptions to point-to-point eBGP session. It would be really great if we could discuss some scenarios that call for such multi-hop BGP sessions.
Good comments. Re: peering arrangements, these are kept confidential for business reasons. If A peers with B and C, they may not want D is know this if he results in better reach for A and a competitive advantage versus D. There are examples of global peering alliances that are publicly disclosed: ATT+NTT+BT for examples, but the details of where and how are confidential. As for BGP Dynamics, you might be interested in looking at: 137. Z. Mao, R. Govindan, G. Varghese, R. H. Katz, “Route Flap Damping Exacerbates Internet Routing Convergence.” ACM SIGCOMM Conference, Pittsburgh, PA, (August 2002).
ReplyDelete