The project will span six IBM data centers at locations in Poughkeepsie, N.Y.; Southbury, Conn.; Boulder, Colo., Portsmouth, England; Osaka, Japan; and Sydney, Australia. Bernard Meyerson, the VP for strategic alliances and CTO for the systems and technology group, said that the data centers themselves will not be consolidated but that space would open up at each site to make room for expansion. He added that the move will consolidate the data center space by 99%.
Meyerson said the existing 3,900 servers are a balanced mix of x86 and Unix boxes, though he didn't know the exact proportions of each. The Unix servers primarily run AIX, while the majority of the x86 servers run Linux. Once consolidated onto mainframes, they'll be running Linux on top of z/VM, the mainframe's operating system. The mainframe can support thousands of Linux instances on one box.
Dan Olds, a principal analyst at Beaverton, Ore.-based Gabriel Consulting Group Inc., said that IBM has long touted the benefits of server consolidation onto the mainframe. Now the company can see for itself how consolidation works.
"It's a hell of a big consolidation," he said. "This is a bold move in terms of eating your own dog food."
The predicted $250 million savings will come from three areas:
- Reduced energy costs: IBM will run less than one one-hundredth the number of servers as it did previously.
- Software licensing costs: Often priced per processor, licensing costs will be slashed, since fewer processors are needed to run the 30 mainframes than to run the 3,900 distributed servers.
- Staff costs: IBM predicts that the new infrastructure will free up IT staff to work less on systems administration and more on projects with customers.
IBM operates about 8 million square feet of total data center space. It could not estimate the total floor area of this project, although the U.S. sites take up about 184,000 square feet. As for the fate of the 3,900 servers, Meyerson said that IBM would send less than 1% to a landfill; many will be recycled through IBM's own asset disposal service.
Why would IBM trash thousands of its own servers to consolidate on the mainframe? Doesn't that make its other server platforms look bad?
Olds said that IBM's decision to consolidate isn't necessarily because the Unix or x86 hardware itself is poor. Like many other businesses, most of IBM's x86 servers are probably running at low utilization rates -- at 10% or less. Rather than virtualizing the x86 servers with software from VMware Inc. or XenSource Inc., IBM has decided to consolidate on its own mainframe, which has been doing server virtualization for half a century.
The move is also an example of a massive scale-up project and is best contrasted with a scale-out project like Google Inc.'s, which relies on thousands of cheap, custom x86 servers. Google assumes that the underlying hardware will fail but that the software on top will provide redundancy.
Which way should you build your server infrastructure? "What is the one true path? There isn't one," Olds said. "What Google is doing is the right thing for Google. They have very specialized workloads. Google is an interesting case to talk about, but it's really different from what regular customers do with their systems. With a company like IBM or any other company that doesn't run one of the largest Web-serving workloads in the world, it's different."
Let us know what you think about the story; e-mail Mark Fontecchio, News Writer.