Microsoft Makes HPC Announcements at Supercomputing Conference
Microsoft this week at the SC10 supercomputing conference
in New Orleans announced updates to Windows HPC Server 2008 R2 designed to enable more flexible and economical high-performance computing (HPC) environments. Service Pack 1 (SP1) of Microsoft's HPC platform will enable shops to extend the capacity of on-premise data centers by allowing them to draw additional computing power from Windows Azure-based servers -- a feature Microsoft calls "cloud bursting."
"With the cloud bursting concept, that job scheduler knows how to talk to Windows Azure," explained Bill Hilf, general manager of Technical Computing at Microsoft. "Now in the same management console you can bring in more nodes to add to that queue of work."
The cloud bursting functionality is expected to ship next year, according to Microsoft.
Microsoft also announced the availability of NCBI Blast, a Windows Azure application for biological research that allows researchers to blend on-premise and cloud-based system resources. NCBI stands for National Center for Biotechnology Information. Microsoft President of the Server and Tools Business Bob Muglia at the conference said the application demonstrates the "platform-as-a-service capabilities" of Windows Azure.
Microsoft also announced that Windows HPC Server 2008 R2 had achieved performance in excess of one petaflop.
"There are less than 10 computers in the world that can actually do this type of performance. It's a gigaflop per watt. We are able to run a very large system in a way that is very power efficient." Hilf said.
Hilf said the HPC sector plays an increasingly important role in enabling capabilities in mainstream application development. He said HPC environments have for years dealt with issues of scalability and parallelism that mainstream apps today are just starting to address.
"In many ways, it's how people re-thought applications when they started looking at three-tier applications and Web-based applications with fat clients. I think the same thing is going to happen with the cloud," Hilf said. "I very much think that the trend of parallelism, big data, cloud -- if you kind of squint at it a little bit, you can say the high end of HPC has been attacking these problems for several years," Hilf said.
"We are really at this impasse or huge challenge. If you look at the number of applications that are written today, they are largely serial," Hilf continued. "Great, solve the problem for making parallelism easy, but don't solve it for a single system. Try to abstract it up high enough so people think about parallelism for multi-core and the cloud in a uniform way."
About the Author
Michael Desmond is an editor and writer for 1105 Media's Enterprise Computing Group.