The program is part of IBM's Project Big Green, announced in May, which aims to redirect $1 billion a year to make data centers more energy efficient. IBM today also announced energy consumption data it collected in August and September for roughly 1,000 mainframe customers.
Dave Anderson, an IBM green consultant, said the data indicates that a mainframe running virtualized Linux instances can do the work of about 250 x86 processors while using as little as 2% of the energy. In addition, IBM found that the mainframe's typical energy use is about 60% of the maximum listed on the machine's label. For the most popular mainframe, the z9 Enterprise Class S18 – which can scale between one and 18 central mainframe processors – nine out of 10 servers measured used less than 50% of the maximum energy amount allowed by the label rating.
The energy numbers make sense, said Brad Day, an analyst at Forrester Research Inc. First, he said, the mainframe is an attractive server consolidation target, especially when it comes to Linux. By installing a specialty processor called the Integrated Facility for Linux (IFL), a mainframe can take hundreds or thousands of Linux servers -- and on a single mainframe that runs z/VM, the mainframe's virtualization operating system.
Still, Day said, you can be sure that IBM picked the right machines to be measured to get the data it wanted.
"Make no mistake, the machines they're going to pull from to make their case are going to be very advanced, with high utilization rates and collapsing a lot of workloads," he said. "Those System z environments will be virtualizing as many workloads as possible. And it's going to be on that level of argument that the mainframe has no peer."
Day recommended keeping tabs on IBM, which over the next five years plans to consolidate 3,900 Unix and x86 machines onto 30 mainframes, a project that IBM estimates will save it about $250 million. Day said it's a case where IBM has taken its own consolidation medicine. As IBM investigates how much in space, power and cooling – measures that translate to cost – it saves, customers can also see the benefits of a consolidation project.
Try this at home
With the announcement today, customers can also begin measuring their own mainframe energy use. Using internal sensors, IBM can now gather power and cooling data from a mainframe and present it to a customer. Users can then correlate that data with how much work the mainframe pumps out. IBM also announced a product called the Power Estimator Tool, which calculates how changing a mainframe's configuration -- adding an IFL, for example -- will affect its energy consumption.
Whether users need or want these offerings right now is another question. According to two mainframe users, they're not worried about the energy efficiency of their machines; indeed, they said, they already know that the mainframe is efficient.
"Power is always nice, and having machines that use less power is always something that we think about, but it isn't our major focus when we're looking at new machines," said Ted Sikorski, acting director of computing and networking services at the University of Toronto.
Meanwhile, supermarket chain Hannaford Bros. worries more about how much energy its x86 servers consume than its mainframe. But that doesn't mean that CIO Bill Homa doesn't realize that the mainframe is more efficient.
"Our Windows servers' power consumption dwarfs our mainframe power consumption," he said. "It's not 100 x, but it's at least 10x in power and air conditioning, space, management. It's pretty much everything."
Hannaford is moving workloads from x86 onto a mainframe and has consolidated about 70 Linux servers onto an IFL. Day said when you "sweep the floor" of all these physical x86 servers -- which take up processing, memory and I/O that all require power and cooling – and consolidate them on one platform, that's where the mainframe flexes its muscles.
"Power and cooling becomes truly a much lower cost of the overall lifecycle cost of the system environment," he said.Influencing federal guidelines?
With this announcement, IBM claims it is the first vendor to support recent federal recommendations that encourage vendors to publish their server energy consumption figures. This past summer, the Environmental Protection Agency (EPA) issued a report on server and data center energy efficiency, along with suggestions on how to start measuring the energy efficiency of servers and data centers.
Through its Energy Star program, the EPA, plans to begin releasing benchmarks and labels at the end of this year. These measures will aid data center managers in determining the energy efficiency of their facilities and which vendors offer the most energy-efficient servers. These metrics could include an Energy Star label from the EPA and an energy-efficiency benchmark from the Standard Performance Evaluation Corp. (SPEC).
But does IBM's way of measuring the energy efficiency of its mainframes jibe with Energy Star and SPEC?
As it stands, IBM's preferred method for measuring energy efficiency is to look at watts per image, said Anderson, IBM's green consultant. That approach "allows you to have an energy efficiency metric that is also related to you running workloads," he explained.
By releasing its mainframe energy-efficiency metrics now, IBM is trying to influence the direction of federal agencies. "What IBM is trying to do is create its own tool to reinforce how the EPA should look at one technology over another," Day said. "They're hoping that what they're doing will help amplify the new metrics, that the EPA will take notice."
Let us know what you think about the story; email Mark Fontecchio, News Writer.