By Kevin Davies
November 17, 2009 | Microsoft is making news at the Supercomputing 2009 (SC 09) conference in Portland, Oregon, and not merely for the free cocktails and flight simulator drawing attendees to its booth. It has announced the availability of betas for Windows HPC Server 2008 R2 and distributed Microsoft Office Excel 2010 for the cluster.
According to Vince Mendillo, Microsoft’s senior director for high performance computing, the overall goal is to make supercomputing more accessible to many users, including scientists, engineers, life scientists “to help many of those users model the real world more accurately – whether they’re doing research, development, or scientific discovery.”
Speaking to Bio-IT World, Mendillo said Microsoft’s strategy is to “make general purpose platforms and tools and rely on our partners to customize their specific applications.” As momentum grows, Mendillo counts more than 150 independent software vendors (ISVs) that partner with Windows HPC server and reporting notable increases in performance.
This is the third release of Windows HPC server. It is based on the core Windows Server operating system, but with added components, such as a job scheduler, system maintenance tools, and programming model, to make it conducive to the needs of HPC users in the marketplace.
Ryan Waite, product unit manager for Windows HPC server, says there are many users who are not expert computer scientists. “They’re biologists or computational biologists or statisticians. For them to be able to set up a cluster is a pretty high bar. When we built Windows HPC server, it was an opportunity to integrate things like deployment, diagnostics, job scheduling, and the programming models associated with these different computational workloads.” There is support for computational chemistry packages, as well as batch workloads in life sciences for large BLAST jobs, or proteomics packages. Current partners include Acclerys and NCBI.
“We want to make it seamless, very easy for end users to take advantage of clusters,” Waite continues. “We want them to use these end applications. Just as you have network resources like a shared printer, we want to have that same model apply with clusters and cluster-based supercomputing. Someone could be inside an Accelrys package like Discovery, go to the file menu and choose ‘compute’ and it would be able to hand off to the cluster on the back end seamlessly. That user would then get their results in the application.”
End Users
Mendillo says there are three principal types of customers: developers, end users and administrators. “It’s a market today that has a lot of Linux on the server side, so we offer better integrated management. We can not only manage Windows HPC servers but also Linux servers… That’s a great boon for administrators, especially as Windows momentum continues in the space and we get more apps running on our platform.”
For developers, Mendillo says that parallel development is simply too hard. “We work with about 10 million developers that use Visual Studio and other Microsoft tools, but there are only 2% capable today of writing really good, safe parallel code. We think that’s a big problem. They’re asking for easier programming models for multi-core and multi-server.”
As far as reaching the life sciences community, Mendillo says customers and end users are spending too much time mired in technology, and they want to get back to science. “They need an easier way to get started with HPC and better integration with familiar tools. We provide familiar Windows tools, including things such as Excel 2010.” Users can take an Excel workbook and run it on an HPC Server cluster and get an enormous boon in performance. He cites computations reduced from 40 hours to 2 hours. “That productivity [boost] they get is enormous,” he says.
“Obviously we’re embracing software as services and cloud computing in a very big way,” says Mendillo. “One of our tenets is that we have an on-premise and an off-premise/cloud solution. We’re hearing loud and clear from many customers that there’s some things they want to run on premise and other elements they want to run on the cloud.”
Mendillo says its important not only to consider the computational needs of the app but also how the data are being used. “Some apps will lend themselves to a cloud solution, e.g. lot of parallel computation, with small data sets… But there are other scenarios where there are huge amounts of data that are not really conducive to piping it up into the cloud and piping it back. By offering a multi on-premise/off-premise solution, we can offer best of both worlds.”
He adds that the strategy is to take the best of Azure and extend Windows HPC Server to take advantage of that computational fabric going forward.
Waite adds that it is critical to consider the types and size of the data being computer. “Some of the genomic data sets are enormous. Moving those data sets from a research facility into the cloud and back again can be difficult. We’ll be very sensitive to understanding large-scale data analytics as part of the cloud computing infrastructure.”
More information on Windows HPC Server is available at http://www.microsoft.com/hpc or the Windows HPC Server 2008 blog at http://blogs.technet.com/hpc.