A Case for Relative Diierentiated Services and the Proportional Diierentiation Model

The Internet applications and users have very diverse service expectations making the existing forwarding model, which treats all packets the same, inadequate and limiting. Consequently , there is now a widespread consensus that the Internet architecture must be extended to allow certain users and applications to get a better service than others at a higher cost. One approach, called absolute service diierentiation, uses an admission control scheme to provide users with assurances of requested absolute performance measures such as rate or a maximum end-to-end delay. That is, users specify their desired performance levels and the network either accepts or rejects the request based on the availability of the resources to provide the necessary assurances. An alternate approach is to ooer users a relative service diierentiation in which the network supports a small number of service classes with assurances of only a speciied relative ordering between performance levels observed in each class. The applications and users must adapt their needs and (dynamically) select the class which best meet their needs. A pricing or policy constraints prevent all users from choosing the same class for their traac. The relative diierentiation model can be further reened and quantiied using a model we refer to as Proportional Diierentiation. The key feature of this model is that the the quality spacing between classes is controllable from the network operator's point of view and predictable from a user's point of view. Furthermore, as highlighted in this paper, there are scalable forwarding mechanisms (packet schedulers and buuer managers) to realize Proportional Diierentiation in the network.