Quality of Service Performance Analysis based on Network Calculus

Data flows belonging to multimedia applications are gaining importance in the Internet. A key characteristic of such data flows is that they require Quality of Service (QoS). An Internet populated with data flows requiring QoS constitutes a paradigm change from the Internet in its early days. This has been accounted for in many research endeavors proposing new architectures, algorithms and protocols. However, one area that has been relatively underexposed is the development of new models for QoS. Hence, the vision that has inspired this dissertation is the development of a unified model for the performance analysis of QoS in the Internet. The potential benefits of such a model can be observed in other fields: Linear system theory is widely used in the analysis of communication and control systems. In this dissertation, contributions are made towards developing a unified model for the performance analysis of QoS in the Internet. The basis of the work is network calculus. Network calculus is a system theory for deterministic queuing systems, which was developed in the 1990s. The underlying rationale is that deterministic QoS guarantees can be obtained by traffic regulation, deterministic scheduling and admission control. Beyond that, this work integrates elements of system theory and queuing theory. The latter has been the method of choice for modeling data flows in the Internet since its infancy. The main requirements of the envisioned model are that it should give insight on relevant characteristics, should have a wide range of applicability and should be transparent and easy to use. Recent research results in network calculus which address these requirements are presented. These include statistical network calculus and transforms. Further, some open issues are identified, which are then dealt with in this dissertation. A network calculus analysis is conducted for dynamically reconfigurable networks. First, the network architecture which can be found in optical networking is presented. The key feature here is that packet forwarding is not only influenced by the routing, but also by the reconfiguration. It is shown how service curves can be determined for different reconfiguration schemes, thus enabling a QoS analysis. On a more general footing, in this chapter it is illustrated how current networking research issues can be translated into network calculus models. The next contribution is the development of a new transform for network calculus and its application. With the new transform the min-plus convolution, which is an important operation in network calculus, obtains a graphical interpretation and thus becomes easier to use. Based on the transform, theorems on the computation of the min-plus convolution are set up. These theorems are then applied to network design using service curves, with an emphasis on bandwidth/delay decoupled scheduling. Furthermore, network calculus and queuing theory are brought together. While network calculus focuses on the worst case analysis, queuing theory deals mainly with average behavior. It is examined whether the best of both worlds can be combined to achieve better models. First, analytical approaches are presented, which are then followed by a simulation. Finally, the achieved progress is summarized and some conclusions drawn