A New Blueprint for Network QoS
David C. Reeve · Kent Academic Repository (University of Kent) · 2003
In this thesis we present a new approach to the provisioning and delivery of Quality of Service in packet switched data networks.Such networks are endeavoring to support an increasing range of applications.For an application to deliver a successful outcome to its user it requires statistical guarantees of data transport performance from the network.These guarantees must be 'strong' in the sense that they hold over short timescales.With such a statistically strong approach to the planning, provisioning and delivery, it becomes possible to bridge the gap between the performance aspirations of applications (and hence their users) and the operational performance of the network.This thesis is founded upon the following ontological basis.Finite queues have two degrees of freedom in the three parameters throughput, loss and delay; any QoS scheme must account for this fact at all its stages to be successful.All network elements have a nite capacity and, as they process the load oered to them, introduce quality degradation.They possess an 'intrinsic' quality, which can only be shared among their customers.It is the conguration and management of this sharing that delivers dierential quality of service.Ultimately delivering quality of service is assuring a bound on the degradation experienced.Using this basis we develop a methodology for creating and reasoning about statistically multiplexed data networks.In these networks the operational behavior can be made to mirror the mathematical foundations of queuing theory more closely than is commonly believed.We illustrate how the provisioning, design and implementation process can be approached in a consistent fashion, using new packet handling mechanisms which have evolved from the ontological basis.We demonstrate the use of this approach by conguring examples of multi-service networks and demonstrating that their target properties are achieved through simulation.