IBM customers looking for some extra processing power can now lease supercomputer-sized Linux and Unix clusters over the Internet, thanks to a new computing facility in Poughkeepsie, N.J., that Big Blue will open on Tuesday.
The new "Deep Computing on demand" facility will initially house 600 clustered servers: 500 eServer x335 and x345 machines, and 100 p655 computers. But the building is large enough that it could be expanded to house thousands of servers, should that be necessary, according to IBM Vice President of Deep Computing Dave Turek.
The idea is to give customers with intensive but short-term computing needs the option to rent out a large clustered computer rather than build it themselves. IBM is now booking the center for compute projects varying between two weeks and three months in length, Turek said.
The new center is not for everyone, however. "This is not a facility that can be envisioned as meeting all possible kinds of needs," Turek said. It will initially serve customers in the petroleum and life sciences industries, he said.
One of the center's first customers is Houston's GX Technology, which expects to begin using the center to run its EarthWave geological analysis software on IBM's Intel clusters running Linux within the next few months.
The center will let GX take on new work when its own 4,000-node cluster is already being used, said GX Chief Executive Officer Mick Lambert. Company technicians will be able to remotely configure IBM's systems to run an EarthWave simulation "within a few days," he said.
"We miss out on a lot of opportunities because of the three months it takes to build out a facility," Lambert said. "Being able to get access instantaneously creates opportunities that we would have to pass on otherwise," he said.
The first company to sign up for this on-demand supercomputing service was Petroleum Geo-Services ASA's PGS Data processing division, which signed up to do a three-month seismic imaging project in January.
IBM is now internally discussing the possibility of opening other such facilities in Europe, the Middle East, and Asia, as well as other locations in North America, Turek said. The worldwide centers will eventually be "grid-enabled," in order to share compute power, he said.