Summary of Network-Layer Discussion

The last few weeks have been instructive as our 660 class has discussed a dozen or so significant publications dealing with the network layer of the Internet's architecture. From BGP replacements to efficient queueing to network neutrality to multicast, we have covered a large set of the current problems in the network-layer.

One theme that seemed to recur was that there are lots of people and politics influencing the policies that are implemented in the Internet. For example, the network neutrality debate involves a number of parties; ISPs, customers, content providers, etc. Multicast is not just implemented at the network layer because it works well; router designers must be convinced it is worth their effort to modify their product, ISPs must be willing to deploy it, and there must be demand from end users.

Another theme that I noticed is that ideas often are not deployed immediately because of the risk and consequences involved in switching to an entirely new architecture. HLP is a protocol designed to overcome many problems inherent with BGP; but there is considerable risk switching inter-domain routing protocols; any problems would affect the whole Internet. It's for that reason that research looks for accurate ways to model and simulate performance of a protocol. The IPv6 upgrade that always seems to be on the brink of the future will carry with it some negative consequences for those that cannot upgrade. Because of the difficulties involved with adopting new architectures and protocols, many of these designs serve as building blocks and ideas for future protocols.

While it is not easy to get a network protocol deployed in the Internet today, there is considerable insight from the ideas suggested in the papers we have read. I was interested in new routing methodologies suggested. Routing table size is growing rapidly, so research is looking at new routing designs that can scale indefinitely. I was impressed with the ideas that came from ROFL and Compact Routing.