Let us say you have a fundamental idea that could accelerate your corporate network - how could you test it? Here is an example: one student at Stanford wanted to experiment with load balancing. Load balancers usually direct traffic to the server with the lightest load, regardless of how congested the traffic to that server. What would happen if the balancer also took the traffic into account?
In theory this is a very simple experiment, but would anyone dare try it out on a large working network? Just think of the labour, the potential disruption during set-up and consequent risk that the experiment might unbalance such a complex system and lead to unforeseen consequences - with so many users relying on the service.
In fact, this experiment was actually carried out, first on Stanford's own network and then nationwide on the GENI test network. Professor Nick McKeown, speaking at the 2011 Open Networking Summit put it like this: "For the first time I've seen a graduate student able to take an idea and run it on a national network".
The reason that this was allowable was that Stanford was pioneering a new approach to SDN, based on a standard now named OpenFlow.
OpenFlow, and the separation of the control plane
Professor McKeown is the man behind the creation of OpenFlow - an industry-standard protocol that allows network operators to reprogram a network's control plane from a central interface. Instead of having to go into the physical network and tweak masses of boxes, general instructions can be sent out across the entire network, or subsections of the network - just as the introduction of software made it possible to automatically configure a computer for different purposes, without having to manually alter its structure. This makes the network into "a software-defined network".
Whereas in a normal router or switch the fast packet forwarding (data path) and the high-level routing decisions (control path) happen in the same device, with OpenFlow these two functions are separated: the data path still resides on the switch, while the high-level routing decisions are moved to a separate controller. The open flow switch and controller communicate via the OpenFlow protocol.
OpenFlow began in Stanford University, with version 1.1.0 released in February 2011, and was first demonstrated at InterOpNet Lab in May 2011. Then the ONF was created and now has responsibility for further development of the standard.
OpenFlow-enabled switches are already available from Extreme Networks - the first Ethernet switch vendor to integrate OpenFlow into its entire product portfolio - and other major vendors are increasingly supporting the standard. Incorporating these switches into a network allows the easy deployment of innovative routing and switching protocols, not only to optimise performance, but also to address specific issues such as network flexibility to support virtual machine mobility, high security networking and next generation IP-based mobile networks.
Sign up for Computerworld eNewsletters.