How Carriers Can Evolve to SDN: A Step‑by‑Step Migration Guide

Telecom operators cannot switch to SDN overnight because millions of existing devices and traditional services must continue operating, and the established network management practices cannot be replaced instantly.

SDN deployment in carrier networks must follow an evolutionary route, gradually migrating while the SDN and legacy networks coexist, similar to the IPv4‑IPv6 transition; understanding how to interoperate safely is a key research topic.

Because customers are risk‑averse, a reliable migration plan is essential. It is recommended to start with closed, small‑scale networks such as data centers, IPRANs, or metro networks for pilot trials; once performance and benefits are validated, the operator can consider full‑scale SDN adoption.

The network can be divided into two functional dimensions: intra‑domain path computation (the fabric) and edge access service handling. This article focuses on the evolution of the fabric toward SDN.

SDN‑Enabled Fabric

SDN‑enabled fabric means centralizing the path‑computation function of the intra‑domain switching network. The diagram below illustrates this architecture.

PCE (Path Computation Element) is a typical centralized control solution for intra‑domain switching paths. Although less familiar to IP engineers, PCE has long been used in optical networks. Defined in the ASON (Automatic Switched Optical Network) standard, PCE provides a dedicated server that performs end‑to‑end path computation, leaving other network elements uninvolved.

You might wonder why PCE is needed when optical networks are already centrally controlled. In practice, optical networks are managed via OSS or manual static configuration, while ASON pre‑configures E2E paths and triggers automatic rerouting on failures. Early ASON implementations performed computation on each node, leading to scalability issues due to full‑topology synchronization and E2E calculations. PCE resolves these limits, and multiple PCEs can be cascaded if a single instance lacks sufficient processing power.

The centralized computation concept of PCE aligns well with modern SDN, making PCE a viable approach for migrating MPLS networks to SDN.

A reference architecture for applying PCE to centralized path computation and scheduling is shown below.

Between the SDN controller and routers, the BGP‑LS protocol runs; routers convey topology and bandwidth information collected via IGP to the SDN controller through BGP‑LS. Typically, only one or two routers (often ABRs) per domain need to establish neighbor relationships with the controller.

The SDN controller uses the received BGP‑LS data to perform path computation and then distributes the results to routers via the PCEP protocol.

Centralized path computation avoids inefficient bandwidth usage and reduces system scaling pressure that plague distributed models.