The applications are probably the No. 1 factor and, in particular, how many Linux applications you have, according to IBM, analysts and mainframe users. If you have only a single Linux server running one application, it isn't worth buying an Integrated Facility for Linux (IFL), the specialty mainframe processor that will run you about $100,000, depending on your mainframe.
But if you have thousands, hundreds or even dozens of Linux servers, you might want to consider the mainframe. That's because all your Linux servers can run as virtual instances on top of z/VM, the mainframe's virtualization operating system, on the IFL engine.
John Phelps, a research vice president of servers at research firm Gartner Inc. who recently wrote a report exploring the argument for putting Linux on the mainframe, said that "if you're talking 20 to 30 applications, you're probably not looking at the mainframe. If you keep going up, it has great consolidation capability."
In a similar vein, Brad Day, an analyst at Forrester Research Inc., said that the sweet spot for breaking even on the z9 Business Class – a smaller mainframe that starts at about $100,000 – is 25 to 30 Linux instances.
Mainframe attractive consolidation target
Whatever your break-even point, Linux on the mainframe is an attractive consolidation target. That's what the Canadian province of Quebec did when it consolidated hundreds of midrange servers running Oracle on Unix to a single mainframe running Oracle on Linux.
That consolidation onto five IFL engines reduced the Quebec government's overall hardware and software licensing costs. Though the cost of the mainframe hardware was about the same as buying Unix server hardware, the agency estimated that it saved $800,000 in licensing costs in the first year alone. Oracle licenses are based on the number of processors used, so the agency saved money by being able to host several Oracle instances on each IFL specialty processor.
David Kreuter, president of Ontario-based VM Resources Ltd., which consulted on the project, said the only other hurdle was training some staff to learn Linux and z/VM.
"We worked out a lot of pressure points by spending a good deal of time developing a very solid architecture," he said. "The savings and license reduction was what was attractive."
Supermarket chain Hannaford Bros. is also consolidating Linux instances on its existing mainframe. Right now CIO Bill Homa said the company has about 70 Linux instances on big iron, and setting up a new one is easier than configuring a physical Linux server.
"You can set up a virtual Linux server on the mainframe in about a minute, just with a few commands," he said. "In fact, our corporate Web server for our Internet is moving to the mainframe tomorrow from a couple servers, and that's not a very difficult move."
These consolidations have also helped both organizations reduce their data center footprint, allowing Hannaford to save on power and cooling costs. That's something that IBM has touted for its mainframe. Earlier this month, IBM announced that over the past two months it had measured the energy efficiency of about 1,000 mainframes and found that they were much more efficient than a scale-out x86 infrastructure.
Workload matters – or does it?
Phelps said that the decision to move to Linux on the mainframe may depend on the kinds of Linux applications running there. Typically the applications that Phelps said should stay on Unix boxes or in an x86 grid structure would be heavy compute-bound workloads – for example, a number-crunching research application that universities run for simulations, which takes up a lot of CPU. If you're going to have several other applications running at the same time, the mainframe may not be the ideal consolidation target.
But Charlie Burns, a vice president at Westport, Conn.-based research firm Saugatech Technology Inc., didn't necessarily agree. If the person administering the mainframe knows how to manage it so the research application doesn't sabotage the server's CPU, then it's possible to put those workloads on big iron.
"What you need to do is make sure that you don't allow that extraneous workload to consume the entire mainframe," he said. "That's a matter of setting priorities and workload objectives, and the tools exist to do that quite handily."
Phelps said that another factor is whether your Linux applications will be supported running on the mainframe. Even though the transition requires only a simple recompile, some independent software vendors (ISVs) may certify their applications to run only on x86 Linux and not the mainframe. Phelps said that IBM is making time available on some of its own systems for ISVs to certify their apps.
If you're writing your own Linux applications in-house, adjusting them from x86 Linux to mainframe Linux shouldn't be a big hassle.
"If it's not supported on Linux on the mainframe, people aren't going to run it there," Phelps said. "So are you developing your own Linux applications, or are you depending on ISVs?"
Having a mainframe helps
Then again, if you're not a mainframe shop, how can you justify that up-front capital cost to your company's financial officers? IBM can make all the sales pitches it wants, but if companies are looking at the decision from an immediate bottom-line perspective, it may matter.
"The mainframe is an environment that many people aren't even used to running," Phelps said. "One of the things it depends on is if you have experience on it already. If I already have a mainframe, putting an IFL on there is not that big a deal."
Even for those companies that do have a mainframe, transitioning can be difficult from a personnel perspective. At the province of Quebec, staff accustomed to Unix environments like AIX and Solaris had to be re-trained in z/VM and Linux. Kreuter said most of the issues involved "improving the skills of the staff to accept new z/VM and Linux technology."
In addition to learning new operating systems, staffers need to know how to manage a virtualized environment. Saugatech's Burns said that some things you might do to increase the performance of a physical Linux server could harm a virtual Linux server on z/VM.
IT staff, for example, might be accustomed to adding memory to a server to increase performance. But if you add virtual memory to a virtual server, you could affect other virtual servers running on the same physical host.
"You have to be more broad in your thinking in terms of 'How do I manage, how do I run a virtualized infrastructure?'"