Modeling and Simulation of Traffic Control Mechanisms in ATM Networks

Abstract

Broadband Integrated Services Digital Network (B-ISDN) represents the most important development in the evolution of telecommunication systems. The aim of B-ISDN is to provide an all-purpose, flexible, efficient and cost effective environment for all new emerging services based on voice, video and data in an integrated fashion. High speed networks provide real-time variable bit rate service with diversified traffic flow characteristics and quality requirements. The main challenge here is the efficient use of network resources and mechanisms in order to achieve a satisfactory quality performance. In order to achieve the aggressive goal, which B-ISDN aims at, a promising transfer and switching technique called Asynchronous Transfer Mode (ATM) has been adopted. The Quality of Service (QoS) measures, such as, cell loss and delay in ATM nodes are the parameters that significantly contribute to the degradation of network performance. It is worth noting that the control of each of these parameters has a special impact on the other parameter, for example, voice traffic can tolerate some cell loss but is very sensitive to cell delay, while video and/or data transfer applications can allow for some delay but not the cell losses. For video applications, the loss of consecutive cells becomes even more critical when reproducing the original data. Buffer management in queueing systems plays important role towards effective control of QoS for various types of applications. Traffic can be shaped by placing cells into buffers to compensate for a difference in rate of flow of data or time of occurrence of events. With larger buffers, the probability of losing cell decreases but the overall delay increases. Hence, it becomes necessary to utilise buffering and scheduling algorithms to regulate QoS attributes in a high speed network. In this thesis, the focus is given on two important design factors, namely, buffer management and scheduling algorithms. We assume that one can always improve performance by increasing speed, selecting a more efficient operating system, or even building specialised hardware to replace inefficient software components. However, for a pre-determined set of available resources, we examine the impact of buffer management algorithms on most important QoS attribute, i.e., cell loss. The network is prone to cell loss in the situation of congestion, when multiple cells blasting away simultaneously at peak rate through different incoming links attempt to reach the same outgoing link during the same cell slot time. In this case, only one cell is allowed to go through the network while the others must be stored in buffers. At this time, a switch buffering strategy as well as the buffer size becomes important since buffers are required to achieve low cell loss rate by providing a place to guard against cell loss when the switch is overloaded with bursty traffic. The control schemes usually categorise and manage the cells based on the cell loss priorities assigned to them, which determines the cells to be dropped in case of congestion. The main function of cell discarding mechanisms in congested network is to control the relative cell loss probabilities of different cell discarding algorithms. It has been shown in the literature that priority-based cell discarding improves the system performance for voice and video traffic. The present thesis considers the space priority mechanisms for optimising the network utilisation in broadband networks and evaluates the impact of controlling traffic and improvement of network’s throughput; when priorities are used. Two buffer management schemes, namely, Pushout scheme and Partial Buffer Sharing (PBS) scheme, use selective cell discarding of cells in buffer. It has been shown that the PBS scheme not only performs well to meet the QoS requirements for multiple priority classes but is also easy to implement. As such, this scheme has been proposed as a candidate for overload control mechanism. In a number of research works, the PBS scheme is analysed with different types of Poisson arrival processes. The challenge in designing a PBS scheme is to select optimal threshold value to obtain desired relative cell loss ratio among the different classes of traffic. The objective of present work is to seek optimal solution of PBS mechanism for reducing cell loss and hence improving QoS when the threshold is able to adapt based on input traffic. In order to accomplish these objectives, a comprehensive study of various priority-based methods used for controlling cell loss in high speed networks has been carried out. The analysis of PBS scheme revealed that due to fixed threshold in buffer, the cell loss control is effective only for single priority class, irrespective of the input traffic model and its characteristics. To make the buffer adaptive for adjusting relative cell loss ratios according to input traffic conditions, the dynamically controlled threshold method has been designed to be called as ADaptive Partial Buffer Sharing (ADPBS) scheme. In this scheme, the threshold is dynamically varied in run-time based on consecutive cell loss behavior for two priority classes. Its queueing performance is analysed using two different traffic models, namely, Poisson input traffic model and Autoregressive process based video model. Also, different traffic load values, input traffic mix, threshold control parameter combinations and different buffer sizes are used for analysing the performance of the proposed method. All the simulation experiments are conducted using MATLAB 7.0 and the results for dynamic threshold are compared with First In First Out (FIFO) queue and PBS queue having Static (or fixed) threshold (SPBS).

