Remaking the Internet

The Time Warner Research Program has published my essay on the Internet architecture, Remaking the Internet: Taking Network Architecture to the Next Level. The essay examines the issues that make it difficult if not impossible for the Internet to continue to grow and meet the challenges that innovation throws at it. The main issues, as I see them, are:

  1. IPv4 address exhaustion
  2. Routing scalability
  3. Security
  4. Congestion management

Let’s look at them one at a time.

Addressing

The problem with IPv4 addressing isn’t so much the fact that the 32-bit IPv4 can only address 4 billion systems, it’s the fact that each device that provides a service on the Internet (or on any network connected to the Internet) needs to publish its address within the context of a routing system that knows how to reach it with no preconditions and doesn’t adapt well to variations in routing. The alternative is for each system to have an address of a system within its own administrative domain that knows how to reach the other system. This sort of arrangement is similar to what we do with our home gateways today, but it can be made much more sophisticated.

Routing

The Internet’s routing problem is related to its addressing problem. It should be possible for a system to be either mobile or multi-homed (directly connected to more than one network at a time.) Multi-homed systems would be capable of surviving failures on any of their connections, so a server attached to both Global Crossing and Verizon, for example, would have more resilience and greater capacity than one with a single connection. Large servers to this today, but they require a unique IP address for each network interface, and that defeats much of the potential of multi-homing. This is an example of the benefit giving each system its own identity on the Internet independent of location. The phone network manages to do this with portable phone numbers but the Internet doesn’t.

Security

Internet security is troublesome for a number of reasons, but the main one is the idea that security should be enforced only at the network endpoints. Security at this stage is valuable (and probably indispensable) but it’s better to check for authorization and authenticity at multiple points in the path between one end system and another. Denial of Service attacks are a key vulnerability that the present system of end-to-end security creates. If it were necessary to communicate authorization at each network boundary, DoS attacks could be stifled at the edge of the source network. This doesn’t impose an end to free speech, but it does provide a way for networks to limit inbound traffic and to throttle systems that are known to be abusive.

Congestion Management

As Internet applications become more diverse, the variations in their cost sensitivity and performance requirements becomes apparent, but the current convention of moving packets across network boundaries at the same service level prevents rational pricing from becoming a reality. This harms consumers, especially those who are first to adopt innovative new gaming and communication platforms and those who use high-volume data backup services and the like. There is some great work being done in the Internet Engineering Task Force (IETF) on congestion management that may never see the light of day because the Internet’s architecture makes it difficult to explore alternative approaches.

There is a lot of talk about the horrors of monopoly among Internet policy scholars in the law schools, but I’ve yet to see any of them address the disadvantages of the “design monopoly” represented by TCP/IP. Like any other technical system, this one was created by human beings with limited knowledge of the future, so there are bound to be mistakes. When there is one and only one protocol to rule them all (IP) every mistake is magnified to the utmost possible degree. That’s why the things that I’ve mentioned have serious consequences even though they may seem minor to the untutored eye.

The paper addresses a number of other issues that are outside the scope of TCP/IP but nonetheless have big side-effects on network cost and utility. It’s probably hard reading, but it covers some important ground. Once again, the link is Remaking the Internet: Taking Network Architecture to the Next Level.