The Old Guard Was Showing Its Age
To understand why Segment Routing (SR) was necessary, you have to look at the problems it solved. For years, large networks, especially those run by service providers, relied on a collection of protocols, most notably MPLS. Within MPLS, two protocols did
the heavy lifting for traffic management: LDP and RSVP-TE. Think of LDP (Label Distribution Protocol) as the simple, automatic option. It was easy to turn on but offered little control over the path traffic would take. RSVP-TE (Resource Reservation Protocol - Traffic Engineering), on the other hand, was the polar opposite. It offered granular control, allowing engineers to dictate exact paths for traffic. The catch? It was incredibly complex. RSVP-TE required every router along a path to maintain "state," meaning it had to remember every single traffic-engineered tunnel passing through it. In a large network, this created a massive scalability problem, adding immense operational overhead and making the network brittle.
A Radically Simple Idea: Source Routing
Segment Routing's creators, primarily from Cisco, proposed a different approach based on a classic idea: source routing. The core concept is simple: instead of letting intermediate routers decide where a packet goes next, what if the first router (the "source") decided the entire path from the beginning? It's like the difference between asking for directions at every intersection versus having a full list of turn-by-turn directions from Google Maps before you even leave the house. The source router stitches together a list of "segments"—which are basically instructions—and attaches them to the packet. Each router along the way just has to read the next instruction, execute it, and forward the packet. The intermediate routers don't need to know the full end-to-end path, which drastically reduces the amount of state they have to hold. The intelligence moves from the core of the network to the edge.
Goal 1: Drastic Simplification
The primary design goal was simplification. By eliminating the need for LDP and RSVP-TE, Segment Routing collapsed multiple complex protocols into one streamlined system. Instead of a separate protocol for distributing labels, SR cleverly extends the existing network mapping protocols (like IS-IS and OSPF) to carry the segment information. This means there's no LDP/IGP sync problem and no need for the state-heavy RSVP-TE signaling. The result is a network that is fundamentally easier to manage, troubleshoot, and scale. For operators, this translated directly into lower operational costs and a more reliable, predictable network. This simplicity also made networks more ready for automation and Software-Defined Networking (SDN), as a centralized controller could easily program paths without wrestling with legacy protocols.
Goal 2: Control Without Complexity
While simplicity was paramount, network operators couldn't give up the granular control needed for traffic engineering (TE). The genius of Segment Routing is that it delivers the precise path control of RSVP-TE without the scalability nightmare. Because the ingress router defines the path, an operator or an SDN controller can easily steer traffic to meet specific service level agreements (SLAs), such as sending voice traffic over a low-latency path. This is achieved without any per-flow state in the core network, a massive win for scalability. This stateless traffic engineering also enables advanced features like Topology-Independent Loop-Free Alternate (TI-LFA), providing extremely fast failover (under 50ms) if a link goes down, a critical requirement for modern services.
The Two Flavors: SR-MPLS and SRv6
The architecture was designed to be flexible, leading to two main implementations: SR-MPLS and SRv6. SR-MPLS was designed for a seamless transition. It uses the existing MPLS data plane, meaning network operators could adopt Segment Routing on their current hardware without a forklift upgrade. The "segments" are simply standard MPLS labels. SRv6, on the other hand, is the native IPv6 implementation. It was designed for the future, leveraging the massive address space and extension headers of IPv6 to embed routing instructions directly into the packet. This eliminates MPLS entirely, simplifying the protocol stack even further and enabling a powerful concept called "network programming," where the packet header can carry not just a path, but also functions to be executed at each hop. The dual-design approach allowed for both pragmatic adoption and a forward-looking vision.
