A Prediction-Based Model for Consistent Adaptive Routing in Back-Bone Networks at Extreme Situations

Abstract

To reduce congestion, numerous routing solutions have been proposed for backbone networks, but how to select paths that stay consistently optimal for a long time in extremely congested situations, avoiding the unnecessary path reroutings, has not yet been investigated much. To solve that issue, a model that can measure the consistency of path latency difference is needed. In this paper, we make a humble step towards a consistent differential path latency model and by predicting base on that model, a metric Path Swap Indicator (PSI) is proposed. By learning the history latency of all optional paths, PSI is able to predict the onset of an obvious and steady channel deterioration and make the decision to switch paths. The effect of PSI is evaluated from the following aspects: (1) the consistency of the path selected, by measuring the time interval between PSI changes; (2) the accuracy of the channel congestion situation prediction; and (3) the improvement of the congestion situation. Experiments were carried out on a testbed using real-life Abilene traffic datasets collected at different times and locations. Results show that the proposed PSI can stay consistent for over 1000 s on average, and more than 3000 s at the longest in our experiment, while at the same time achieving a congestion situation improvement of more than 300% on average, and more than 200% at the least. It is evident that the proposed PSI metric is able to provide a consistent channel congestion prediction with satisfiable channel improvement at the same time. The results also demonstrate how different parameter values impact the result, both in terms of prediction consistency and the congestion improvement.