A Bayesian k-NN based scheme for QoS provisioning in IP/MPLS networks

The Quality of Service (QoS) provisioning paradigm has become highly complicated due to the rising of networks complexity, heterogeneity and unpredictability. This complexity resulted in the need to satisfy stringent user´s QoS requirements which depends generally on network capabilities. To deal with this complexity, an autonomic management of the global network seems to be mandatory. Indeed, autonomic network management can offer a new way to master end-to-end performances by providing self-* properties to networks. Self-* properties include self-optimization, self-organization, self-protection and many others. This paper proposes an autonomic and flexible end-to-end QoS provisioning scheme which is based on the per-hop classes of service discovery and its mapping to user´s QoS requirements basing on the k-nearest neighbor algorithm (k-NN). Simulation results show that the proposed scheme significantly outperforms performances of parallel solutions such as the IntServ-over-Diffserv model and the EEAC-SV approach proposed in [14] with respect to the request delay bounds and the packet dropping ratio capability, respectively.