Dynamic class selection and class provisioning in proportional differentiated services

The relative diierentiation architecture does not require per-ow state at the network core or edges, nor admission control, but it can only provide higher classes with better service than lower classes. A central premise in the relative diierentiation architecture is that users with an absolute QoS requirement can dynamically search for a class which provides the desired QoS level. In the rst part of this paper, we investigate this Dynamic Class Selection (DCS) framework in the context of Proportional Delay Diierentiation (PDD). We illustrate that, under certain conditions, DCS-capable users can meet absolute QoS requirements, even though the network only ooers relative diierentiation. For a simple link model, we give an algorithm that checks whether it is feasible to satisfy all users, and if this is the case, computes the minimum acceptable class selection for each user. Users converge in a distributed manner to this minimum acceptable class, if the DCS equilibrium is unique. However, suboptimal and even unacceptable DCS equilibria may also exist. Simulations of an end-to-end DCS algorithm provide further insight in the dynamic behavior of DCS, show the relation between DCS and the network Delay Diierentiation Parameters, and demonstrate how to control the trade-oo between a ow's performance and cost using DCS. In the second part of the paper, we consider the related problem of class provisioning. At the provisioning phase, the network manager conngures the link to support the QoS requirements of all traac types. Each traac type is speciied by an expected arrival rate and a delay requirement. The objective of the provisioning phase is to jointly determine: the minimum link capacity needed to support the given traac types, the nominal class of service for each traac type, and the appropriate resource allocation between classes. Our class provisioning methodology is also based on Proportional Delay Diierentiation (PDD). The major advantage of PDD is that it avoids the computation of an explicit bandwidth share for each class. The class provisioning methodology is illustrated with examples.