Could storms in Virginia dispel some cloud momentum? I received the last of the responses to my spring survey of enterprises on Friday, and operators had already sent in their responses so the process of data analysis and interpretation can now begin. For subscribers to our journal, Netwatcher, the results will be in the July issue. For now, though, I was struck me by the synchrony between enterprise comments on the cloud and the experiences of Twitter, Amazon, Netflix, and Instagram users during the east-coast power problems.
If you ask senior management at enterprises what the benefits of the cloud are, they cite flexibility in worker empowerment as number one, flexibility in assigning resources to applications based on optimum cost/performance as number two, and availability as number three. Nine out of ten believe the cloud to be more reliable than their own data centers, and yet only 2% reported any massive data center outage in the last year. That’s actually a bit lower than the rate of reported cloud outages. What gives?
It looks to me like there are two clouds out there; the idealized fuzzy cloud of our imagination and the real deployed cloud. You can read about the first in any publication, but you never hear much about the second. For example, you don’t get an answer to the question of why, given the supposed geographic diversity of the cloud, a user like Netflix could be taken down by a power outage in one area. Actually, the cloud isn’t usually that geographically diverse. Many cloud providers serve customers out of a single data center, and most serve customers out of no more than a couple in any market geography.
Cloud operators will admit in private that a small number of large data centers is more economical than a bunch of distributed ones, which we all ought to realize or the whole server consolidation trend would be based on a lie. The facilities are cheaper and can be located in areas where costs are lower, power is similarly likely cheaper, and so is communications access. Thus, even though in many cases cloud data centers are so large that they can’t be powered by backup facilities at all, we are seeing risk concentration along with server concentration.
One reason we may be having these problems is that we lack a systemic vision for how an application is deployed on the cloud. Absent automatic commissioning of even complex applications, there’s a tendency to build non-resilient app models even on resilient infrastructure. Another reason is that the notion of the cloud demand an intimate partnership between IT and network to facilitate optimized distribution and access. That partnership also has to be automated, and the progress toward addressing these issues is very limited at this point. People talk about cloud networking but progress there is elusive.
Maybe because we don’t know what it means, period. We do have a notion of networking’s evolution in SDN but we haven’t linked it to the cloud. This despite the fact that nearly all the enterprises who think they understand cloud networking understand it in SDN terms. The recent Cisco announcement and some competitive and market byplay following it make it clear that there’s still a major debate even on what SDN is, much less how it’s related to the cloud. Even Cisco drew a more casual link than an architectural one, citing VPNs as the basis for both SDN and cloud but going no further.
Could an SDN-centric vision of the network combine with a DevOps-centric vision of the cloud to create the Great IT Movement? Sure, but we need to develop that vision. I suggest that to secure the goal we have to unite the efforts, not pursue them individually then try to harmonize the result. There is only one IT budget, so you can’t drive it in two directions. Unite and prosper. If you’re a vendor in the network or cloud space, that would be a good motto to consider according to my surveys.