This thesis is divided into six chapters. A brief outline of each chapter is given in the following paragraphs. First chapter introduces the B-ISDN and ATM networks. It gives an overview of the role of traffic control and management functions of the ATM layer. The QoS parameters are discussed in relation to different ATM service classes. The main focus in the presentation is on various priority mechanisms under Usage Parameter Control (UPC) that have been devised so far and are used in this thesis. In second chapter, review of the literature relating to the traffic and network resource management has been done. A detailed survey on algorithms and input traffic models for buffer management under PBS scheme in ATM switch has also been carried out. Third chapter is divided into two parts. In the first part, PBS priority mechanism has been implemented using a fixed-size buffer being serviced by a single server for traffic streams of high priority (real-time) and low priority (non-real-time) cells. A recursive algorithm is implemented in MATLAB to calculate loss probabilities for block of consecutive cells. The effect of different threshold values on cell loss probabilities is also examined. The second part discusses the proposed model with adaptive threshold for PBS queue. The algorithm for adaptive partial buffer sharing scheme is implemented in MATLAB that assumes a finite queue of fixed-size with the initial value of threshold set at 70% of buffer size. The source traffic with two classes of priorities; class 1 with high priority and class 2 with low priority are considered. The threshold in ADPBS scheme is not fixed and is varied dynamically according to the type and burstiness of incoming traffic. The cell loss ratio is controlled by adaptive threshold that depends on a set of control parameters. Chapter four provides the analysis of adaptive threshold controlled partial buffer management scheme in which the threshold is dynamically varied in runtime based on consecutive cell loss behavior for two priority classes. The performance of ADPBS queue is analysed with Cell Loss Ratio as a major QoS parameter using Poisson traffic model as input source to the FIFO, Static threshold PBS and ADPBS queues. In the source model, the inter-arrival time of the cells is distributed exponentially. The source traffic module and the queueing system are simulated using MATLAB. The cell loss data of each queue is captured by a common module and this is used for performance analysis of all the three queues. This chapter is structured in a manner to present results of each traffic condition in a separate section. It is commonly observed for various combinations of threshold control parameters that the ADPBS queue manages to adapt the threshold to allocate sufficient buffer space for the kind of traffic class which forms majority in the incoming traffic; as such, high and low priority cell loss is reduced upto 93% and 28%, respectively, in comparison with other two types of queues. In fifth chapter, the performance of ADPBS queue is analysed with Variable Bit Rate (VBR) based video traffic model for frame sizes of MPEG encoded video sequence based on second order nested autoregressive processes. With the nested autoregressive processes in traffic model, the empirical video sequences can be captured at both small and larger lags. The implementation of complete model along with the three queues is done using MATLAB. The simulations are carried out under different traffic conditions, such as, various combinations of threshold control parameters, varying traffic load, input traffic mix variation and different buffer sizes. For performance analysis, the simulation is carried out by taking 30 samples under each category and the results of simulations are captured, compiled and compared for all three queues. The relative cell loss ratio of high priority cells and low priority cells, when compared for different input traffic mix ratios, is 14 times and 2.2 times higher in SPBS queue than that in ADPBS queue, respectively. The overall consecutive cell loss in ADPBS scheme is 83% less than the consecutive cell loss in SPBS queue for high priority traffic and 54% less for low priority traffic. Finally, Chapter six presents the inferences drawn as a result of the various simulations conducted in this thesis. Also, some pointers to the future research on the topic under consideration in this thesis are discussed briefly in this chapter.