A new year: the time to recap and also to look ahead into what is coming. So how far did we get in 2013?
For me, the biggest achievement of the industry was the recognition of the fact that it is about what you can do with software-defined networking (SDN) and not how it is done or how cheap the switches to do so can be. When you distill it down, it is about lower operational cost through automation; higher agility through orchestration of network configuration with and through other IT systems; and better visibility and control of devices and systems, applications, and traffic flows throughout the network. The initial architectural components — especially the centralized nature of management and control (not necessarily a single centralized controller) as well as an open northbound API to program the network from that central point of management and control — are still common to most solutions that are out there.
The northbound API: another main theme in 2013 that will keep us busy for the coming years. Where is the standard for this? Can or should there be a standard? Can such a standard address the diverse needs that exist out there? Maybe “open” and “highly capable” are more important attributes than a standardized API, which is either utopia when it should address all requirements or insufficient when it lands on the lowest common denominator. A great article about the “Open Awakening” in the industry can be found at GigaOm.
That also means that application developers, and so customers as well as system integrators, need to pick and choose a solution with its northbound API that fits best their needs and requirements today and tomorrow.
As there are so many different use cases in so many different network types — from network virtualization to network functions virtualization (NFV) to traffic engineering, from data center to WAN to campus networking — the industry came up with different architectures, as said before, which still have some key components in common. So how does this evolve over the coming years? Will there be a convergence of architectures, controllers, tools? I’m in doubt about it. I expect that we will continue to see multiple architectures like overlays, strict separation of control and data plane in the physical network, and hybrid solutions even within the same architecture (i.e. for a strict separation with an OpenFlow controller-based architecture).
We will see diversity. Which is good for innovation and to address specific needs, but it leaves the customers with the obligation to choose the solution that fits best. So I highly recommend that customers do ask in the foreseeable future in their SDN RFPs for features and capabilities and not for a specific architecture.
Getting Abstract With SDN
What challenges does the industry need to address beyond this? I highlighted a few during this year already in various blogs and those are still valid. For example: the abstraction model, where I am convinced that the right model to use at the access layer to automatically apply policy is the device or endpoint and its relationship to other endpoints derived through the integration with other IT systems managing those endpoints and their applications.
An abstraction in the core of the network might be the application flow itself, but this requires an intelligent, application-aware data plane that can provide a huge differentiation in the market. If the existing switch hardware is not capable of doing this, then the question becomes how a network topology and aggregated traffic flows can be properly abstracted and how, for example, traffic engineering can be achieved without the need to know each and every little detail of the network infrastructure. Solutions solving this problem have a clear differentiation potential. I expect that we will see a lot of interesting approaches to that problem in the coming years.
With all of the abstraction and overlay discussions going on, we tend to forget that there are still real and mostly finite network resources to work with. In particular, pure overlay solutions without proper integration with the physical “underlay” network will run into resource contention. There is still a need to orchestrate. The examples of working overlay solutions like Skype, VPNs, and so on cannot applied one-to-one to today’s cloud and data-center fabrics, as the bandwidth requirements and growth as well the dynamic traffic patterns can saturate the existing fabrics quickly. I do expect that after the initial hype around overlay networks, the industry will come to that conclusion as soon as late next year.
The race and contention for resource allocation will continue to be a challenge on the northbound side as well. How do the central management and control plane of an SDN solution handle conflicting requests from multiple applications for a limited resource? And how is access control of those applications into the system managed, how are denial of service attacks mitigated, etc.? This is for sure a pretty complex problem space that will keep us busy for the coming years.
Overall, exciting times are in front of us in the SDN industry and as we are entering a year where we can expect more production deployments of SDN. This will allow us to learn more about what needs to be done to address further customer requirements.