Enhancing Traffic Engineering with AI: Comparative Analysis of Mpls, Sd-WaN, and SRv6

Abstract

Modern networks must manage dynamic traffic driven by 5G, IoT, and cloud services. Traditional traffic en- gineering (TE) technologies such as static routing cannot react in real time, leading to congestion and degraded performance. Predictive and adaptive capabilities come through artificial in- telligence (AI) to overcome these shortcomings.
This article compares three classic TE technologies: Segment Routing over IPv6 (SRv6), SoftwareDefined Wide Area Network- ing (SD-WAN), and Multiprotocol Label Switching (MPLS). Each has unique trade-offs: MPLS provides deterministic QoS at a high cost and limited flexibility; SD-WAN provides cost-effective flexibility but does not provide guaranteed QoS; SRv6 makes source routing programmable at the cost of header overhead and scalability demands. To address these drawbacks, we present a TE framework based on AI that leverages predictive analytics for predicting flows and RL to provide adaptive path selection choices. The model was evaluated with simulated enterprise-scale topologies supporting composite traffic mixtures of voice, video, and data. Outcomes demonstrate that AI-driven TE significantly reduces latency and packet loss while improving throughput and cost savings over static TE controls. Predictive rerouting, in particular, achieved double-digit latency savings, while RL dynamically distributed load between MPLS, SD-WAN, and SRv6 paths.
These findings confirm that AI-based TE enhances perfor- mance, scalability, and flexibility and is a suitable solution for future heterogeneous and high-traffic networks.