Softly Defined Networks

Software defined networks were a technique developed around 6 years ago. The original structure of IP based network scaled by using additional routers that would forward packets based on partial information about the network topology. Inside each router was a dataplane, where the packets themselves flowed through, and a control plane that analyzed the packet headers and made decisions about how to handle each packet. For performance reasons, in all except the lowest-powered routers, these had a large hardware component. In particular, the dataplane was always implemented in hardware, often very complex hardware. However, with the growth of mobile and of virtualization inside data-centers, the need grew to add a new layer to handle the dynamically changing topology as servers and mobile devices came and went, and people wanted better support for technologies such as VPNs. A new standard, openFlow, was defined that made it easy to implement this new layer on top of the traditional IP foundation.

Software Defined Networks (SDN) allowed the control to be distributed in a different manner from the dataplane hardware so that switches and routers could forward packets and network administrators could have central control of the network through a controller without requiring access to the networks physical switches.

Now that Xilinx programmable devices are more like SoCs than traditional FPGAs, they have started to provide full implementations of important functionality. You probably know that networking is Xilinx’s biggest market so this is obviously one area that they have taken an interest in. They call it the softly defined network which they abbreviate to SDNet.

Unlike in a traditional SDN where the dataplane is fixed hardware, the softly defined network uses the programmable fabric to provide wirespeed data-transfer that is completely protocol agnostic. The control plane has all the network intelligence and provides virtual network services, network flexibility and integrated management. The dataplane is now programmable so can be updated on the fly to keep it protocol complexity agnostic.

This gives network operators a lot of flexibility to better manage the economics of the network. Carriers can dynamically provision unique, differentiated services without any interruption to the existing service or the need for hardware re-qualification or truck roll. This provides service providers higher revenue potential with unprecedented CapEx, OpEx, and time to market savings.

• Improved, highly flexible Quality of Service (QoS)
• Flow and session aware capabilities
• Fully programmable hardware data plane and I/O
• Support for network function virtualization (NFV) at wire speed including user defined, custom capabilities
• Scalable line rates from 1G to 400G

The diagram above shows how system architects can unleash the benefits of All Programmable technologies to realize smarter, softly defined networks without requiring a detailed knowledge of the underlying All Programmable device architecture. This implementation flow also allows system architects to focus only on the services they are looking to provision, without having to focus on exactly how those services are being implemented.

More details, including a video introduction, on the Xilinx website here.

More articles by Paul McLellan…