The first service pack for Windows HPC Server 2008 R2, due before the end of this year, will let customers connect their on-premises high performance computing systems to Windows Azure, giving them "on-demand scale and capacity for high performance computing applications," Microsoft said.
Microsoft is also providing an Azure resource for scientists that will not require an installation of Windows HPC Server. The service makes the National Center for Biotechnology Information's BLAST technology, which lets scientists search the human genome, available on Azure. At SC10, Microsoft said it will demonstrate the NCBI BLAST application on Windows Azure performing 100 billion comparisons of protein sequences.
The new Windows HPC Service Pack's integration with Azure, meanwhile, gives Microsoft what it believes is a key differentiator between itself and the likes of Amazon's Elastic Compute Cloud: the ability to run supercomputing workloads across both in-house software and over the Internet on the cloud service.
"There is no on-premise Amazon, there is no on-premise Google computing resource," says Bill Hilf, general manager of Microsoft's technical computing group. "It's one of the big advantages we have."
HPC software is "really just a job scheduler that knows how to break up work and distribute it across a bunch of other servers," Hilf says. Integrating Windows HPC Server with Azure lets a customer's data center "talk to the Windows Azure system," and spread work across the two, he says. This makes sense for workloads that have large, temporary spikes in calculations.
Essentially, Microsoft is taking the concept of "cloud-bursting," the ability to access cloud-based computing resources in an automated way when applications need extra processing power, and applying it to the HPC world.
"This burst demand has been at the top of our HPC customers' requirements," Hilf says.
As for the NCBI BLAST announcement, Hilf notes that the code is in the public domain, but says running BLAST calculations on the Azure service will give scientists the ability to run gigantic database queries without investing in expensive hardware. In addition to porting BLAST to Azure, Microsoft built some Web-based user interfaces to make running calculations a bit easier, he says.
The cost of running BLAST on Azure will be the same as running any Azure workload, the price of which goes up as customers use more computing power. The 100 billion comparison BLAST workload, for example, was a query that took place on 4,000 cores over about six days for a price of less than $18,000.
While BLAST is the first HPC application Microsoft has offered on top of the Azure service, the vendor says more will come in the future. Even without such specific offerings, Microsoft says some customers have already begun running their own HPC workloads on the Azure cloud.