Robin Harris
Hype
Hope
Steven J. Vaughan-Nichols
Best Argument: Hype
Audience Favored: Hope (56%)
The moderater has delivered his final verdict.
Opening Statements
Destined to be another unfulfilled promise
Robin Harris: Of course SDN is hype. Yes, there is a good idea: separating data and control planes, and making the control plane an open software product.
But the bottom line is that while SDN may be good for customers, it isn't in the interest of the large switch vendors. They have a thousand ways to sabotage OpenFlow while at the same time pointing to their "progress" in making SDN a reality.
SDN is a wonderful idea. Many startups will emerge to enable it. But it goes against the switch and router vendor's commercial interests and it will fail.
I saw this same process decade ago with SANS. Once the fog of hype clears, SDN will be another unfulfilled promise.
SDN holds networking's future
Steven J. Vaughan-Nichols: If you're about networking -- and in today's IT world how can you not be? -- then just as surely as IPv6 lies in your future so does Software Defined Networking (SDN).
It's really very simple. As the Open Networking Foundation says, SDN brings direct software programmability to networks. Typically this is done with the OpenFlow protocol, but other protocols can be used.
With this ability, you can use SDN to centrally manage and monitor your network across not just routers and switches from a single vendor but across any networking hardware that implements standardized SDN protocols. For network administrators this enables them to create efficient virtual networks that are independent from physical networks.
This, in turn, means for the CFO that a company's network infrastructure can be used more affordably. Thus, SDN, which now has broad industry support—even Cisco has finally jumped on -board, is a win-win both for the CIO and the CFO.
SDN won't arrive overnight. There's too much legacy equipment. Still, with advantages for both the technicians and the bottom line, it will come.
Talkback
Unless you're talking about the data like layer, it's incoherent.
You do need to transport information via wires or some other physical method. I can't program an isolated computer to talk to another isolated computer merely by writing code. At some point or another, I have to connect the computers somehow to each other.
So you'll always need a physical layer. That can't be moved to software.
Above the physical layer is the data link layer - which is pretty much hard wired into the networking device, be it an Ethernet card, router, Bluetooth device, etc. This is the only layer that is normally hardware that could maybe be moved to software.
But to be honest - I don't see a big need to move the data link layer to software. It's not broken. It's not performing poorly. It does one thing, and it does it well - it provides the lowest level packets that higher levels can use. Anything more complex should really be handled at a higher level.
Above the data link layer - well above that, it's all software anyways. Be it in router firmware or as a part of the OS's TCP/IP stack, it's all software, and it would be quite absurd to NOT call it "software defined networking," because anything above that is in fact defined by software.
So - it's impossible to move the physical layer to software, and anything at or above the network layer is already software, so . . .
. . . so I'm interested in why you think the data link layer needs to be moved to software. Don't disappoint ;).
Data LINK layer
Better name for it . . .
I think the concept is real and solid - but the buzzword is horrible, and a terrible description of what's really going on. Can I plead for ZDNet to quit using buzzwords? Please?
Pretty much all networking is software defined
Television solidified it's 525 line standard in 1941.
I don't get the point of SDN anyways.
Overhyped
When and if you can use separate management products for router/switches then SDN will come of age.
Manageability more than performance
Robin is ENTIRELY right to recall the storage mess of the last *decades*. SDN is shaping up much like that. VMware and Cisco are *clearly* angling to make SDN into "lockin-by-another-name".
Still, I see great promise in SDN. To all those emphasizing that there must be physical reality underneath it all: true, but SDN is NOT just about performance. Equally important, in some contexts, is manageability. For some situations, SDN is the difference between a five-minute deployment, and a five-day one.
Yet another perspective: SDN is a necessary part of SDDC, and Cthulhu knows how much we need *that*.
I get centralized control, but don't get why it needs to be "virtualized."
That actually sounds like a formula for failure: Since the virtual network may not reflect the topology of your physical network, optimizing the virtual network may actually kill performance, as it's not matching the topology of the real, physical network.
Okay, so you've made your dream network by virtualizing everything. Except it's ignoring the topology of your actual, physical network, and thus is actually hugely inefficient.
I can imagine trying to create a virtual ring network on top of a standard hierarchical physical network. A virtual ring network sounds good in theory, but on top of a hierarchical physical network will be an enormous waste, as it has to has to constantly go up and down the hierarchy to visit each node on the way to the destination.
Not such a hot idea IMO. If you want to optimize your network, it should be done at the actual physical level, not at some abstracted virtual level.
Having a centralized, easy to access control panel to control routing for the entire network? Great idea. Layering some other "virtual network" on top of your physical network, so that you can play around with typologies that may not reflect the realities of your physical network? Not such a great idea, IMO.
SDN is just another good hyped Open Source idea