Future SDN-Based Network Architectures

Sdn: An Introduction And Some Historical Background

More than a decade has passed, since the advent of what is called, Software-Defined Networking (SDN). Network research gained a burst of activity and innovation in 2008 when SDN was introduced as a term by Stanford University researchers (McKeown et al., 2008) as an attempt to operate networks in a more programmable fashion and to run experiments (such as new protocols, interfaces, or algorithms) on real production networks.

The main premise was that simulations and emulations can only provide insights of whether a proof of concept may be applicable. To actually deploy and test new protocols or architectures on real hardware, the researchers would have to convince hardware manufacturers to adopt their ideas, which they are very understandably reluctant to do so as the design cycle of a new device could take a lot of time and provide limited to no return of investment. The only path left was to develop hardware themselves, using custom-based hardware as described by Lockwood et al. (2007). This custom-based hardware approach was embraced by the networking community.

SDN initially begun with the precept of separating the forwarding plane, the portion of the device that moves packets around, from the control plane, the portion of the device that decides how packets should be moved around, for research purposes. SDN gave the applications to program the forwarding behavior of network devices via a standardized interface. However, based on research and demonstration of SDN-enabled technologies, the industry realized that by utilizing the concepts proposed by the SDN proponents, they could solve real-world problems. For example, in environments such as Data Centers (DCs) where it is important to optimize resources and, thus, the capabilities to customize the forwarding behavior of the network and automating configuration tasks are key factors.

The separation adopted by SDN is achieved by abstracting the forwarding plane and providing an open interface to the control plane. Such a separation incurs many benefits to both planes as it allows research and innovation to occur independently in each plane. Developers could create network applications that directly program the forwarding functionality of network devices irrespective of the device and these applications can be physically located outside of the device itself.

SDN, while at that time a disruptive and empowering technology, was not a new concept. On the contrary, the concept of separating the control plane from the forwarding plane has been present in the networking field for a long time, as documented by Feamster et al. (2013) and Mendonca et al. (2013). As discussed in Feamster et al. (2013) and later in this chapter, the main reasons for SDN adoption was the need for programmability, especially in DCs, where it was more cost-effective to write sophisticated control programs than using proprietary switches that could not support new features without engagement of the equipment vendors. It is important for the sake of historical accuracy and given the context, to take a step back and discuss the precursors to SDN.

The concept of separation to create new services dates even back with ITU’s SS7 (ITU, 1993) networks, where the signaling of telephone calls was separated from the actual phone call to setup and tear down phone calls, thus, enabling new services to be formed such as local number portability and number translations. In the networking domain, ITU’s ATM recommendations (ITU, 1990) were also based on the concept of separating signaling and data path, with signaling being used to set up virtual circuits prior to sending any packets and TINA (as discussed in Lengdell et al. (1996)) where network resources where modeled using the Network Resource Information Model (NRIM) to provide abstractions to service software.

Next came the era of Active and Programmable Networks (A&PN), as surveyed by Tennenhouse et al. (1997) and Campbell et al. (1999), where network programmability was the focus. A&PN was based upon a richer model than programmability and presented two alternatives: in-band and out-of-band controls.

In-band control was the most representative approach of the active networking school of thought from those years, where the concept was that code would be traversing the network alongside the packets to be executed at specific nodes in the network such as Active Node Transfer System (ANTS) discussed by Wetherall et al. (1998). However, it was the out-of-band control (i.e., the programmable networks approach) that was the privileged topic of research and experimentation. The programmable networks concept was to allow software to control how the devices process packets, a concept that would later be the rationale of SDN development.