SDN Architecture: Control and Data Planes
Software-defined networking is the formal architecture behind controller-based networks. It defines clean layers, names the interfaces between them (northbound and southbound APIs), and introduces the modern data-centre design — a spine-leaf underlay with an overlay fabric built on VXLAN.
What is SDN?
Software-defined networking (SDN) is an architecture that separates the network’s decision-making (control plane) from its forwarding (data plane), and exposes the decision-making through programmable interfaces. The word “software-defined” means the behaviour of the network is defined by software running in a controller, not hard-wired into each device’s firmware.
SDN grew out of research (notably OpenFlow, from Stanford and the ONF) that wanted to make networks as easy to program as computers. The essential move is the same one from Lesson 2: centralise control, distribute forwarding. In a fully SDN architecture, the controller is the only place intelligence lives; devices are programmable forwarders.
Three ideas carry the whole architecture, and you will meet all three in every diagram this week: the planes (who decides, who forwards, who manages), the APIs that connect them (northbound and southbound), and the fabric (underlay plus overlay) that carries the traffic.
Control, data, and management planes
SDN formalises the three planes we touched on earlier. Getting these straight is the most commonly tested part of this lesson.
| Plane | Job | Examples |
|---|---|---|
| Management | Configure, monitor, and manage the device | SSH, SNMP, syslog, NETCONF/RESTCONF |
| Control | Decide where traffic goes; build forwarding tables | OSPF, EIGRP, BGP, STP, SDN controller |
| Data (forwarding) | Move the actual user packets/frames | ASIC forwarding, switching, routing at line rate |
The management plane is where your automation tools live. When Ansible or a Python script SSHes into a switch, or when a REST call configures a controller, that is management-plane traffic. SDN does not remove the management plane; it makes it programmable.
Northbound and southbound APIs
An API (application programming interface) is a defined way for one piece of software to ask another to do something. SDN controllers expose two kinds, named for their position relative to the controller.
- Northbound API — sits above the controller. Applications, scripts, and orchestration tools use it to tell the controller what they want. It is typically a REST API exchanging JSON over HTTPS, which is why REST and JSON dominate the rest of this week.
- Southbound API — sits below the controller. The controller uses it to program the devices with the forwarding behaviour they must implement. Examples include OpenFlow, NETCONF, RESTCONF, and vendor-specific protocols such as Cisco’s southbound interfaces.
Underlay, overlay, and fabric
Modern design separates the physical network from the logical network it carries. Three terms describe that separation, and the exam expects you to keep them apart.
- Underlay — the physical, IP-routed network: the switches, links, and routing (usually a simple, fast protocol such as OSPF or BGP) that provide basic IP reachability between every device. The underlay is the road system.
- Overlay — a logical network built on top of the underlay. It tunnels traffic from one edge device to another, hiding the physical path in between. The overlay is the delivery service that drives on the roads.
- Fabric — the combined system of underlay plus overlay, managed as one programmable unit. A “fabric” is what the controller provisions.
The point of an overlay is freedom from the physical topology. A host in one rack can be in the same logical Layer 2 segment as a host ten rows away, even though the traffic between them is routed across the underlay. That is what makes workload mobility and segmentation possible in a data centre.
Cisco implements this idea as SD-Access (software-defined access) for the campus and SD-WAN (software-defined WAN) for the wide area. In SD-Access, the fabric is built from underlay switches, routed with a control-plane protocol (LISP), and carries an overlay tunneled with VXLAN; the controller defines who can talk to whom as a policy rather than as box-by-box VLAN and ACL configuration. SD-WAN applies the same controller-driven, overlay-based approach to branch and WAN links.
Spine-leaf underlay
The standard data-centre topology for a fabric is spine-leaf (also called a Clos network). It replaces the old three-tier core/distribution/access design.
- Leaf switches connect to servers and other endpoints. Every leaf is the boundary of the fabric for its attached devices.
- Spine switches connect only to leaves. They do not connect to servers.
- Every leaf connects to every spine, giving a predictable, fixed number of hops (usually two) between any two leaves.
| Property | Three-tier (core/dist/access) | Spine-leaf |
|---|---|---|
| Hops server to server | Variable | Predictable (leaf-spine-leaf) |
| Adding capacity | Add bigger core/distribution boxes | Add a leaf or a spine horizontally |
| East-west traffic | Can bottleneck at core | Spread evenly across spines |
| Design | Hierarchical, oversubscribed | Flat, low-latency, scalable |
VXLAN at a high level
VXLAN (Virtual Extensible LAN, RFC 7348) is the tunnelling technology most fabrics use to build the overlay. The problem it solves is a limit inherited from VLANs: a VLAN tag is 12 bits, so only about 4,000 VLANs are possible, and VLANs do not naturally stretch across a routed data centre. VXLAN removes both limits.
| Feature | VLAN (802.1Q) | VXLAN |
|---|---|---|
| Identifier size | 12-bit VLAN ID | 24-bit VNI (VXLAN Network Identifier) |
| Segments possible | ~4,096 | ~16 million |
| Scope | Layer 2 domain, bridged | Over a Layer 3 IP network (overlay) |
| Encapsulation | Adds a 4-byte 802.1Q tag | Adds a UDP/IP header + VXLAN header |
VXLAN uses VTEPs (VXLAN Tunnel Endpoints) — typically the leaf switches — to encapsulate a frame, send it across the IP underlay, and decapsulate it at the far end. The original frame is unchanged; the underlay only ever sees ordinary IP packets, which is exactly what lets the underlay stay simple and fast.
Review
Check what you have learned with these review questions.
| Key term | Meaning |
|---|---|
| SDN | Software-defined networking; centralised control plane with programmable devices over APIs. |
| Northbound API | Interface between applications and the controller, often REST/JSON. |
| Southbound API | Interface between the controller and devices (OpenFlow, NETCONF, RESTCONF). |
| Underlay | The physical IP-routed network providing reachability between devices. |
| Overlay | A logical, tunneled network built on top of the underlay. |
| Fabric | The combined underlay and overlay managed as one programmable system. |
| Spine-leaf | A flat data-centre topology where every leaf connects to every spine. |
| VXLAN | Tunnelling protocol that encapsulates L2 frames in UDP/IP using a 24-bit VNI. |
| VTEP | VXLAN Tunnel Endpoint; the device that encapsulates/decapsulates VXLAN traffic. |
- Cisco. CCNA (200-301) Exam Topics, Domain 6.0 Automation and Programmability. cisco.com.
- Mahalingam, M. et al. RFC 7348: Virtual eXtensible Local Area Network (VXLAN) (2014). rfc-editor.org/rfc/rfc7348.
- Open Networking Foundation. Software-Defined Networking: The New Norm for Networks. opennetworking.org.
- IEEE. IEEE 802.1Q: Bridges and Bridged Networks (VLAN tagging). standards.ieee.org.
Sources & references
- Cisco. CCNA (200-301) Exam Topics, Domain 6.0 Automation and Programmability. cisco.com.
- Mahalingam, M. et al. RFC 7348: Virtual eXtensible Local Area Network (VXLAN) (2014). rfc-editor.org/rfc/rfc7348.
- Open Networking Foundation. Software-Defined Networking: The New Norm for Networks. opennetworking.org.
- IEEE. IEEE 802.1Q: Bridges and Bridged Networks (VLAN tagging). standards.ieee.org.
Target exams: CCNA 200-301 v1.1 · CyberOps Associate 200-201. Last content verification: 2026-10-11.