Introduction to SDN Architecture

The purpose of any network is to enable data communication between host endpoints via network protocols. The network operational model can be described using the concept of planes, where each physical network device has a data plane , control plane and management plane. It is a functional model that describes the dynamics of data communication and networking services for traditional or controller-based network architecture. There are differences however between traditional and newer controller-based architecture, that is evident with the location of operational planes.

Data Plane

The data plane is only responsible for forwarding of endpoint data traffic between network interfaces. All data plane traffic is in-transit between neighbors, and not associated with communication protocols. It is not handled by the processor as a result. For example, routing tables created by the control plane are used by the data plane to select a route. The packet is then forwarded to a next hop neighbor.

  • MAC learning and aging
  • MAC address table lookup
  • Routing table lookup
  • ARP table lookup
  • MAC frame rewrite

Figure 1  Traditional Network Architecture

Similarly, MAC address and ARP tables created by the control plane, are used by the data plane to forward traffic. While all three planes exist on all network devices, the services provided are based on the device class. For example, only routers and L3 switches support routing tables, ARP tables and frame rewrite. Conversely, all switches create MAC address tables while routers do not. 

Control Plane

The control plane is responsible for building network tables used by the data plane to make forwarding decisions. Control plane protocols only communicate with directly connected neighbors. It is only the processor that handles inbound and outbound control plane traffic. There are routing protocols that build routing tables from neighbor advertised routes for Layer 3 connectivity. Some common examples of Layer 3 control plane protocols include OSPF, EIGRP, BGP, and ICMP.   

  • Network tables
  • Path selection
  • Frame switching
  • Link negotiation
  • Error messages

Control plane protocols also enable interconnection of switches within Layer 2 domains. For example, STP enables a loop free topology between multiple switches. There is dynamic trunk negotiation between neighbor switches and EtherChannels. Examples of Layer 2 control plane protocols include STP, DTP, LACP, and CDP. Network switches create MAC address tables for frame switching within Layer 2 domains.

Management Plane

The management plane is responsible for configuration and monitoring of network devices. There are various application protocols that are used to manage the network. For example, SSH is initiated to the management plane of a router to configure network interfaces. SNMP sends traps to a network management station to alert on operational status of interfaces.

  • Configuration
  • Monitoring
  • Automation
  • Programmability

There are newer protocols available such as NETCONF that enable automation of management functions. Similar to the control plane, is the fact that all management plane protocols must be handled by the processor. Some other examples of management plane protocols include TFTP, Telnet, RESTCONF, Syslog, NTP, DNS, and DHCP.

    1. The management plane initiates a session with the local router to configure OSPF and enable network interfaces.
    2. The control plane has a routing table with a route that includes a next hop address and local exit interface.
    3. The data plane does a routing table lookup for the next hop address associated with a destination subnet. The data plane then forwards all packets to neighbor with next hop address.

Software-Defined Networking (SDN)

Software Defined Networking (SDN) is an architecture that separates the control plane from the data plane. Cisco IOS software is moved to an SDN controller. That decouples the control plane from hardware and enables programmability of all network devices. The controller communicates via agents installed on devices. The same functions are provided as with traditional networking architecture for each operational plane. Figure 2 illustrates how the management plane is also moved to the controller.

Figure 2  SDN Operational Planes

It is similar to a hypervisor layer that abstracts (separates) server hardware from application software. There is a software-based centralized control plane with a physical underlay transport. SDN and fabric overlays enable programmability of devices from a centralized policy engine.

  • SDN decouples the control and data plane.
  • Control plane is software-based and not a hardware module.
  • SDN controller is a centralized control plane with a policy engine.
  • Network infrastructure is an underlay for programmable fabric.

Figure 3  SDN Architecture Layers

SDN Components

The SDN controller provides centralized management where the network appears as a single logical switch. Network services are dynamically configurable when the control plane is moved from physical infrastructure to a software-based SDN controller with API modules. The northbound and southbound APIs enables communication between applications and network devices.

Table 1  SDN Components

SDN controllers communicate with physical and virtual network devices via southbound APIs. Conversely, communication from controller to SDN applications is via northbound APIs. There is a policy engine configured on a controller for orchestration and automation of network services.

