New job for mainframes: Hosting private clouds
"A mainframe is a cloud," contends Jon Toigo, CEO of Toigo Partners International, a data management consultancy in Dunedin, Fla.
If you, like Toigo, define a cloud as a resource that can be dynamically provisioned and made available within a company with security and good management controls, "then all of that exists already in a mainframe," he says.
Of course, Toigo's isn't the only definition of what constitutes a cloud. Most experts say that a key attribute of the cloud is that the dynamic provisioning is self-service -- that is, at the user's demand.
But the controlled environment of the mainframe, which is the basis for much of its security, traditionally requires an administrator to provision computing power for specific tasks. That's why the mainframe has a reputation as old technology that operates under an outdated IT paradigm of command and control.
It's also one of the reasons why most cloud computing today runs on x86-based distributed architectures, not mainframes. Other reasons: Mainframe hardware is expensive, licensing and software costs tend to be high, and there is a shortage of mainframe skills.
Nevertheless, mainframe vendors contend that many companies want to use their big iron for cloud computing. In a CA Technologies-sponsored survey of 200 U.S. mainframe executives last fall, 73% of the respondents said that their mainframes were a part of their future cloud plans.
And IBM has been promoting mainframes as cloud platforms for several years. The company's introduction last year of the zEnterprise, which gives organizations the option of combining mainframe and distributed computing platforms under an umbrella of common management, is a key part of IBM's strategy to make mainframes a part of the cloud, say analysts.
The company set the stage 10 years ago when it gave all of its mainframes, starting with zSeries S/390, the ability to run Linux. While mainframes had been virtualizing since the introduction of the VM operating system 30 years earlier, once IBM added Linux, you could run virtual x86 servers on a mainframe.
Over the past several years, some organizations have done just that, consolidating and virtualizing x86 servers using Linux on the mainframe. Once you start doing that, you have the basis for a private cloud.
"You have this incredibly scalable server that's very strong in transaction management," says Judith Hurwitz, president and CEO of Hurwitz & Associates, an IT consultancy in Needham, Mass. "Here's this platform that has scalability and partitioning built in at its core."
Plus, the mainframe's strongest assets -- reliability, availability, manageability and security -- are the very characteristics that companies are most concerned about as they consider rolling out major business applications in the cloud, she says.
The Sticking Point: Provisioning
But that lack of support for self-provisioning is glaring. "The mainframe is very well controlled in most organizations, often to the point where it's locked in a room and people can't access it," says Julie Craig, an analyst at Enterprise Management Associates. "[Mainframe vendors] are going to have to do some developing to allow the self-service features of the cloud."
Reed Mullen, IBM's System z cloud computing leader, says that the lack of self-provisioning is cultural, not technological. Companies could enable self-provisioning in mainframes either by using IBM's Tivoli Service Automation Manager or through custom development, he says.
The five characteristics of a private cloud:
1. Scalable: High levels of utilization (e.g., through virtualization), with large, mature data centers.
2. Accessible: Users can provision resources on their own.
3. Elastic: Appearance of infinite capacity on demand.
4. Shared: Workloads are multiplexed; capacity is pooled.
5. Metered consumption: Ability to pay for use with no commitment.
Source: Corporate Executive Board's Infrastructure Executive Council, Arlington, Va.
And yet he acknowledges that such implementations would still depend on the IT department -- users wouldn't have full self-service autonomy. Specifically, mainframe systems with self-provisioning options would require a user to submit a request by email, and IT would have to approve the request before the resources were provisioned, Mullen explains. This reflects the "old habits" of the mainframe world, he says. But he also notes that any kind of cloud implementation, including those on distributed systems, would include an approval process.
"I know the perception is that the user doesn't have to bother anybody in IT -- that I just have to point and click to get my service," Mullen says. But in every cloud scenario, he adds, there's some kind of approval process -- a way to prioritize the requests -- even though that process may not "require human eyes."
As for the licensing costs, Mullen says that IBM's current generation, System z, has a little-used "on-off" feature, whereby mainframe administrators can turn a processor core on for a limited time, paying short-term day rates for IBM software rather than buying an expensive annual license based on the number of processor cores. "We are looking at taking advantage of this infrastructure to make it even more suitable for a cloud environment where there is a lot of unpredictable usage," says Mullen.
But it's hard to find an organization that's using a mainframe in a self-provisioned cloud computing system. Some analysts say the talk of the mainframe as cloud is just hype. The technology may indeed exist, but the question is whether companies are actually using it, says Bill Claybrook, president of New River Marketing Research in Concord, Mass. "If they are not automating things, if they don't have a self-service portal, then it's not a cloud architecture; it's just a virtualized environment," he says.
