Design of new real-time models for tight upper bound approximation of cell loss ratio in ATM networks

ATM as a high-speed cell switching technology can support multiple classes of traffic with different quality of service (QoS) requirements and diverse traffic characteristics. A main QoS requirement is the cell loss ratio (CLR). We need a real-time expression for the CLR calculation in ATM networks where the statistical multiplexing is an important factor. The existing analytical methods for the CLR estimation are mostly based on fluid-flow and stationary approximate models. In this paper, we first evaluate these methods against the results obtained through simulation. The simulation is done at the cell level that provides very accurate results with buffer size as a variant. It is shown that the CLR estimation based on existing analytical models are widely overestimated. We have, then, proposed three new approaches that yield significant improvement in the accuracy of the CLR approximation. First, we have found global correction coefficients to compensate for the error of the current analytical methods. Second, we have proposed a new upper bound based on exact modeling of system behavior in the finite buffer case. This is a novel approach that combines fluid-flow and stationary approximate models and outperforms all the previous ones. The accuracy of the proposed model is verified by simulation. Third, we have found a tight piece-wise linear approximation that can be calculated in real-time. We have studied application of these bounds in non-homogeneous as well as homogeneous cases. (C) 2002 Elsevier Science B.V. All rights reserved.