The Internet evolution together with the popularity of real-time on-line applications has resulted in quality of service (QoS) being an important research topic. Over the last decade, the concept of QoS has evolved dramatically. Starting from the unsuccessful origins in the type of service byte (ToS) in the initial IPv4 specification, the networking research community has explored a wide variety of techniques ranging from per-flow QoS in Integrated Services (IntServ) to aggregate or per-class QoS in the Differentiated Services (DiffServ) model. This dissertation studies several schemes in the DiffServ environment to provide predictable QoS for scalable networks/systems.
While techniques to aggregate state information for scalability have been proposed in the literature such as with Differentiated Services, the per-stream effects of such aggregation are poorly understood. This dissertation first explores the effects of aggregated state information and proposes a scheme, called Granularity Aware $(m,k)$ Queue Management (GAQM), that improves control over the trade-off between scalability and per-stream QoS performance.
As the correct admission of flows in the DiffServ environment is critical to provide stable and predictable QoS to end users, this dissertation then proposes an approach, Edge-centric Resource Management (ERM), for admission control that exploits the unique architectural aspects of DiffServ. Through the use of periodic heartbeats emanating from edge routers to probe the network state on available egress paths, edge routers are able to quickly conduct admission control with a tunable degree of precision.
In addition to ERM, this dissertation presents two other admission control schemes as well, Fast Admission for Short Flows (FASF) and End-wise delegation for End-to-End Admission Control (E3AC). By enabling accelerated admission control at the edge rather than via centralized or in-path mechanisms, FASF not only reduces the burden on admission control, but also improves flow completion time and hence network goodput. E3AC, an approach to make significant strides in practical end-to-end QoS, provides a framework for catalyzing QoS amongst end ISPs while dramatically reducing the setup latency for QoS negotiations.