Queueing characteristics for markovian traffic models in packet-oriented networks

Abstract

This monograph presents the author's contribution to the field of performance evaluation of packet buffering processes in network nodes. In particular, a detailed characterization of finite-buffer queues fed by Markovian traffic models is presented by means of theorems and formulas. The Markovian traffic models were chosen due to their ability to mimic a very complex statistical behaviour of the traffic, including the self-similarity and long-range dependence. For analytical purposes, a new powerful method that can be used for all Markovian traffic models is proposed. This method combines the Laplace transform technique with the use of special recurrent sequences to solve large systems of linear equations. The method has several important advantages. Firstly, it enables an analysis of finite-buffer queueing models. This issue is crucial from the practical point of view as in all real devices (routers, switches) the buffering space is limited. On the other hand, most previous studies covered only infinite-buffer queues, less demanding in analysis. Secondly, it makes possible both transient and steady-state characterization of the queueing process. Thirdly, it allows for finding virtually all important performance characteristics and gives results in a closed, easy to use form. The following characteristics are studied by means of this method: the queue size distribution, the queueing delay distribution, the blocking probability, the loss ratio, the buffer overflow period, the time to buffer overflow and the statistical structure of packet losses. For each characteristic a new formula describing its transient and steady-state behaviour is obtained. The analytical results are illustrated by numerical examples, most of them obtained for traffic parameterizations based on IP trace files. The monograph consists of five chapters organized in the following way. In Chapter 1 the potential method is presented. This method permits finding closed-form solutions for large systems of linear equations in a special form. The systems of equations in this form appear frequently in the remaining part of the book and the potential method is used to solve them effectively. Chapters 2, 3 and 4 are the main part of the monograph and they present the queueing characteristics for different Markovian traffic models, with an emphasis on growing complexity of the model in every next chapter. In particular, Chapter 2 is devoted to the simple and compound Poisson processes, Chapter 3 focuses on the Markov-modulated Poisson process (MMPP), while Chapter 4 presents results for the batch Markovian arrival process (BMAP). All these chapters are structured as follows. Firstly, the description and the properties of the traffic model are given. Secondly, the theorems presenting formulas for the queueing characteristics are proven. Thirdly, the numerical examples are presented. Each chapter ends with a bibliographical note. Chapter 5 presents a set of mathematical and numerical tools that enable or make easier obtaining numerical results from formulas proven in the previous chapters. Namely, a set of algorithms for inverting the Laplace transforms and generating functions is shown, the uniformization method for computing coefficient matrices for MMPP and BMAP is presented, the expectation maximization algorithm for MMPP and BMAP parameter fitting is discussed and the continuous version of the total probability formula is recalled.