One reason why it's hard to find a self-provisioned mainframe-based cloud may be because we're still in the early days of cloud computing. "There is incongruity between what's out there in cloud today and what these big mainframes do," says Phil Murphy, an analyst at Forrester Research.
Business units might use a credit card to buy some extra compute cycles for a one-time project, for example, but most companies wouldn't run mission-critical transaction-processing applications in the cloud.
The one cloud scenario that includes self-provisioning is the model used by global outsourcing companies, where far-flung developers have the ability to automatically set up their own testing and development platforms. Those aren't all mainframe-based, but Murphy thinks some of them must be.
Mullen agrees that the offshoring model is a good example. A platform-as-a-service setup like that "is perhaps the dominant usage of a cloud infrastructure in mainframe environments today," he says.
But as cloud computing matures and as new models of mainframes begin to offer more computing power at lower costs than they do today, more companies will experiment with the mainframe-based cloud. Hurwitz, for one, says many of her clients are looking into it, although none are ready to talk about it publicly. "It's something we're going to see a lot more of," she predicts.
The Very Early Adopter
Marist College is a poster child for IBM mainframes. The college is right down the road from an IBM mainframe manufacturing plant in Poughkeepsie, N.Y. Marist has had a research-and-development partnership with IBM for more than 20 years, and it helped IBM develop and roll out System z Linux.
Marist has rewritten many x86-based applications to run on Linux on its two System z mainframes. The college runs 80 Linux servers, mostly handling administrative tasks, on one mainframe, and it has more than 600 Linux servers running academic applications on the other.
The college runs other applications on an IBM System p midrange computer and IBM blades as well. But the mainframes are "the real engine," says Bill Thirsk, Marist's CIO.
Marist is getting big cost benefits from virtualizing on the mainframe. The college avoids purchasing extra server hardware, plus it saves on space, power and IT staff to manage the data center. It not only avoids having to pay extra for each application it adds to the mainframe, but also benefits from increased utilization of the mainframe, resulting in a very good return on assets, says Thirsk. He calls Marist's setup a cloud.
Skeptics would say it's not a cloud, because it has no user provisioning. But there is some provisioning going on: When students enroll to study computer science, for example, they are automatically provisioned with a mainframe partition, Thirsk says. And when they leave the school, he adds, "that's sucked back into the fold and re-allocated automatically."
Though critics might disagree, Thirsk says the lack of user provisioning isn't important.
"The fact is that if you wanted to change the policy [to] where the student could just order it, it would come down to the same auto-provisioning routine," he says. "We do it more explicitly because it's an academic institution. The faculty decide what resources get used by students, depending on their courses."
Marist has advantages that make building a mainframe-based cloud easier. It gets an academic discount on the mainframes, although the price breaks aren't any larger than those available to other universities, says Thirsk. And thanks to an IBM-sponsored mainframe academic program at the college, Marist has a built-in, cheap source of IT labor with mainframe and System z Linux skills.
"Where one CIO might have to hire very expensive professionals to run their data center, I have an entire internship program, and my labor's fairly inexpensive," Thirsk notes. "I only have three professionals to supervise."
Marist's cloud is starting to get some attention. "Four years ago, when I started talking about this, everybody looked at me like I was crazy," Thirsk says. But as the years have passed, others have taken an interest in Marist's computing environment. He notes that he has hosted lots of visitors eager to learn what the college is doing, including representatives from 21 companies and several universities last year. "We're talking to a college in the Middle East that has over 200,000 students," Thirsk says. "There's only one way to meet that load: with a mainframe."
Along with several concurrent developments, zEnterprise could make the mainframe into a true cloud platform, says Susan Eustis, president of WinterGreen Research in Lexington, Mass. Just in the past several months, she says, IBM has improved WebSphere, improved z/VM and adjusted its pricing structure -- all moves to make the mainframe more cloud-friendly, she says. Eustis thinks that IBM now has all the pieces in place to enable business units to self-provision a mainframe-based cloud.
At the very least, zEnterprise could change the traditional thinking about mainframes. "I think you'll start seeing the mainframe viewed in a different way," says Hurwitz. As mainframes begin to run more of the same software as other high-end servers and gain expanded service-management capabilities, "people are going to see it as the high end of the server market as opposed to a world unto itself."
Harbert is a Washington, D.C.-based writer specializing in technology, business and public policy. She can be contacted through her website, TamHarbert.com.