Mythical, Magical SDN
“The best way to sound relevant is to put ‘dot-com’ after everything you say dot com.”
– Conventional wisdom circa 1995
Last week at Interop you couldn’t swing a stick without hitting a vendor booth emblazoned with “SDN” or “Software Defined Networking.” Clearly the hype cycle is still in ascension phase and buzzword compliance is in full swing.
It’s not letting up, either. Some vendors are leapfrogging their own SDN messaging by now declaring the era of the Software-Defined Datacenter. What happened to Network? Forgive my pragmatism, but I’d like to focus on some details of the networking part, which is yet to come to fruition.
It’s obvious that we at Vyatta are big believers in software networking, as are our 1,000+ customers around the globe. And we can attest to the R&D challenges of creating a feature-rich, scalable and reliable system; it’s hard and takes time. Now add in the separation of control plane and forwarding plane (or “pitcher and catcher” as I think of it) to achieve the distributed functionality required to enable SDN concepts, and it’s no wonder there’s hype and confusion.
To that end, I offer my top 5 candidates of SDN Myths that I’ve encountered in the wild over the past few months:
1. SDN = Switching
The stated objective is “networking,” which runs up and down the OSI stack. Switching solves one type of networking issue; routing another; security and load balancing yet more. Envisioning an SDN architecture that stops at Layer 2 is like putting low-profile sport tires on a bulldozer.
2. It’s All About the Controller
A controller without a forwarding plane is like a pitcher without a catcher. Somewhere, something has to receive the distributed instructions. Critically important point: The forwarding plane needs to be architecturally linked to the remote controller. They need to sprechen.
3. Top of Rack Forwarding Is Good Enough
The root driver of SDN is compute virtualization. Over 50% of the x86 installed base is virtualized, which is atomizing the compute layer. The new Sandy Bridge Intel servers have 24 cores; you can easily assume 10-20 virtual machines per server (Gartner told me they’ve seen densities up to 70 VMs). This means the forwarding challenge goes past TOR… it needs to penetrate all the way into the server.
4. Virtual Switches Are The New Forwarding Plane
vSwitches work at Layer 2. They don’t segment or secure traffic; that happens at Layer 3 and above (See Item #1 above).
5. SDN = OpenFlow (and Vice Versa)
SDN is an architectural concept and construct. OpenFlow is a potentially enabling collection of protocols. Vyatta is pulling for OpenFlow to succeed, but we think it will take time.
I’ll give it a few months and maybe more myths will crop up. Past the myth, however, there is magic and immense benefit to come from SDN. It will evolve in waves over time. And it’s getting very interesting right now…
Entry filed under: Uncategorized.