Monday, November 17, 2008

DTN

This paper is kind of short (I'm not complaining), and generally outlines difficulties with modern routing across the internet, especially with respect to wireless networks, ad-hoc networks, and sensor networks. It then suggests a partitioning system in order to deal with these difficulties, and presents some design decisions that it thinks would be best.

The first thing that is suggested is a partitioning into various independent networks, which can be addressed by name. One of these, of course, would be the internet as we know it today; the rest would be networks such as military ad-hoc networks. The key to communications between different networks would be DTN gateways. In general, a message would be destined for a particular network, with a particular address within the network. The DTN gateway then do late resolution, where the latter half of the address is resolved once the data enters the appropriate network.

I like some of these ideas, and must disagree with others. Partitioning the network into multiple address spaces is exactly the kind of thing that the original internet was supposed to overcome. The idea of IP over everything still has merit (in my personal opinion). However, I think the idea of using DTN gateways to control traffic moving into a wireless network is a relatively good idea, for reasons which are below.

The main idea behind DTN gateways is that they will control all traffic headed into their particular network. This includes long-term storage for data that is destined for nodes that are currently disconnected, as well as address translation/lookup. The long-term storage makes reliability much easier for long-term connections (essentially, the responsibility for delivering the data is delegated to the DTN gateway), and address translation is good for networks such as sensor networks, which like to use address compression to save bandwidth.

The authors have a security suggestion that involves edge routers in the internet verifying public-key encrypted information from end nodes. This seems like a generally bad idea, given the intensity of computation, and the ability to crack public-key encryption given sufficient time and resources (of course, it depends on the strength of the encryption). Still, they recognize the need for authentication, which is something that the internet doesn't do well right now.

On the whole, I'm not sure this paper provides very good solutions to today's problems, but it does seem to identify the problems very well.

1 comment:

Randy H. Katz said...

So interesting enough, elements of this architecture have recently become operational on a deep space probe. Primarily the deferred gateway ideas. As you can seen from both papers, there is a lot of thinking on naming and transport as next steps in a new generation Internet. This paper sets its flag in one possible direction -- rather than something that hides it all, make the differences of the pieces more visible.