It’s recapitulation Friday, and there were some items of interest that were crowded out of my blogspace during the week. One deals with the important topic of SDN and OpenFlow. Extreme became the new kid on the OpenFlow block, which at one level is a good thing. The reason it’s not an unqualified good thing is that the whole announcement reveals that we’re still looking at SDNs and OpenFlow from the wrong perspective. Vendor positioning and vapid coverage are perpetuating a myth that can only hurt SDN credibility in the long run.
What Extreme is doing with OpenFlow is about as much as any vendor can do, which is to support the standard. They’re also enhancing their exposure of network APIs, which they have been doing for quite a while under the covers. Extreme was our partner in a test of ExperiaSphere control over the network several years ago, in fact. Their assertion that they have some special sauce in terms of SDN is at least historically credible.
The problem is that if you have goals for SDN, for OpenFlow, then the present processes really aren’t going to meet them. Debates over whether a vendor’s architecture is better for OpenFlow or not disguise the broader question of whether OpenFlow is enough to create an SDN, and if not where the rest is going to come from. In fact, Cisco’s ONE, which doesn’t even include explicit support for OpenFlow, is a more credible SDN strategy than the “standard” ones are.
SDN success comes from uniting two conceptual layers, ones that in prior writings I’ve called the “Cloudifier” and the “Topologizer”. The first of these owns the collection and abstraction of application connectivity needs so that they can be expressed in a systematic and consistent way. The second owns the abstraction and dissemination of network information, from topology to status and capability. The central intelligence, my “SDN Central” element, that many SDN definitions hypothesize is responsible for uniting these layers. Only we don’t have any firm conception of the layers, nor do we have any for the central intelligence. No, the OpenFlow controller isn’t that, it’s something that simply commands individual switches.
You can’t create a connection from individual forwarding changes; only a consistent strategy for path-building that’s implemented as a set of single-switch rules creates a connection. Where does that strategy live? In the Cloudifier, the Topologizer, and that mystery intelligence in the middle. Where are they defined? Nowhere, not even for Cisco’s ONE in any organized way. Cisco actually finesses the issue by drawing on a distributed vision of SDN, which is contrary to many definitions, but unless somebody offers a centralized vision Cisco’s approach is the only game in town.
My spring survey results for enterprises are already showing that among the most cloud-literate of the firms, the role of SDNs as the mechanism for creating “cloud networks” is becoming clear. Since the cloud is the principle IT driver of our time, the network vendors either have to link to it or be buried under the trend. Cisco’s ONE also puts pressure on, and as a result we’ll likely see a number of vendor-proprietary visions of SDN…before we ever see a standardized one.
I would love to see a standard framework for SDNs with central application control. I think the chances of that happening are about the same as Angela Merkel’s suddenly announcing the collectivization of all EU debt under a German guarantee. The standards process today is broken; only something harmless can advance. What we can then hope for is at least reasonable dialog on the issues, and we’re not having that now. There is no SDN without central intelligence linking application needs to network capabilities. There is no framework to provide that. Can we talk about the issues? If you think your company can make a contribution there, then email me and I’ll write about you. No smoke, and I’m past the age of fascination with mirrors.
Speaking of smoke and mirrors, we have Facebook and social advertising. Nobody needs to be reminded that Facebook’s IPO was highly disappointing, and as more Street researchers cover the stock things aren’t looking much better. Another report just out says that a third of social-network advertisers don’t know if they get any return on their ad investment because they don’t know if it influences buying decisions.
This whole mess is something like an “Emperor’s New Clothes” scandal that’s complicated by the fact that even though the clothing is only virtual we’re still not only wearing it but buying and selling it. Further, since there’s a finite limit to what the total adspend of the globe could be (and it’s proportional to GDP) we also have to account for virtual shortages of raw material. The root problem is that nobody wants to step on the explosive growth of the Internet by insisting on anything as pedestrian as pay-for-what-you-get. Ad sponsorship is “free” in the minds of users, and to a point that’s right. The question Facebook has presented us with is whether we’re approaching that point.
There’s ad reality. There’s SDN reality. Whatever reality there is, you eventually have to face it. RIM taught the lesson this week; can everyone else learn from that, or will they have to be burned to appreciate fire?