Achieving Equality is Critical to the Future of the Internet


Dr. Lawrence Roberts led ARPANET, the team that birthed the Internet as we know it today. The ARPANET cluster included distinguished individuals such as Vint Cerf, who created the core protocol TCP/IP that underlies the infrastructure of our modern IP-based communication systems. We are proud to have Dr. Roberts speak at our Structure 08 conference and proud to present his thoughts here on the problems P2P and inequality in network capacity cause for consumers.

Inequality, or “unfairness” in how network capacity is allocated between different homes or computers, is causing major reductions in the actual realized speed of Internet service for almost every user. The magnitude of the problem is well beyond what most people understand, with realized access speed often reduced to as little as a tenth of its potential. For the Internet to truly support all of our imagined uses — video, voice, gaming, social networking and the like — we must eliminate the basic inequality inherent in TCP/IP. To put it simply: Each user must receive equal capacity for equal payment.

Let’s consider the residential ISP market. The real goal should be to provide equal capacity to all homes that have paid the same amount, and on some scale, more to those that paid more.

In the current situation, pricing is flat, and any user, via a “greedy” program like P2P, can capitalize on TCP’s preference for multi-flow traffic and drag down the average capacity of all other users. So far, the most common approach to addressing inequality problems is Deep Packet Inspection (DPI), which literally inspects packet contents to find P2P applications — and then slows them down or kills them.

However, this inspect-and-destroy approach has led to a new kind of arms race: P2P applications add encryption and rapidly changing “signatures,” and DPI constantly races to catch up. In a typical network, DPI finds roughly 70 percent of the P2P traffic, and things will only get more difficult as encryption becomes the norm and signatures change even faster. Even at 70 percent detection, the remaining P2P still slows down all the normal users to a third of potential speed. The problem affects residential users, but it can be even more serious in a school or corporate environment. It is clear that DPI is doomed as a solution for containing P2P. However, a totally different solution is possible.

Each cable or DSL concentrator has a maximum capacity which must be shared at any moment. If all the traffic from each home was rate-controlled to share the total capacity equally, a P2P user with 10 flows would get 10 percent of the capacity per flow when compared to a neighbor downloading a new application with one flow. Both homes would get the same number of bytes delivered in the same amount of time. A third neighbor doing something simple, such browsing the web or checking his email, would get much faster service than before, since his short-duration flow would not experience any delay or loss. That is, unless he extended his session long enough that the total use neared that of the file transfer users. In that case, he would be treated the same as the others who are consuming the same amount of capacity for the same price.

Since this “automatic rate equalization” does not require inspection of every packet, it operates at full 10 Gbps trunk rates quite inexpensively compared to using many DPI systems, and the result is complete network usage equality for all users paying for the same service.

Once inequality is eliminated in the network, application vendors can stop devising techniques that unfortunately harm other users and start discovering techniques that deliver improved service. Easing traffic snarls will also bring down the cost of reliable, high-speed Internet service susbtantially. Without solving the TCP/IP inequality problem, providing affordable Internet service will become extremely difficult — if not impossible.


Dave Burstein

Dr. Roberts

I write the DSL industry news, and most of my data suggests that DSL networks are non-blocking much or most of the time. Can you point me to the data that led you to the conclusion of 1/3rd and 1/10th throughputs? Is that just the cable local loop, or something outside the U.S.?

Help appreciated
Dave burstein

Comments are closed.