| Abstract |
The traditional Internet provides only the best effort class of service, with no guarantees in terms of packet loss, delay or bandwidth. On the other hand, recently new applications, such as real-time video or Internet telephony, with Quality of Service (QoS) requirements have been developed rapidly. Furthermore, Internet Service Providers, require that their networks provide traffic engineering and resource management capabilities, in order to be able to support and charge applications demanding QoS. This Master's thesis deals with the problem of routing flows and managing resources in MPLS (MultiProtocol Label Switching) networks. MPLS is a label switching technique that is applicable to various kinds of networks, including IP-based networks. We assume that the MPLS network studied is capable of supporting a number of different classes of service. Incoming flows belong to one of the supported classes and can have different bandwidth requirements. Also, flows should be routed in a manner that optimizes network performance, while situations where some parts of the network are congested and some others are underutilized should be avoided. It is important for the network to support mechanisms that control the resources allocated to classes, in order to guarantee the QoS that each class is supposed to provide. We designed a system's architecture that provides mechanisms for traffic engineering and resource management in MPLS networks. For the support of different classes of service, the Differentiated Services (DiffServ) architecture was adopted. We introduced a method for defining DiffServ classes of service and their respective quality. We first studied the use of implied costs theory in routing and resource sharing among the supported classes. This theory deals with the problem of routing calls and allocation resources in circuit-switched networks, which have significant differences from the type of networks that we study. Thus, it was necessary to modify and adjust the implied cost theory in order to be applied to MPLS networks. The system that we designed for routing and dynamic resource management was implemented in an actual MPLS network. Two additional systems were developed and implemented. These systems use simpler algorithmic approaches in order to perform flow routing and resource management. The first system supports dynamic resource management and performs route selection based on advertisements of the available resources of each path. The second system uses estimates of the mean traffic per class on each path in order to compute the resources that should be allocated. The resource allocation is performed statically, i.e. resources are not dynamically reallocated during the operation of the network. Incoming flows are routed so that the amount of actual traffic carried by a path converges to the estimated mean traffic. We conducted a number of experiments in order to evaluate each system. In particular, the evaluation was based on the degree of assurance of the QoS provided by each class and on the revenues acquired by the network provider in each case. In order to compute the provider's revenue, we defined a method for charging flows based on the class of service and the bandwidth requirements of each flow. The experimental results revealed the need for dynamic resource management mechanisms for assuring the QoS provided by each class. In particular, the two dynamic systems outperformed the static one. Regarding the comparison of the two dynamic systems, i.e. the system based on implied costs and the one based on available resources advertisements, their effectiveness was seen to be similar, although the latter was considerably simpler to implement. For the purposes of the experiments, a traffic generator was implemented to generate the traffic that was inserted in the network during the experiments. Flows with different transmittion rates were used to emulate a number of different network applications.
|
Δρομολόγηση Ποών και Διαχείριση Πόρων σε Δίκτυα MPLS που υποστηρίζουν Πολλαπλές Κλάσεις Υπηρεσίας
