Supporting Rate Adaptive

this notion by saying client (i, r) has a minimally acceptable adaptivity parameter ai,r, which can be interpreted as a client’s individual QoS requirement. To emphasize this distinction between provider and client definitions of minimally acceptable stream resolutions, consider the example of a content provider offering a certain media stream with adaptivity a and maximum subscription level s. The adaptivity a is such that the provider’s assessment of the minimally acceptable stream encoding has an average instantaneous rate of as. Now consider two clients, denoted i and j, with individual adaptivity parameters ai and aj , such that a < ai < aj < 1. Both clients have a subjective minimum adaptivity exceeding that decided upon by the content provider; client i requires a minimum subscription level of ais and client j requires a minimum subscription level of ajs. The point is that optimal adaptation, which gives large volume streams the minimally acceptable stream resolution, as determined by the content provider, may prove unsatisfactory to clients with more stringent QoS requirements. In this chapter we propose a network architecture comprising K service classes, where each service class k carries a distinct normalized subscription level αk. Streams selecting service class k are guaranteed a minimum normalized subscription level of αk, but may receive higher subscription levels if sufficient capacity is available. For simplicity we will restrict our attention in this chapter to the single link case, but the development may be generalized to an arbitrary network. We also assume the link admits as many clients as possible while still respecting all of the QoS guarantees of the already admitted clients. Note that we assume clients are allowed to select their own service class, which means we will need to price the service classes in order to ensure client selection corresponds to actual client QoS requirements. Formally, we assume a price vector u = (uk, k = 1, . . . ,K), and a set of demand functions λ(u,α) = (λk(u,α), k = 1, . . . ,K), where λk(u,α) is the mean