Although actual implementations of software-defined networking (SDN) are uncommon in mid-2014, this may be the next wave of virtualization that organizations aggressively pursue. The SDN approach to building networks, which actually got its start around 1995, makes a lot of sense to IT professionals who are weary from ripping out routers and switches to upgrade every few years.
SDN is also very likely to be a major disruption to the existing networking market. The complex and expensive networking devices we see today will be overtaken by SDN controllers that maintain their rich features on a standardized platform with the routers and switches largely a commodity that will be available from numerous manufacturers.
Those existing devices are often proprietary—written in the manufacturers own internal and typically copyrighted code. The fact that most of these current network devices are fixed-function boxes using application-specific integrated circuits (ASIC) and embedded processors, means that the hardware itself has to be physically changed out in order to achieve upgrades.
The proprietary nature of these devices can also cause the network planner difficulty in product selection and making everything work together. SDN will likely eliminate these issues too.
The effect of SDN on the networking market should therefore lead to far more standardization and lower-cost hardware with software taking the lead role. Manufacturer differentiators will change and the apparent need to stick with the incumbent brand when growing a network or replacing devices should wane.
The SDN software will be centralized and therefore highly controllable, visible, and verifiable over the network. Programming, reconfiguration, routing rules—everything can be managed from this central controller and then migrated out across an entire network. Imagine the simplification over making such changes one switch at a time and maintaining records in spreadsheets.
Consider an example of what networking administrators have to deal with in a large university setting. There can be over 100,000 users, including students, faculty, and staff. Nearly 95% BYOD, they elicit high volumes of data traffic of every imaginable type, from e-mail to HD video, documents, Web browsing, and even experimental application communication having the potential for generating errors. With all of that and the constantly changing network needs of research projects, administrators in this situation likely face more complexity than many global enterprises.
SDN holds out promise that, using a single GUI running on open standards, the administrator in that complex environment can access a central controller with a global view of the entire network. Rather than multiple, vendor-specific devices and protocols, network design and operation are simplified because the administrator is programming the entire network, not configuring it piecemeal.
SDN does this by decoupling the network control and forwarding layers, or stacks, as defined in the Open System Interconnection (OSI) Model that has been a standard in networking for many years. Control is an abstraction that is virtualized in the sense that it is software that performs the functions of networking resources such as switches and routers. The hardware is simply a physical forwarding device. As such, the hardware can be an off-the-shelf platform in the same manner as computer hardware that hosts virtual servers today. The switch therefore becomes a simple forwarding table.
The first standard of SDN, OpenFlow, is finally maturing. This is the initial open communications protocol for SDN that enables the controller to communicate with any brand of forwarding device such as a network switch or router.
The SDN software, which defines the policies and rules of packet forwarding and transmission, is moved off the individual switches and routers. Due to the centralized view, the administrator has more control and has visibility into the network.
As evidence that SDN with OpenFlow is ready for commercial use, Google and Facebook have both adopted them in their own networks. In Google’s case, their own production G-Scale network with custom 10 Gbps switches built on standard chip sets spans the globe.
Startup companies and some existing networking vendors are already working on switches optimized for OpenFlow. So the market shift is impending and many expect the fortunes of the existing manufacturers to change. That has all the vendors talking about SDN. The problem is, many talk about SDN differently and continue to put a proprietary spin on it in order to maintain differentiators and protect those fortunes.
The shakeout in the networking market will be an interesting series of events. In the end, we can expect that network virtualization with SDN will ultimately benefit administrators, IT bottom lines, and users much as other forms of virtualization have been good friends.