'Good enough' could have been
If you talk to a vendor about SDN or NFV, this isn't at all the bit they want to talk about. Vendors want to talk about how many changes per second that a network fabric can achieve. They want to talk about this mostly because people who care about such things buy switches by the pallet and don't ask too many questions about cost.
These are the same people who, for decades, have told us you need high-end Cisco switches and routers to run a five-man small business. They are the people who will sell you the worst IT deal of all time.
Legacy network vendors are so used to milking a market comprised of endless money and unquestioning obedience they don't even know how to start addressing concerns about usability and interoperability of management applications.
The thing is, if they got their act together on this, it would probably save them. Managers and purchasing who enjoy the swank lunches, boat trips could keep on keeping on, while IT attained 21st century management capabilities and the ability to work with gear from multiple vendors without grief.
They won't, and so grassroots interest is going to make other vendors a lot more attractive.
One winner is Cumulus Networks. I have been playing around with a new Supermicro switch that runs Cumulus. I'm supposed to be getting a nice review together, but I'm honestly running into issues because I can't break the thing.
There's nothing particularly special about the hardware. Supermicro have put together a 4GB Intel Rangely platform, attached it to reasonably decent Broadcom Trident II switching silicon and they sell it for next to nothing.
On my existing SSX-X24S and SSE-X3348T Supermicro switches I can – under very exceptional circumstances – cause iSCSI microbursting issues to occur, although it's actually pretty hard to make it happen. I have not been able to get the Cumulus-enabled Supermicro SSE-X3648S to exhibit this behaviour yet, and I've thrown my entire lab at it.
This concept of "good enough" is critical. The previous generation non-SDN Supermicro switches stand up to every real-world workload I can throw at them. I need to bust out a lab that pulls enough power that I am bringing in additional circuits from other floors to power the thing in order to get those switches to behave even remotely oddly.
Clearly, the previous-generation Supermicro switches qualify as "good enough" for most circumstances. The biggest issue is management. Oh, they have a usable GUI if you want to work on a per-switch basis. They have a CLI that will behave like Cisco IOS if you want to work that way.
Managing a network with a few dozen older Supermicro switches? No problem. Managing a network with thousands of them? Suddenly I have a lot of questions.