QoS and traffic shaping in ATM networks

Abstract

This research work proposes a generic traffic-scheduling algorithm, deployed within the network, which can be applied to all the six service categories defined by the ATM forum. It can handle both real time and non-real time traffic each defined in terms of a different set of traffic attributes and QoS metrics. The traffic scheduler continuously strives to prioritize real time traffic in the relative order of 'urgency' defined in terms of the difference between the maximum limit committed / alio ted for the connection specific peak-to-peak CDV / buffer and the actual accumulated value in every round of calculation. Thus a delay difference function (DelayDF) is computed for each cell of every connection with real time traffic and cells are scheduled in the ascending order of the value of this variable. Similarly a dummy difference function (DummyDF) of the same order is computed for non-real time traffic as well so that they can also be serviced fairly along with the more demanding real time traffic. The CLR/FLR is utilized for arbitrating between the connections with conflicting (Delay)DF function values and for taking cell/frame discard decisions for over delayed cells, if any. All the exceptions and worst case scenarios that could be thought of have been considered and taken care of in the detailed algorithm. A MS VB6 based simulator was developed for rt-VBR services and the performance of this scaled down basic version of the proposed approach compared with that of benchmark GPS and Burst Index based dynamic multiplexing algorithms. The results showed clear qualitative improvements in traffic shaping performance over the reference mechanisms. The implemented and tested elementary algorithm can also be used for CBR services although testing for the same could not be carried out. The simulation of the detailed algorithm, which includes shaping of non-real time services, is recommended for future work. As an aside a rudimentary buffer management scheme for all ATM service categories is also proposed. The proposed DF algorithm is highly scalable, dynamic & efficient.