A new Bayesian Model of QoS provisioning in DiffServ over MPLS networks

Natively, Multi Protocol Label Switching (MPLS) allows LSP/Tunnel setup using a round trip model. This Model consists of sending a request message, like the PATH message in the Resource Reservation Protocol(RSVP), and standing for a response message, like the RESV message in RSVP, upon which required resources are reserved. Moreover, crossed routers do not participate in flows classification according to DiffServ Model, since this classification is stringent and is driven by the ingress node at network entry. This paper proposes a new flexible one-way model of the LSP/Tunnel setup in MPLS networks which are based on the mapping of incoming requests on flows admission using a Naive Bayes model and the weighted K-nearest neighbor algorithm. The proposed model is compared to parallel solutions such as the IntServ-over-Diffserv model, the classical model using RSVP-TE and the EEAC-SV approach proposed in [12]. Simulation results show that the new model outperforms other solutions with respect to the request delay bounds and the packet dropping ratio capability, respectively.