Software Defined Networking (SDN) is the hot topic in the networking industry these days. There’s no conference or trade show in the datacom and telecom world where SDN isn’t discussed in various flavors. Speakers and panelists from both enterprises and network operators analyze the applicability of SDN to their networks and discuss its benefits across all layers of the protocol stack.
While people agree that SDN has the potential to change our world of communication networks to a – hopefully – more flexible, more automated and more efficient environment, there are quite some different perceptions of how SDN is defined and where and for what it should be used in the network. This reminds me of cloud computing a couple of years ago. When the hype started, everybody discovered that the cloud will have a positive effect on his business and will ultimately fuel the industry – even though every two people talking to each other had a different understanding about its meaning.
Let’s go back to SDN. The good news, people agree on what it means technically on a very high level. The concept of SDN is based on the separation of data and control plane with a central control entity that can be programmed by the user. Just like an open central device where algorithms and applications can be developed autonomously or downloaded from portals. And then there’s OpenFlow – a control protocol for facilitating the communication between the network elements and the central control entity.
Applications for both SDN and OpenFlow are manifold though, and this is where it becomes a bit fuzzier. Let’s start with a view on Google and what they do in this area. Google operates a network called “G-Scale Network”. It’s the company’s internal network interconnecting its twelve data centers around the globe. Google has implemented a software-based OpenFlow networking system that makes it easier to change network routing and switching. The benefits of such architecture: When knowing all available network capacities and volumes of data that need to be exchanged between data center locations, network utilization can be increased to a high value. Google claims to achieve 95% utilization on average by continuously optimizing the routing.
Clearly, such high network load is only achievable when you have a good understanding of the data sources in the network and when you can control those sources to a certain extent – and most important, when your network is small enough like in Google’s case. That’s because the optimization process is an NP-complete problem – as computer scientists would classify it. That is, the time required to arrive at the optimum configuration using any currently known algorithm increases very quickly as the size of the problem – in our case the size of the network – grows. While helping Google to utilize their long haul network most efficiently, this approach does not work for the network infrastructure of a commercial telecom network operator or carrier – too many nodes and thousands of users that cannot be completely controlled.
Carriers generally see SDN to suit their needs in a different way. Selling plain connectivity services to enterprises is tough and differentiation is needed. For network operators and service providers, SDN can open the doors to designing new services that are much more integrated with the enterprise IT landscape. COLT is one of the service providers clearly evolving in this direction. SDN can enable simple bandwidth adjustment by the user and also the automation of much more complex tasks, ultimately leading to better network utilization and higher efficiency for the end user. SDN can therefore help tying the datacom and telecom world much closer together, enabling virtualization of the network and automated communication of demand between both worlds. Just as we see it in any advanced IT landscape for virtualized storage and compute today. SDN and cloud computing therefore complement each other.
Last but not least, SDN is also seen as architecture to control physical network resources in an efficient way by adapting to load conditions and bandwidth profiles. Rather than operating network nodes at full speed all the time, processing power can be reduced and put asleep when not needed. This will help to reduce power consumption making SDN a heavy contributor to a greener IT concept. And there are many more examples of what can be achieved by more software-driven network element architecture. These include increased flexibility and versatility by adapting to the application, better usage of optical spectrum by adapting modulation schemes – often discussed in the context of Software Defined Optical Networking (SDON) – as well as service intelligence by classification of flows, simplified multi-vendor integration and many more.
SDN is omnipresent and promises the revolution of our networks – at least from today’s perspective. While we are currently in the hype phase according to Gartner’s Hype Cycle Research Methodology, we will see the noise around SDN cooling off at some point followed by the trough of disillusionment. We will have learned a lot by then and will soon enter the slope of enlightenment, with people identifying many productive applications of SDN in the datacom and telecom world.