An efficient network architecture motivated by application-level QoS

We consider application-level quality of service (QoS) requirements in design of networks that provide delay and loss guarantees. Using this approach, we have designed a novel network architecture that in many aspects is more efficient than ones motivated exclusively by packet QoS. An overview of our work is presented in this paper. The centerpiece of our design is a traffic model that enables delivery of application specific information to the network. Based on the traffic model, we have developed (i) efficient network techniques for providing application-level delay guarantees, (ii) an admission control algorithm for a statistical service that bounds application data losses below a specified value, and (iii) network techniques for managing application data losses to achieve fairness and protection of high priority data units marked by applications (e.g., I frames of MPEG applications).