Q&A: Opsware’s Utility-Computing Trump Card

Even though HP, IBM, and Sun have monopolized the utility-computing limelight, Opsware believes it has a trump card up its sleeve

Opsware Inc. has been developing data-center automation software since the late 1990’s. Thanks to high-profile deals with Electronic Data Systems and Hewlett-Packard Co., the companyis one of the most recognized vendors in the nascent data center automation—or utility computing—space.

We spoke with Dr. Time Howe, Opsware's chief technical officer and founder, about the company's surprisingly storied history—it has Internet royalty in its executive bloodline, after all—and about the advantages Opsware brings to the table. Even though HP, IBM Corp., and Sun Microsystems Inc. have monopolized the utility-computing limelight, Howe believes that Opsware has a trump card up its sleeve: It’s shipping software now—today— that delivers on the vision touted by these and other vendors.

Not all of our readers are familiar with what Opsware does, so I was hoping you wouldn’t mind giving us a bit of background on the company’s history.

Sure. We are a software company, and we make data-center and network automation software that’s really aimed at helping IT organizations lower their costs, increase quality, and improve security, through regulatory compliance. We started life under a different name. We were called LoudCloud, founded in 1999 as a managed services company.

That was Marc Andreessen’s big project after Netscape, right?

Yes, it was. The idea in that business was that companies that had large, mission-critical Web sites, we would run those sites for them. Companies like Nike and Ford, News Corp., USA Today, a bunch of other large companies with very large Web sites, we ran [their Web sites] for them. The genesis of our software was actually in that business. We created the software for our own use in the managed services business.

How did you get out of the managed services business and focus more on being a software company?

Two things happened, really, that convinced us to do that. The first-worst thing was the economic downturn, which affected everyone across the board. The second-worst thing that happened is that our large competitors, Exodus and WorldCom (and others), started to go bankrupt, which kind of made us look questionable by association.

But we still saw very strong interest in the software. Customers would say, “I’m not into the outsourcing business right now, but can’t you put that in a box and sell it to me to make my business more efficient?” And that was something we’d been talking about doing since the very beginning of LoudCloud.

You’ve had one very high-profile win, with EDS, where they licensed your software to help manage their own data centers. You’ve also notched an agreement with HP, I believe, for a similar purpose. What kind of interest are you seeing from enterprise customers—that is, enterprise organizations that want to implement your software in their own environments?

First of all, yes, EDS licensed that software in a separate transaction for their use across their entire hosting business, and that gave us a $52 million contract to start off life with when we changed our name to Opsware and focused just on being a software company.

We have over 250 customers right now, and we have about $100 million in the bank. We’ve expanded from our core server-automation product line to where we’re now doing network automation, too, and we’re seeing a lot of interest in that from customers, too.

At the same time, the data-center management and automation niche that you more-or-less claimed as your own four or five years ago is suddenly very competitive. IBM has On Demand, HP touts Adaptive Enterprise—but where do you propose to fit into the picture?

From a customer perspective, there are really two things that drive customers to choose Opsware versus one of the large players. We actually have something today, a product to offer today, as opposed to much of what you get from the systems vendors, which is really more of a services offering, and some products that were built for the client-server world.

If [customers] want to get started today, on the path for a utility-computing vision, they need a tool that was built for that world, and Opsware is that vision. Also, we hear from customers that they don’t want to get locked in to a [single] systems vendor, they don’t like the idea of buying the entire stack from Sun, or from IBM, or from HP, and then getting locked in the cycle of continually having to upgrade hardware.

Do you have any limitations in terms of the kinds of systems and applications, or the hardware resources, you can support?

We support all of the major Internet-related platforms, Windows, AIX, Solaris, HP-UX. We support virtually any software that sits on top of those things, all of the hardware that they run on, basically.

One important thing to note about Opsware is that we have a very broad range of tasks and technologies that we automate the management of, so it’s not just at the hardware and operating system level. We also do patching. We go all the way up to the ultimate application that sits on top of those things, and again that breadth of coverage is really what you need to make your IT infrastructure more responsive.

Okay, but how do you manage that? To the uninformed, it kind of sounds like you’re painting a too-pretty picture of what we all know is a difficult problem. How do you propose to manage, automate, and provision across different applications, platforms, and hardware resources?

One way we do this is by using the vendor’s own [management tools] where they’re available. It’s a general philosophy of Opsware to use the tools that are provided by the vendors on each platform. So, for example, we’ll use the underlying Jumpstart technology provided by Sun to assist in operating system provisioning. We use underlying tools when we can get them, but we also have built our own tools, so regardless you can deploy [applications, operating platforms, or hardware resources], you can install them, you can uninstall them, and you can control them afterwards.

Like it or not, there’s a perception that data-center automation, or utility computing, or whatever you want to call it, is still a pretty immature technology area. What are some of the issues customers must grapple with today if they’re going to use their software to automate their data centers?

Most of the sexy part of data-center automation is all about that capacity on demand and being able to scale up [and] scale down applications, and that is relatively immature. IBM bought ThinkDynamics, Sun bought another vendor.

There is actually a bigger part of the overall utility-computing puzzle, which is the automation piece. It’s one thing to be able to add capacity and remove capacity in response to what your monitoring system tells you is the load on your systems. It’s quite another to be able to do the 98 percent of what has to be done in response to change management on these systems. When you want to get information on these systems or do a mass change across them, you’ve got to write scripts to do it. That’s what system administrators spend the vast majority of their days doing, and that’s where Opsware has focused most of its time.

I’ve seen some initiatives around utility-computing standards, like the DCML, but where are we today in terms of defining and approving standards that can allow all of the different utility-computing visions to interoperate and coexist?

There’s kind of the de facto standards on platforms, but in terms of vendor-independent standards in the space, there is not much today, which is why we along with EDS and CA went off and started the DCML effort.

If you look at what needs to happen, it’s in part because of the huge amount of legacy, not only hardware and software, but also management systems that customers have in their environments. You want those systems to talk to one another. You’re not going to get from any single vendor an application that can manage your provisioning and your monitoring. So we’re probably not going to have just one standard that does all of these things, either.

Sort of a quiver of interoperable standards, then?

Yes. Look at just the Internet itself. There’s not just one standard for the Internet. There’s the whole raft of them, and I think that you look at what’s going to be required to make everything work together in a utility-computing model, you’re bringing together a hugely diverse set of technologies and tools.

DCML is one. We’re focused on the data exchange between the different actors in the utility-computing exchange, [but] there [are] other standards related to how you exchange the DCML information, how you represent it in systems, and so on.