We feel that the cost and priority structure which many users have enjoyed thus far, transparent to their behavior and often facilitated by government bulk subsidized funding, is becoming untenable in the current environment that is rapidly evolving towards a market driven infrastructure. Presently users perceive little if any incentive to limit their use of network bandwidth, and the appetite for bandwidth is growing far faster than the government can reasonably support. We propose a strategy to address this situation in the short-term, taking advantage of an existing but unused aspect of the Internet Protocol (IP).
On the surface, the technical changes required for our proposal are twofold: software in Internet routers would require modification to support the IP Precedence field for queuing; and participating application software would also need modification to set non-zero values of the IP precedence field. However we see the social obstacles as at least as great as the technical ones. Suggesting voluntary compliance with a policy to lower one's own favor in the eyes of the network may seem absurd in the Internet community. The present decentralized structure of the Internet has not only allowed incredible technological advancements in application technology (video, audio) and overall network usability (gopher, mosaic), but has also had major social impacts on the notions of individualism and autonomy. In short, users who inject real-time video may be able to clog the backbones no matter what. Incentives are necessary to address such behavior, while at the same time still allowing the Internet to evolve.
However, we also see the potential for the culture to shift somewhat, as people see lower performance for small applications (e.g. massive video streams slowing down simple ftps). Tracking and publishing statistics on usage and congestion can also raise awareness and perhaps accelerate this cultural shift. We advocate phasing in an implementation, over several months and possibly years. A typical Internet communication may involve two campus networks, two regional networks, and one or more very large scale backbone networks, for a total of five or more concatenated networks. Yet if any one of them implements precedence queuing, performance will improve. That is, incremental implementation yields incremental improvement; there are no ``flag days'', and implementations are backward-compatible with the existing Internet technology. We anticipate that long before network and application conformance reaches 80%, or even if it never does, the system will be providing substantial benefits to those users and routings that are compliant.
We recognize that an IP Precedence-based service, even in conjunction with accounting and/or charging, is not going to protect the Internet from applications that do not fit the basic premise of the current Internet: stochastic sharing. That is, traffic priorities, or even accounting, are not a solution to this problem. But accounting can help this problem in the transitory phase of the current environment. In addition to billing applications, maintaining statistics for the variety of multiple services can also yield a better understanding of network demands, a valuable knowledge base to develop interim policies, capacity planning, and fairer and potentially more competitive cost recovery in the future.