| Abstract |
Today's Internet provides only Best Effort service. Traffic is served as quickly as possible, while there is no guarantee as to timeliness or actual delivery. However, new applications that are demanding in terms of QoS emerge rapidly. This fact in conjunction with the continuously increasing amount of Internet users necessitates the provision of class guarantees as well as the control of the traffic inserted in Internet. Also, certain algorithms for the negotiation of the traffic contracts have to be introduced, in order both specific policies for network management and the interests of the end-users to be served. To accomplish the above, certain solutions are needed that use the current Internet framework. Such solutions should not modify in principle the network operation and nature, and they should be feasible and scaleable in the whole Internet, without inducing great overhead costs. None of the proposed solutions for QoS provision in Internet (RSVP, DiffServ, RSVP/DiffServ, RSVP/MPLS, MPLS, Bandwidth Brokers) serves all of these goals so far. In the present thesis, we define, implement, and analyze an architecture for the control and negotiation of the traffic to be inserted in an administrative network domain that provides differentiated services in its users. We have employed the Differentiated Services (DiffServ) architecture for the QoS provision, due to simplicity in its implementation and the scaleability it provides. The traffic inserted in this network is categorized in the various network QoS classes by means of a control environment according to the results of the negotiation algorithm. The control environment consists of a Policy Server (PS), the ingress routers of the network and the QoS functionalities of the Windows 2000 operating system. We use the RSVP signaling for the communication of the QoS provision requests for new traffic flows of the end-users to the network provider in a clear and controlled way. The PS functions as the network provider's representative, and thus all QoS provision requests for new traffic flows are sent to the PS by the ingress routers of the network using the COPS protocol. The PS negotiates with the User Agent (UA) of a user in order to select efficiently the Service Level Agreement (SLA) for a new traffic flow of the user, i.e. the categorization of the traffic flow in a QoS class in an efficient way. The UA functions as an "intelligent" user representative that serves the interests of the user in the SLA selection process. In this process, under our approach,a UA selects the SLA for a new traffic flow by maximizing the net benefit of the user that represents. For the purposes of negotiation, we develop an appropriate utility model that expresses user preferences in a simple yet informative way. Although our approach is more generally applicable, we assume that the PS charges proportional to the effective bandwidth of the traffic flows, thus providing the user with the right incentives for SLA selection. We retain free usage of Best Effort service, as in today's Internet. An important feature of our approach is the distribution of information enabling the efficient SLA selection. According to this distribution, each of the components involved only possesses those peieces of information for which it has an incentive to store. Another important feature of our approach is the simplicity of user's procedure for selecting optimal Service Level Specification (SLS) parameters. In order to achieve efficient resource management in the network, we employ Multi-Protocol Label Switching (MPLS) protocol for the implementation of the network. MPLS has such routing features that facilitate the traffic engineering procedures. We describe how, using our approach, the network provider can employ network management policies (in conjuction with appropriate charging) for traffic categorization into QoS classes. In the SLA control and negotiation environment that we have developed any specific negotiation process can be used. Using our SLA selection process we achieve individual optimization, by maximizing the net benefit of the user, and at the same time we attain improved economic efficiency. In particular, experimental results indicate that users are provided with the right incentives in SLA selection process. Having implemented our system in a real DiffServ/MPLS network environment, we proved its feasibility, and assessed its efficiency in terms of performance. Last, we studied possible extensions of our approach.
|
Σχεδιασμός και Υλοποίηση Αρχιτεκτονικής για την Αποδοτική Επιλογή και Παροχή SLA σε Δίκτυα DiffServ
