Q&A: Charles Babcock on cloud computing

Cloud computing so far in the federal government is more potential than reality, but initial clouds at the Defense Information Systems Agency and NASA hold a promise of widespread future capability.

Information Week Editor-at-Large Charles Babcock has a timely book out on the issue, "Management Strategies for the Cloud Revolution: How Cloud Computing Is Transforming Business and Why You Can't Afford to Be Left Behind," and it includes a chapter on what NASA's experience might portend for the rest of government.

It's a chapter that we're proud make available as unabridged .pdf on our site, a link to which comes right after the text of a interview we recently conducted with Babcock. You can also always follow FierceGovernmentIT's coverage of cloud computing in the federal government.

FGIT: What is that makes NASA ahead of other federal civilian agencies in terms of cloud adoption?

Babcock: I think they've been selected as a test bed, or a pilot case. They've got a lot of innovative expertise in house, particularly at the Ames Research Center, but there are probably highly skilled people at all the NASA centers.

Getting a cloud experiment off the ground there, something that can serve as the prototype, makes a lot of sense. It makes more sense to do it there than the Department of Veterans Affairs, or something like that, that's struggling just to keep its business in order and keep up with the times. I believe that other government agencies are watching the NASA experiment. I've heard that the Office of Management and Budget makes use of the NASA cloud just to illustrate how one federal agency might make use of another's data center.

FGIT: What exactly is cloud computing?

Babcock: Cloud is a convergence of technologies that lets us do things in new ways. There's nothing particularly new in the technology involved. What's new is the distribution pattern. We assemble a large number of servers, than manage them as a group. Google would call that the data center as a computer. Instead of 20 or 30 managers looking at 20 or 30 different machines, you strive for uniformity, similarity, common practices and procedures across the whole group. That way, you simplify the task in front of you in terms of managing the resource.

It's particularly easy if you can virtualize those servers and can generate virtual machines among the physical machines without worrying too much about the characteristics of any one physical machine.

These things combined lead us to a new resource that we can distribute on a per-hour basis. There are ways to measure who is using it, the length of time they use it, and then simply charge back for that use. This is a common practice inside the enterprise already, but with the cloud, we can build the data center out of commodity hardware, as opposed to a high-end Unix server running a database and a mainframe and then a bunch of Intel commodity servers running Windows Office applications and email. If you build it out of those commodity servers and manage it like a mainframe, you've got the best of both worlds, and you've suddenly you have to invest to get a very large amount of computing power.

FGIT: From a federal perspective, what are the major impediments to its widespread adoption?

Babcock: The existing way of doing things is the major impediment for everybody, whether you're in the federal government or in private enterprise. People have to be committed to what they're doing today to keep it up and running, so there's learning, there's computer skills, there's practices and procedures invested into that. Cloud computing is something different, and it takes some getting use to it, it takes a learning curve, it takes a reinvestment into the data center where you begin to convert away from high-end servers toward commodity servers.

A psychological shift is that you will tend to empower the end user to provision the server themselves through software. And this is not the way it's done in today's data center, again, whether it's a federal agency or a private enterprise. There's a great deal of mistrust of what end users might do--the whole idea of giving them the ability to create their own virtual machines on your server cluster is a little bit like giving the inmates the keys to the asylum. You tend not to want to do that. It takes a shift in thinking that the end user will be responsible for what they provision, they'll pay for it, and you the IT manager will make sure that good practices are followed, that the thing is safe and that it is terminated when the end user's job is done.

FGIT: Does the cloud create a single node of failure, in that you're dependent on the network connection for all the functionality?

Babcock: You could view it that way. The cloud data center will be built at the junction of two major networking territories, like two telecom providers, where their territories abut, that's where the cloud data center will build. And so there will be two ways to access a cloud data center, always. One network can fail utterly, plunge everybody in its district into darkness, but it's highly unlikely that the one next door with independent equipment  and power supply will fail.

But it's true, if there is a problem in the network segment between you and your cloud data center, you have to rely on the Internet to route you around that failure, or if it's right at or next to the data center, there may be some disruption. Networks can fail, and other things can fail at the cloud data center. Generally, they are built with redundant systems, they are built with independent zones. If you wished, you could rent a server in one zone at the data center and a second data server as a fail-over server in a different zone. Everything could go awry with your primary server, and it would fail over to a secondary one that has a different power supply, separate hardware, etc. And chances are you would stay up and running.

Amazon encourages people to do that with its EC2 cloud. There are redundant systems built into the cloud itself, and users are encourages to establish their own redundant servers.

FGIT: What about data portability? How likely am I to be able to move my data from one cloud to the next?

Babcock: Uh, not too likely, as things stand today. Different clouds tend to have distinct methods of operating. Amazon has come up with a distinct virtual machine file format, it only works for the Amazon cloud. But I think the day will fast approach where you will be able to transfer between clouds. It may take a virtual file format conversion. This exists in a rudimentary form from the Distributed Management Task Force. It's a standards-setting group that has a migration format, or an import format, where if you put your virtual machine in that format, no matter where you send it, the hypervisor there will tend to recognize it. This is very rudimentary mobility, because once it's into that cloud, you would have to reconvert it out of that cloud's format to move it to another cloud.

We've got a one-way street up and running. What we need is a two-way street. We don't have that, yet. But, I think it's inevitable that it will come and already there are front end cloud service providers, like RightScale. Elastra is another, that will convert your virtual machine format into one that works in your target cloud. If you decide to move, they will convert back again for you. This incurs more fees--in my view, unnecessary fees because there ought to be some common mobility standard to move things around. Cloud computing will be greatly furthered when that's the case. But that's not the case today.

Read "Nebula: NASA's Strategic Cloud," a chapter from Charles Babcock's new book on cloud computing. The chapter, and the book, are copyright The McGraw-Hill Companies, Inc.