Two different approaches for providing QoS in the Internet backbone

This paper compares two different approaches for providing QoS in the Internet backbone. The first, denoted as simple approach, is a simple method of Admission Control performed by a weighted-fair-queuing scheduler. This approach guarantees QoS with a suitable overprovisioning of transmission resources. The second approach, denoted as basic statistical approach, is a full-featured statistical framework, based on the central limit approximation, for estimating the delay bound violation probability associated to each traffic class, given the offered traffic and the schedulers' configuration. Both approaches adopt the dual-leaky-bucket (p, r, b) regulated traffic model in order to characterize the aggregate traffic flows. At first, we analyze the characteristics of both approaches in order to point out the respective pros and cons. Then, we propose an enhanced version of the basic statistical approach, the enhanced statistical approach. Our novel approach, as opposed to the basic statistical approach, performs a joint optimization of the regulator's parameters and of the network capacity instead of a separate optimization. Our study shows that the enhanced statistical approach outperforms both the basic statistical approach and the simple approach and offers the highest potential for a better utilization of transmission resources. The basic and the enhanced statistical approaches assume that traffic is Gaussian. The simple approach has looser constraints on the statistical behavior of traffic, but it requires overprovisioning of transmission resources to guarantee the required QoS in terms of packet delay and loss.