SDN is Dead, Long Live SDN – SDNCentral@Churchill Club

SDN is Dead, Long Live SDN - SDNCentral@Churchill Club

Okay, so the title is a little misleading, but not by much. This evening, SDNCentral’s own Matt Palmer was on a panel at the Churchill Club event: Open Forum-Is Software Defined Networking the Next Network Revolution? He was in illustrious company, with Najam Ahmad, (Director, Network Engineering, Facebook), Martin Casado (Chief Architect, Networking, VMware), Erik Ekudden (Sr. Vice President, Head of Technology Strategy, Ericsson), and the moderator, Craig Matsumoto (Managing Editor, Light Reading).

The video recording will soon be available on Churchill Club’s YouTube channel, but in the meantime, here’s a couple of highlights from the event which had Craig posing a series of questions to the panel around SDN, its deployability, benefits and maturity:

  • Key tenets of SDN. The discussion around the key tenets of SDN involved elements of separation of the control plane and opening up the network to programmability, to business benefits of simplification, automation and assurance. Suffice to say, the original definition of SDN has since morphed to include just about everything new and current in networking.
  • What does SDN accomplish? SDN allows us to build networks that don’t suck (Martin’s words), while Najam was clear that the status quo in networking is not longer adequate and that FaceBook is aggressively pushing to use SDN to increase their agility and flexibility.
  • When is SDN deployable? Martin cites numerous examples that SDN is already deployed (depending on your definition), and has been for some time–Rackspace was cited as an example. Erik indicates that some SDN is deployable in carriers today. However, all panelists agree that only early adopters have deployed it so far and that it’s not quite mainstream yet.
  • Is it truly a revolution? Yes and no. Najam believes that it is truly a new way of thinking and looking at the network–contrasting old vertically integrated mainframes from IBM with the open systems and x86 servers that we have today. All panelists agree that the role of the network engineer is changing and that programming/scripting ability will be a critical skill going forward, and Matt pointed out that we will see network applications with user-friendly UIs that dramatically reduce the effort require to manage networks. This will happen just in time to accommodate the new network and IT managers, many of whom expect point and click Web 2.0 UIs, not CLIs. Martin brought up an interesting view (one that I hadn’t yet heard) that he no longer believes it will reduce complexity (as a system), but that we have just moved the complexity to a different place.
  • What do we need to do to get SDN deployed? And what is the role of the systems vendor? Given that early deployments will be custom integration efforts (a point that we at SDNCentral have made in the past), Matt points out that the smart vendors should be going to their customers and offering solutions, not boxes. These vendors will recognize that they have to coinvest with their customers to build new frameworks (custom today) to solve problems.
  • What’s the view on physical versus virtual? The panel sees that the new generation of datacenters will have fast hardware, but intelligence will be in the software. Martin was pushing for essentially fast, static (dumb?) networks in the core, with the intelligence pushed to the edge–which makes sense for VMware and Nicira’s overlay approach, but different from using a southbound protocol like OpenFlow to attempt to orchestrate both physical and virtual switches. As the panel reminded us though, network virtualization and SDN are orthogonal and while SDN can be used to achieve network virtualization, it is not required.
  • View on L2 versus L3 hooks into network? Najam’s view is that L2 is dead and that he only cares about L3 networks. Martin concurred.  In reality, what Najam really wants is the switches to just report visibility into queues, local resource utilization etc up to a central control plane and for that control plane to make all the intelligent decisions. Today, all he gets is SNMP. And as he eloquently put it “SNMP is dead, long live SNMP”.
  • What will SDN be like 5 years out? The panel’s unanimous take was that SDN will become integrated into networking and cloud and will longer be separate. We will stop talking about SDN because it is no longer necessary to talk about it–the applications will hide the SDN underpinnings from us.

Fundamentally, it looks like SDN will succeed when we no longer look at it as SDN–perhaps, we will again just call it networking. And so, our industry’s goal is to get to the point when we can say “SDN is Dead, Long Live SDN!”

Leave a Reply