Towards a Measurement Based Networking approach for Internet QoS improvement

Abstract

The Internet is on the way of becoming the universal communication network, and then needs to provide various services and Quality of Service (QoS) for all kinds of applications. We show in this paper that oscillations that are characteristic of the Internet traffic provoke huge decreases of the QoS that flows can get. After having demonstrated that such oscillations can be characterized by the Long Range Dependence (LRD) function (as well as the Hurst parameter), we propose an approach for improving Internet flows QoS based on smoothing sending rate of applications. TFRC (TCP-Friendly Rate Control) is a congestion control mechanism that has been issued for this purpose. This paper then proposes an evaluation of TFRC benefits on traffic profile and flows QoS. Given these first results, this paper deals with, an evolution of the approach based on TFRC, by promoting a new architecture, called MBN (for Measurement Based Networking), aiming at developing new Internet mechanisms for better managing traffic, QoS or network behaviors. The idea of MBN first relies on the fact that traffic is very different from one link to the other, and that even on the same link, traffic is not stationary, exhibiting very frequent ruptures. Issuing a static protocol, or any kind of network mechanisms being optimal on all links all the times, with the presence of huge ruptures in the traffic, is quite impossible. Therefore, MBN proposes to use monitoring and measurement results in real time, to react to any rupture in the traffic, thus adapting to new traffic characteristics and constraints. This paper, then, illustrates the MBN approach on a case study: the development of a new congestion control mechanism called ‘MBCC’ (for Measurement Based Congestion Control) extending TFRC on the basis of traffic monitoring and measurement results. Some preliminary results, based on NS-2 simulations, are presented. They show the perfect suitability of this new networking approach for improving traffic characteristics and QoS in the Internet, given the complexity, variability and versatility of actual traffic.