| Abstract |
This Master's thesis concentrates on the development and refinement of sophisticated tools, based on applied probability models, to be used for the evaluation of the performance of ATM networks, for charging and pricing services, and for the effective traffic substitution. Statistical multiplexing and resource sharing, will be essential aspects in future ATM networks. These can be attained by analysing the performance of such networks on the basis of the concept of effective bandwidth. In fact, this concept has a unifying role as a parameter summarising the statistical characteristics of sources over different time scales. When an effective bandwidth can be appropriately defined and measured, then problems such as Call Admission Control (CAC) and pricing greatly simplify. However, analytical methods are usually hard to apply (if at all possible), particularly when real traffic sources are involved. Thus, simulation and on-line methods are invaluable tools for related investigations. An existing asymptotic result, justified by the theory of large deviations, is employed in our work. This is based on a multiplexing model, where the superposition of groups of traffic sources feeds an ATM link in which the capacity and buffer size also increase proportionally to ; this result is known as the "large asymptotic". A measure of the Quality-of-Service (QoS) offered, is the cell loss probability (CLP), which is used as the primary QoS criterion. This thesis focuses on two main problems related to the concepts mentioned above: charging and pricing network services, and substitution of real traffic streams with an equivalent stream of modelled traffic. The first part, deals with evaluating charging schemes for broadband network services. This is an extremely vital subject in modern telecommunications an has recently attracted considerable attention. More specifically, we study, implement and validate two charging formulae based on models that address needs such as fair charging according to effective usage of network resources. The charging formulae are based on bounds for the effective bandwidth. We compute these charges, and compare them to those based on the accurate calculation of the effective bandwidth of the source at the end of the call. Through this comparative study, we assess appropriateness of these schemes for charging different traffic sources. Simulation experiments for typical calls from real (MPEG video) and Markov-Modulated Fluid (MMF) sources, are performed. The schemes investigated have low implementation cost, and can lead to efficient usage of network resources when pricing is combined with CAC. The second part of this thesis concerns aspects on traffic substitution, derivation of equivalent traffic, and assessment of models for real MPEG video sources. In particular, in deriving the equivalent traffic, we are seeking for a modelled source (with known statistics) that may replace a given subset of real sources feeding a multiplexer, without affecting the QoS. Two different models, which are widely used to characterize real-time services are considered. We generate traffic from AutoRegressive(1) and 'ON/OFF MMF' sources, in order to achieve equivalent traffic. Two methods, a traditional one and a more elaborate new method, are employed to derive values of the appropriate model parameters. Evaluation of these methods is based on simulation experiments for real and modelled input traffic in an ATM link which serves a single buffer in a FCFS manner. A simulation study was conducted which investigated both the proposed new methodology and the traditional, and compared their respective performance.
|
Θέματα σε Χρέωση και Ανάλυση Επίδοσης Κίνησης ΑΤΜ
