Felten, Edward W. “Nuts and bolts of network neutrality.” Center for Information Technology Policy, July 6, 2006.
Summary: An impartial introduction to the structural characteristics of the Internet that are most relevant to the issue of network neutrality. The key points:
The economic heart of the issue is control over the direction and speed of innovation on the Internet. By design, the Internet traditionally favors innovation in applications and services intended for end users, while requiring the devices and applications that carry and route information from one end user to another to follow a relatively simple, relatively static, relatively conservative and consistent design. This design was chosen partly because end users have much more computing power available, and partly because end users have a much better idea of what they want. The corporations that provide the means for carrying and routing information oppose network neutrality, because they want to have more opportunities to innovate, which they believe will lead to greater profits.
A routing device in the middle of the network may on occasion receive so many requests in so short a time that it is forced to delay a packet of information that is being transmitted from one end user to another. If the requests pile up even more, the router may even be forced to discard some packets. In choosing which packets to discard, the router may apply some criteria that discriminate among packets, treating some of them as more important than others. Indeed, a router may be programmed to discard packets that its programmers regard as unimportant, even when it would be possible to handle all the requests. Discrimination of the latter kind (“non-minimal discrimination”) doesn't improve the router's performance or decrease its cost of operation.
Even delaying packets can operate as a kind of discrimination, and Internet services that involve transmission of audio or video in real time are particularly vulnerable to discrimination of this kind.
However, it is often difficult to determine whether a company that provides facilities for carrying and routing information has programmed its routers in a discriminatory way. There can be subtle technical reasons for companies to adopt routing strategies that have seemingly discriminatory effects.
Traditionally, the plan of keeping routers simple and non-discriminatory has encouraged an important measure of cooperation from end-user applications: When a router is forced to drop packets, the end-user applications, detecting this fact, lower their transmission rates so that the router has a chance to catch up. The number of requests reaching the router gradually decreases until it becomes possible for the router to handle them all. Routers that practice discrimination violate this social contract; if a router is going to drop more of one particular end-user's packets, there is less incentive for that end user's applications to lower their transmission rate. This could, in principle, be a serious problem for Internet communications generally, because if many users transmit at the highest possible rate all the time, there will be a lot of congestion and everyone's service will be degraded.
Quality-of-service guarantees, which are often cited as an argument against network neutrality, are actually helpful in only a few cases (real-time audio and video). Even in those cases, such guarantees can often be met even in the absence of network neutrality, by providing greater bandwidth; and in the cases where such guarantees cannot be met, they often could not be met even with the help of routers that discriminate among packets.