In this paper, algorithms for resource allocation in an ATM node that serves heterogeneous traffic sources subject to varying Quality of Service (QoS) requirements are proposed. The node can be either a switch port or a multiplexer. Each connection is first individually treated as logical queue. Quick and efficient algorithms allocating service rate and buffer space to each connection based on traffic characteristics and QoS requirement are developed. In order to improve link and buffer utilization, the aggregate traffic is next replaced by an appropriately parameterized new traffic source that still preserves the key characteristics of the aggregate traffic. The most stringest QoS requirement among all connections is selected to be the QoS target of the new traffic source to assure that QoS of each individual connection is satisfied. Resource allocation for the aggregate traffic is determined based on the traffic parameters and QoS target of the new source. Each individually determined service rate and buffer space can be used in cell transmission scheduling and selective cell discarding. In other words, resource allocation together with two related side problems: cell transmission and cell discarding, are treated in this paper in an integrated and efficient manner. The resource allocation algorithms proposed in this paper can also be used to support Call Admission Control (CAC) in ATM networks.

In this paper, a new traffic shaping mechanism for ATM networks, the RC Shaper, is proposed and studied. The objective of a shaping mechanism is, to smoothen the traffic by reducing the burstiness. Using the analogies between the RC low pass filter and the proposed shaper, performance measures such as burstiness reduction factor and cell mean waiting time of the shaper can be obtained. In this current exploratory stage of study, the shaper is used to shape the well-known Markov Modulated Poisson Process. Behaviour of the shaper is established through examining the burstiness reduction achieved by the device. Another performance measure obtained for the shaper is the cell mean waiting time. Difference in shaping each connection individually and a group of connections collectively is also observed. From our results, it seems better to shape VCs within a virtual path `aggregately' instead of shaping each of these VCs individually. Comparison with fixed-rate shaper is reported and difference in multiplexing shaped and unshaped cell streams is also observed.