Confusion Reigns!
I'm not sure Mark has this right.
>Within a few years, every data centre worth the name will be supported by a complex, powerful and fully configurable SDN.
I know it isn't the same, but how many organisations have a really detailed and comprehensive QoS policy? Why not? It is too hard to track all the apps on the network and quite frankly, it doesn't matter too much. VoIP gets a boost, but often that's on its own hardware too. Its easier to run a couple of gb cables and then we know how much capacity we'll use.
Then there's the SLAs. The number one thing about service is that you don't mess with operational systems, especially if they are high-value to the organisations. Seriously important stuff often gets its own kit - switches, even firewalls, cabling and servers. This is because software/people which/who do lots of different dynamic things are more likely to foul up, be it through user error or simply code complexity. Switches move finite, (almost) fixed size blocks of data, so they tend to be very reliable, running one software instance for months even years. A web browser running flash with lots of dynamic content, not so much, even on a good day.
SDN brings dynamic reconfiguration across the network. Certainly, the actual reconfiguring of kit doesn't actually take long, so SDN does two things: it cuts out that annoying change control and means you don't need those expensive techies to reconfigure kit. That leaves two questions: (1) is your change control serving a useful purpose? and (2) do you have anyone left who actually understands your network when it all goes pear-shaped? There's actually a third question - how do you troubleshoot something which is that dynamic? Do you log every flow?
There is another major issue with the article's premise: the data centre is not the same as a carrier network. Data centres have short distances and very cheap network capacity upgrades. The carriers' problem is that running cable over the wide area is enormously expensive. That is why we still have copper and not fibre everywhere. Its hard for the telco to recoup costs - not the fibre costs, but the fibre laying costs. The fibre itself is possibly cheaper than copper and offers more service capability.
In a data centre I can over-provision and use SDN to allocate the use. I can run fast backups overnight, I can configure up new databases and give them priority over MS updates for W10. I can provision new web-servers and out-allocate connectivity. Actually, I can do all of this without SDN, but that's another issue.
The carrier problem is that the Melbourne to Sydney fibres are running at capacity. Without digging a 900km trench I can't do much to add new capacity. SDN doesn't fix my capacity problems, it just virtualises my configuration targets. If the network is overcommitted, no amount of reconfiguration is going to fix it. I could add new services, but everyone encrypts everything so unless they really trust me and we do an encrypt/decrypt function to allow "the network" to understand the traffic we're pretty much out of luck. Personally, I'd be running VPNs across the carriers' networks as a matter of policy. I don't need them accidentally treating my voice UDP traffic as torrents, thanks.
Really, I don't want a "smart" network. I may want a high-capacity network, but that is different. I want a network which does what I've asked it to do, reliably. Intelligence leads to complexity, and I've already got 99 problems without that.