Amazon.com is to high performance computing what McDonald's is to food: fast, cheap but with a limited menu.
But while some HPC users may refer to Amazon's Elastic Compute Cloud (EC2) service as a "CPU farm" or a "CPU bin," there are aspects of the company's model and pricing scheme that may be having an impact on supercomputing centers that have been typically been the domain of researchers.
[ Get the no-nonsense explanations and advice you need to take real advantage of cloud computing in the InfoWorld editors' 21-page Cloud Computing Deep Dive PDF special report. | Stay up on the cloud with InfoWorld's Cloud Computing Report newsletter. ]
Amazon may be helping supercomputing centers indirectly by popularizing the pay-as-you-go model, which makes it easier for the centers to attract business users and clearly contrast their services against the online giant.
Access to supercomputing is increasingly seen as critical to U.S. competitiveness for its ability to test designs virtually and speed products to market. That has encouraged some states, including Ohio, New Mexico, and Montana, to invest in supercomputing programs aimed at smaller businesses.
"You cannot just have the biggest of Fortune 100 companies using supercomputing to become more competitive -- you really need to enable the whole supply chain to become more competitive," said Ashok Krishnamurthy, senior director, research and scientific development at the Ohio Supercomputer Center in Columbus.
OSC is working to make its model easier to use. This month, it announced that it had developed with Nimbis Services, which provides access OSC's compute services through a Web portal. Users can pay for the service using using a credit card, like the EC2 model.
But for the most part, that's where the comparison to Amazon ends.
Nimbis also negotiates terms with HPC computing application providers to enable users to run those apps on OSC's platform. Nimbis is a for profit software company that can make agreements with vendors more readily than an academic supercomputing center can, said Krishnamurthy.
Supercomputing centers say they can provide expert help as well as ISV software, fast interconnects, large memory footprints, and other technologies not available though commercial cloud infrastructure providers. While Krishnamurthy said the broader familiarity with Amazon's model makes it easier to explain their process to customers, "our model is not Amazon's model."
The New Mexico Computing Applications Center in Albuquerque, which was funded by the state and operates as a nonprofit, offers customers a pricing sheet for compute cycles and monthly invoice. That keeps the cost management simple as the company goes "after specific industries that can help bring high paying jobs to the state," said Scott Collins, the center's CTO.
"We're not competing with [Amazon]," said Collins. "Our partnerships with the national labs allow us to leverage expertise to bring in people to work on the hard problems."
In Butte, Montana, Rocky Mountain Super Computing center was established with similar goals. It uses a large cloud computing platform that also allows it to move work to IBM's Computing on Demand Centers for projects that require more compute power than it has on hand. Similar to Ohio and New Mexico, its primary mission to develop jobs and make compute resources easily adopted through a "pay as you go" or "on demand" model.
At the SC09 supercomputing show here this week, Amazon.com had a booth for the first time, tucked away in a corner of the trade show floor. The booth was staffed by people who said they couldn't talk to the press without public relations office involvement.
That was no matter, as there were other firms on show, such as Cycle Computing, that were working to make platform easier to use with management tool that scheduled and execute compute cycles on Amazon's cloud, allowing user to easily scale their codes over as many processors as they need, according to Jason Stowe, the company's CEO.
This story, "Supercomputing centers acknowledge Amazon's influence" was originally published by Computerworld.