  • Programmability – network is directly programmable because it is decoupled from infrastructure and data plane forwarding.
  • Agility – abstracting control plane from data plane enables dynamic configuration to modify traffic flows as network conditions change.
  • Centralized Management – network intelligence is centralized in software-based SDN controllers. The global network appears to applications and policy engines as a single logical switch.
  • Automation – dynamic configuration (provisioning) of network devices and software upgrades is based on APIs.  

Network Functions Virtualization (NFV) increase agility by decoupling network services from proprietary hardware and moving it to software modules on SDN controllers. That makes it easier to provision, automate, and orchestrate network services such as DNS, firewall inspection and network address translation.

Advantages of SDN / Programmability

SDN architecture and programmability has shifted network management from traditional static to a dynamic model. Cisco devices support programmable ASICs to enable programmable services. Open APIs translate between application and hardware to initialize, manage and change network behavior dynamically.

There are a variety of advantages such as on-demand bandwidth, dynamic security and elastic capacity. In addition, there is rapid cost effective deployment of applications and services. The provisioning of wired and wireless services requires automated turn-up of network services, push configuration, automatic monitoring and real-time analysis.

Fabric Underlay

Cisco has recently developed SD-Access fabric architecture for data center and enterprise connectivity. The purpose is to enable automation, programmability and mobility for physical and virtual platforms. It is comprised of an underlay, fabric overlays and Cisco DNA Center.

The fabric is comprised of a physical underlay designed for high-speed transport. It is characterized by network devices, topology and protocols for communication. There is a common underlay that provides transport for overlay traffic. That would include control plane protocols such as STP, DTP, OSPF, EIGRP and ARP.

  • Network infrastructure used for transport of all data traffic
  • Comprised of network devices, protocols and configuration
  • Network devices must support programmability with agents
  • Physical underlay operation is independent of overlays

Fabric Overlay

There is also path virtualization enabled with fabric overlays that are built on top of (or over) the underlay. Overlays create a virtual topology across a physical underlay infrastructure with encapsulation techniques that create tunnels. That essentially enables route and address isolation, that is independent of underlay and other overlays. Encapsulation is nothing more than adding outer header/s to original payload that is not visible to network devices when in-transit.

  • Virtual topology with interconnects between nodes
  • Encapsulation (tunnel) creates path virtualization
  • Network address overlap and route isolation enabled
  • Overlays are operationally independent of underlays

Consider that overlays logically create single point-to-point connections. That same topology has multiple physical connections between switches. The purpose of overlays are to solve limitations inherent with physical switching domains such as STP, routing loops, broadcasts and address overlap. They also enable multi-tenant service, enhanced mobility, seamless connectivity and automation. 

Table 2  Underlay vs Overlay

Layer 2 Overlay

Within the fabric architecture there is support for Layer 2 and Layer 3 overlays. Layer 2 overlays are designed to emulate a physical topology for the purpose of extending Layer 2 domains. For example, connecting two servers on different switches that are assigned to the same VLAN. The solution is a VXLAN overlay to enable a virtual connection between servers. It is common to have web-based applications with multiple servers that are often in different locations.

  • Emulates a physical switching topology with virtual overlay
  • Extend Layer 2 domains between switches and locations
  • Enable address isolation and overlapping between domains
  • Tunnels terminate at leaf switches for campus deployment

Figure 4  VXLAN Fabric Overlay

VXLAN is a data plane overlay that encapsulates host packets for communication across fabric. As an overlay, it requires the transport services of a physical underlay infrastructure. In our example, the tunnels are terminated at fabric edge switches. There is a common underlay for data plane forwarding, however the underlay topology is independent of overlay topologies. As a result, underlay and overlay maintain separate data and control planes.

Layer 3 Overlay

Layer 3 overlays enable data plane forwarding across a fabric between different subnets. There is the advantage as well of isolation from the underlay limitations associated with MAC flooding and spanning tree protocol loops. Tunnels are created with encapsulation of host packets. Some examples include VPN, MPLS, GRE, CAPWAP and VRF.

  • Routing-based overlay for IP connectivity across fabric
  • Isolates broadcast domains to each network device
  • IP tunnel terminates at host endpoint or network device
  • Logical point-to-point topology between tunnel endpoints

Figure 5  GRE and CAPWAP Overlays