This lecture provides an introduction to the global interdomain routing system. The reading discusses BGP4, the current protocol for propagating information on how packets should be routed among the different autonomous systems (AS) that make up the Internet. There are two flavors of BGP:
- External BGP is used to propagate inter-AS routing information from one AS to another.
- Internal BGP is used to propagate inter-AS routing information within a single AS; that is, it propagates information that an AS has learned via eBGP to the other routers within the AS. This is distinct from an AS-local route propagation protocol (“IGP”), which just describes how to route packets within a single AS.
After reviewing the goals of Internet routing and the business practices followed by network operators, the lecture discusses the BGP protocol, which is quite straightforward. BGP configuration within a single AS is also discussed: for large ASs with thousands of routers, route updates must be propagated carefully to achieve both scalability and consistency. From a database perspective, propagating BGP configuration updates within a single AS doesn’t seem like a fundamentally hard problem, so it was interesting to see how it was treated by the reading.
I liked how the lecture discussed the economic motivations of the participants in the global routing system. It would be interesting to apply game theory to the empirical behavior of ISPs, to understand how close their behavior approaches “optimal”: how successful are ISPs at minimizing the cost incurred by their routing choices, and maximizing the value they provide to their customers? Could the efficiency, cost-effectiveness, or reliability of the routing system be improved if a single authority was designated, rather than relying on competition? If theory suggests that the market participants are acting sub-optimally, is there an opportunity for arbitrage, essentially acting as a “virtual ISP”? Alternatively, what if there was auction-oriented approach to matching supply (packet routing capacity) with demand (ISPs and customers that need to route packets from A to B).
I was really surprised by the security and robustness properties of the currently-deployed BGP architecture. Given the massive economic value of the Internet and the obvious presence of malicious actors, it seems amazing that, for example, there is no authenticated mapping from IP prefixes to autonomous systems or real-world identities. In class, I’d be interested to discuss what factors have prevented more widespread deployment of S-BGP and related technologies, which is only eluded to briefly in the reading.