Nimrod - An Internet Architecture

Introduction

Nimrod described a new routing architecture for an Internet, an architecture that achieved scaling by distributing and diversifying the routing. Nimrod distributed the routing by allowing the end nodes to choose their own routes (through map retrieval and path setup) and diversified it because with such a scheme there is no requirement for everyone to run the same routing algorithms.

Back when the IETF was thrashing because the sky was falling, we were running out of IP addresses and we had to do something to prevent the collapse of the Internet as we knew it, I wrote a short paper that tried to lay out the thinking that seemed to me to lead inevitably to a Nimrod-like solution. Of course, that's not the direction the IETF chose to go.

After the Nimrod work, two fundamental problems continued to bother me. One is an ongoing problem in the Internet that started well before Nimrod, that is the DNS. The other is a part of Nimrod and that's how Nimrod's addressing would work. Two rather different pieces of an Internet but they came together in trying to find a solution.

Here I write down my thoughts on these matters and try to do a paper design of a new Internet. It's based on the work of Nimrod so if you're going to read this you should probably first have an idea how that works as I don't cover that ground again.

Note: Noel has put up all the documentation he could find on Nimrod and the BBN work is available from their ftp site.

I also haven't figured out yet how to present this in a coherent, linear fashion so that at any time I'm only talking using terms I've already defined. Thus it might take a couple readings before the pieces start to fit together.

Please, if you have comments on these ideas, feel free to send me email.

• The Problems -- Two problems led me in this direction of thinking, naming and addressing. You can't get much more fundamental than that.
  • Naming - Where the DNS has to go.
  • Naming 2010 - More thinking about the naming issue circa 2010.
  • Addressing - An alternative to hierarchical addressing.
• My Architecture --
  • Endpoints - Where it starts and ends.
  • Sites - Machines under a common administrator. The hope is to make everything beyond this (everything that would be higher in the hierarchy if we had a hierarchy) configure automatically.
  • Coalitions - Sites combine for Aggregation and Abstraction.
  • Naming and Routing - In the first step these end up completely intertwined. There's a second step to routing which is one of those "thrown over the wall" problems that need not be centralized and I haven't written anything about it yet.
  • Transition - How to get from here to there.