Preprocess your Paths -- Speeding up Linear Programming-based Optimization for Segment Routing Traffic Engineering

TL;DR

The paper addresses the scalability of LP-based Segment Routing Traffic Engineering (TE) by surveying preprocessing approaches that prune SR paths before optimization, and by conducting a large-scale comparative study on public Repetita topologies and a Tier-1 ISP backbone. It shows that individual preprocessing techniques offer substantial speedups at the cost of varying amounts of solution quality loss, and that combining SB, DP, and SR path domination yields robust, order-of-magnitude reductions in compute time with minimal impact on MEAN LOAD UTILIZATION $MLU$ or optimality. A key contribution is demonstrating that a carefully tuned combination can reduce LP solve times by around $10\times$ (and up to $\sim 10\times$ with outliers) while maintaining near-optimal TE configurations, thereby making LP-based SR TE practical for large networks. The work also documents dataset-dependent performance, the importance of real-world data for evaluation, and provides guidance on parameter choices to balance speed and quality in operational settings.

Abstract

Many state-of-the-art Segment Routing (SR) Traffic Engineering (TE) algorithms rely on Linear Program (LP)-based optimization. However, the poor scalability of the latter and the resulting high computation times impose severe restrictions on the practical usability of such approaches for many use cases. To tackle this problem, a variety of preprocessing approaches have been proposed that aim to reduce computational complexity by preemtively limiting the number of SR paths to consider during optimization. In this paper, we provide the first extensive literature review of existing preprocessing approaches for SR. Based on this, we conduct a large scale comparative study using various real-world topologies, including recent data from a Tier-1 Internet Service Provider (ISP) backbone. Based on the insights obtained from this evaluation, we finally propose a combination of multiple preprocessing approaches and show that this can reliably reduce computation times by around a factor of 10 or more, without resulting in relevant deterioration of the solution quality. This is a major improvement over the current state-of-the-art and facilitates the reliable usability of LP-based optimization for large segment-routed networks.

Figures (6)

• Figure 1: MLU deterioration for different preprocessing approaches. (A few very large outliers were cut off for better readability.)
• Figure 2: Achievable speedup for the 2SR optimization. (A few very large outliers were cut off for better readability.)
• Figure 3: ECDF of the relative demand sizes (w.r.t to the total traffic volume) of the ISP 2021 and the Repetita Cogentco instances.
• Figure 4: 2SR speedup achieved by the SR path domination preprocessing on the two evaluation datasets.
• Figure 5: MLU deterioration resulting from the combined preprocessing approach for different datasets and SR algorithms.