Traffic Management of the ABR. Service Category in ATM Networks

Abstract

Data traffic has emerged as a big challenge for the standardization of traffic management mechanisms in ATM networks. In April 1996 the ATM Forum published the first version of the Available Bit Rate Service Category (ABR) to give support to this kind of traffic. ABR was designed with ambitious objectives: high network efficiency, fairness and inter-operability of different ABR switch mechanisms.<br/><br/>The major part of this PhD Thesis has been devoted to ABR. Instead of focusing on one aspect of ABR, the main research topics involved in ABR have been covered, namely: (i) switching mechanisms, (ii) conformance definition, (iii) charging, (iv) ABR support to TCP traffic. In the following the main conclusions are summarized. <br/><br/>Maybe, switch algorithms have been the most investigated topic of ABR. This has happened because the specification of ABR given by the ATM Forum allows a diversity of switch algorithms to be implemented. These range from the simplest binary switches to the more complex ER switches. In the PhD Thesis three of these switch algorithms are analyzed by means of simulation, showing the different degree of performance and complexity that can be achieved. The behavior of ER switches is also analyzed by means of real traces obtained with a commercial ER switch. <br/><br/>The conformance definition is the formalism established to decide whether the source transmits according to the traffic contract. The conformance algorithm standardized for ABR is the Dynamic Generic Cell Rate Algorithm (DGCRA). The PhD Thesis gives a detailed description of the DGCRA. Furthermore, traces obtained by simulation are depicted showing that the algorithm given by the ATM Forum has a decreasing accuracy of the rate conformance with increasing feedback delay. A "UPC based on the CCR" is proposed to solve this drawback. <br/><br/>The parameter dimensioning of the DGCRA is addressed in the PhD Thesis by means of two analytical approaches. Numerical results calculated with the analytical models are also obtained by simulation for validation. The analytical approaches are based on a novel queuing model of the DGCRA. The first analytical approach is based on a renewal assumption of the cell inter-arrival process at the UPC. This approach gives a simple but coarse approximation of the cell rejection probability at the UPC. The second analytical method consists of a Markov chain that accurately describes the stochastic variables involved in the queuing model of the DGCRA. The Markov chain is solved applying the matrix geometric technique. The complexity of this mathematical approach only allows investigating a simple network topology. However, the accuracy of the model allows taking into account the influence of the delay bounds that are negotiated with the DGCRA. This study shows that a major degradation of the cell rejection probability may be obtained if these delay bounds are not properly set. <br/><br/>Another issue investigated in the PhD Thesis is the charging of ABR. Charging may have a decisive impact on the deployment, success and growth of a network. In fact, the research community has paid a great attention to this topic in recent years. Furthermore, pricing may be an essential condition for the users when submitting traffic. Some authors have used this fact to propose congestion control mechanisms based on a dynamic pricing. In such schemes, prices vary according to the demand of network resources by the sources. New prices are conveyed to the sources by means of a feedback mechanism. This charging scheme seems to fit well with ABR, since the RM-cells can be used to dynamically communicate the prices. In the PhD Thesis a dynamic pricing scheme is proposed and an analytical model is used to find out the evolution of the prices. Additionally, several charging schemes are confronted by simulation. This comparison shows that the dynamic pricing gives the best expected charging. <br/><br/>Finally, the support of ABR to the traffic generated with the TCP protocol used in the Internet is investigated by simulation. Currently, the data communications are dominated by the Internet traffic transported by a variety of networks. The deployment of ATM technology has been located in the backbone networks and the end-to-end ATM systems appear remote. In fact, it is not clear whether the universal multi-service network will be built on the Internet rather than the B-ISDN. Simulations performed in the PhD Thesis confront the transport of TCP traffic in different scenarios using ABR and the simpler UBR Service Category. The main conclusion is that ABR can solve the severe fairness problems that can arise using UBR.