La ROUTOURNE va tourner

TL;DR

This paper addresses the SR constraint that limits the number of segments ($PMS$) and the loss of isotonicity that hampers traditional route calculations. It introduces ROUTOURNE, a framework that interleaves on-the-fly encoding of distances with an extended-dominance comparison to directly produce deployable segment lists during path search. The work provides formal proofs of correctness and efficiency (in the bramas2023simple report) and demonstrates through experiments that ROUTOURNE achieves deployable solutions with at most linear overhead and outperforms graph-transformation approaches, enabling practical, load-balanced, source-based traffic engineering. The method is presented as a generic augmentation to path computation algorithms, with public code and experiments to facilitate adoption in real networks.

Abstract

Segment routing (SR) offers precise control over the paths taken: it specifies a list of detours, called segments, in IP packets. However, the number of detours that can be specified is limited by the hardware. When calculating segment lists, it is therefore necessary to limit their size. Although solutions have been proposed for calculating these lists, they lack generality and are not always optimal or efficient. We present ROUTOURNE, a method for diverting routing algorithms so that they calculate, not simply an optimal physical path to be translated into a list of segments a posteriori (with no guarantee of its size), but directly the optimal lists of segments deployable by the underlying hardware. ROUTOURNE thus facilitates the deployment of advanced traffic engineering strategies and policies, notably for load balancing from sources. Despite a route fraught with surprising challenges - in particular, the loss of isotonicity induced by SR - ROUTOURNE proves efficient, inducing at worst a linear overhead. Its accuracy and optimality have been proven, and its effectiveness evaluated by generalizing it to several more or less complex path calculation algorithms.

Figures (3)

• Figure 1: Segments nécessaires pour encoder les distances optimales vers 3 et D. ROUTOURNE résout le défi de l'isotonie perdue.
• Figure 2: Évaluation sur topologies réalistes de 2000, 1000, 370 et 300 arêtes respectivement.
• Figure 3: Évaluation du temps de calcul sur des graphes synthétiques et denses.