Case Study: SDN‑Based Overlay/Underlay Network Architecture and Automation at Alibaba

Peng Junying, an engineer from Alibaba's Technical Support Department, reports on the continuous network challenges faced across Alibaba Group's businesses and the progress made since his previous presentations.

He highlights the persistent issues of network complexity, scalability, and VM migration, emphasizing the need to move from traditional Layer‑2 designs to an overlay architecture using technologies such as VXLAN, NVGRE, and TRILL.

The solution separates the underlay (physical CLOS/BGP fabric) from the overlay (VXLAN‑based virtual networks), with automated provisioning handled by Python scripts and Netconf, eliminating manual CLI configuration.

When a VM is created, the cloud control system notifies an SDN controller via a northbound API; the controller then installs forwarding entries on OpenFlow hybrid switches through packet‑in/packet‑out mechanisms, enabling rapid, automated network deployment.

Key features of the architecture include the use of commercial off‑the‑shelf chips, OpenFlow hybrid switches, API‑driven management, support for multi‑tenant isolation via VXLAN, and integration with NFV, Service Chaining, and PCEP/BGP‑LS for WAN virtualization.

Despite successes—such as improved elasticity, automated fault isolation, and reduced manual intervention—remaining challenges include cost reduction, pure OpenFlow switch capabilities, northbound API standardization, and the need for a full‑stack network architect.

The speaker advocates a "P+V" (Physical + Virtual) framework where routers, switches, x86‑based gateways, and software switches all participate in a unified fabric, supporting both overlay protocols and OpenFlow interfaces.

He concludes that SDN can fundamentally enhance network operability, elasticity, and visibility, and calls for industry collaboration to advance open, standardized SDN architectures.