Hop-by-hop QoS Routing using Statistical Distribution-free Approach

Abstract

The advent of Quality-of-Service (QoS) routing has brought a wide range of applications to network users. While precise network state information is critical to QoS provision, maintaining such accuracy is almost impossible. Towards this end, we propose a hop-by-hop QoS mechanism to operate in networks with inaccurate information. The proposed mechanism, namely DF-PI, adopts the distribution-free (nonparametric) approach to construct two-sided prediction intervals. The prediction interval helps infer the future available bandwidth, and is used to generate the proposed QoS metric − statistical available bandwidth δ. “Widestâ€-shortest paths are calculated, by which δ replaces the instantaneous available bandwidth in the traditional widest-shortest routing algorithm (WSR). Relative to WSR, simulation results show that DF-PI achieves satisfactory performance in terms of packet loss, commit ratio, link utilization and average end-to-end delay, together with less update message overhead.