End-to-end absolute differentiated services for real-time traffic

For many real-time applications (i.e. VoIP), the absolute quality of service (QoS) requirements should be guaranteed. In this paper, we propose a novel dynamic congestion control framework for scalable support of absolute end-to-end QoS in differentiated services (DiffServ) networks. With dynamic measured feedback control and traffic regulation at edges and relative DiffServ support at core routers, the proposed framework provides quantitative per-class deadline violation probability (DVP) guarantee. More specifically, we extend relative DVP from a single node to an end-to-end path by developing a bound for end-to-end relative DVP. In order to support the framework better, we present a unified scheduler, the Dynamic Deficit Round Robin (DDRR) scheduler, which combines generalized processor sharing (GPS) with relative DiffServ. Simulations are carried out to illustrate the performance of our proposed framework and scheduler under both single-node and multi-node cases.