Design and Implementation of UCloud's Next‑Generation Backbone Network Using Segment Routing

UCloud’s rapid data‑center expansion created challenges for its MAN and backbone networks, prompting the evolution from backbone 1.0/2.0 to a new generation built on Segment Routing (SR) to achieve an intelligent, reliable, and schedulable infrastructure.

Design goals include automatic path computation by the controller, flexible hybrid access (local dedicated lines + Internet), multi‑dimensional SLA‑aware routing, reduced line‑card overhead, and telemetry‑driven traffic visualization with on‑demand scheduling.

Overall architecture consists of three components: an intelligent controller, backbone edge PE devices, and access‑side CPE/VPE devices. The controller centrally manages resources, collects state via BGP‑LS and Telemetry, stores data in a distributed database, and performs path computation and policy distribution via NETCONF and PCEP.

Control‑plane design features:

Real‑time collection of IGP topology, BGP‑EPE, SR, TE link attributes, and SR‑Policy information.

NETCONF/PCEP for configuration push and path calculation.

Fast fault response, manual fault‑domain isolation, custom path tuning, and per‑flow traffic monitoring.

SR‑TE backbone control plane uses MP‑BGP for L3VPN and BGP‑EVPN for L2VPN, replacing LDP with SR‑allocated public labels. BGP‑EVPN provides integrated L2/L3 services, dual‑homing, and automatic PE discovery.

Forwarding‑plane design includes:

Core layer: PE devices run ISIS‑L2 with SR, advertise Node‑SID/Adj‑SID/Anycast‑SID, and establish MP‑IBGP with RR for VPNv4 routes. SR‑TE tunnels carry L2 traffic.

Edge layer: PE connects to CPE/VPE and M‑Core via EBGP; VPE/VCPE use Internet‑based EBGP with IPsec encryption.

The backbone’s three key characteristics are:

Intelligent : unified controller orchestrates traffic, auto‑calculates paths, and supports per‑destination and per‑flow scheduling based on SLA, delay, bandwidth, and SRLG constraints.

Reliable : carrier‑grade dedicated lines, dual‑PE Anycast‑SID protection, Ti‑LFA fast reroute, SR‑TE primary/backup segment lists, and an Internet‑level backup plane with Flex‑Algo.

Schedulable : five‑tuple flow classification, multi‑type SR‑TE tunnels, color‑based policy routing, and dynamic algorithms for delay, bandwidth, TCO, custom, and public‑tunnel forwarding.

Color resources (default, city, business, local) are used to map traffic to Service‑Class queues, with DSCP or ACL‑based classification feeding the controller’s scheduling engine.

Best‑practice example : a high‑traffic Hong‑Kong–Singapore link is bypassed by steering Guangzhou‑to‑Jakarta traffic through a SR‑TE path Guangzhou→Beijing→Frankfurt→Singapore→Jakarta, demonstrating millisecond‑level convergence, path recomputation, and traffic engineering benefits.

Future evolution plans to introduce Binding‑SID based end‑to‑end SR‑TE tunnels from data‑center tenants, leveraging VXLAN‑to‑MPLS encapsulation and automatic binding‑SID‑driven forwarding.

In summary, UCloud’s next‑generation backbone delivers a smart, reliable, and highly programmable transport layer that reduces line‑card costs, improves utilization, and supports the growing demands of public‑cloud services.