Granular differentiated queueing services and cost model for end-to-end QOS: principle

One weakness of DiffServ is the lack of granularity for QoS guaranteed services, which makes it difficult to cost-effectively support end-to-end (e2e) QoS according to the e2e situation of applications. With the conventional packet-level QoS mechanisms for the regulated traffic, i.e., buffer admission control plus output schedulers in general, increasing service granularity may inevitably complicate implementation and/or impact scalability since sophisticated output schedulers seem necessary in this case. In this paper, a new structure, differentiated queueing services (DQS), is discussed to handle the above issue. DQS tries to provide granular and scalable QoS guaranteed services to be selected by users according to their QoS requirements and e2e situation. Its basic idea is to convert packet delay guarantee into packet loss ratio guarantee with either dropping or marking the packets whose e2e delays are perceived unable to be guaranteed. Packets are queued according to their e2e delay requirements so that various delay bounds can be guaranteed without using sophisticated output schedulers while different packet loss ratios are mainly controlled by call admission control (CAC). A cost model for differentiated QoS services provided by DQS is also discussed with a possible CAC based on the exponentially bounded burstiness traffic.