A Segment Routing Approach to Traffic Engineering In Research and Education Networks

Abstract

Services and applications such as Video-on-Demand, Voice-over-IP, and over-the-top services such as X, YouTube, Facebook, etc. nearly double Internet traffic every other year. A survey conducted among National Research and Education Networks (NRENs) in East and Southern Africa shows that 72% of the NRENs set up their backbones with best effort traffic in mind. However, the same NRENs provide services that have strict bandwidth and latency requirements. Therefore, it is becoming increasingly critical for network operators to efficiently route massive Internet traffic while meeting numerous Service Level Agreements (SLAs) in terms of latency, packet loss, jitter and bandwidth.

Currently, Internet Service Providers (ISPs) use Policy-Based routing (PBR), Interior Gateway Protocol Traffic Engineering (IGP-TE) and Resource Reservation Protocol Traffic Engineering (RSVP-TE) for intra-domain traffic management with the primary goal of effectively utilising the available network bandwidth. However, these traditional methods lack automation and are neither efficient nor scalable. For instance, in IGP-TE, there is a high chance of traffic congestion as the traffic is being routed over the shortest-paths only. Furthermore, PBR and RSVP-TE do not consider all possible paths, and setup times increase as the number of nodes and links grows or in the event of node or link failures.

Considering these shortcomings, this study investigates the applicability of Segment Routing (SR) as a TE technique in response to dynamic user traffic profiles in Research and Education Networks. SR has been increasingly adopted in large commercial ISPs such as Arelion, Bell Canada, among others. It provides a form of source routing, wherein a packet's predetermined path is encoded within the packet header. The study follows an experimental approach where a multi-vendor network model is built in EVE-NG simulation software. The control plane operation is based on a hybrid setup which consists of a combination of distributed and centralised policy implementation. Traffic scenarios are then developed and incorporated into the simulation to enable validation of this proposed approach.

Results show that SR-TE is easier to deploy, more automatable, and scalable compared to IGP-TE and RSVP-TE. It removes the need for label distribution and path signalling protocols, simplifying packet forwarding. Furthermore, SR-TE enhances load balancing by allowing NRENs to implement weighted traffic distribution leading to an optimisation ratio of 2.