High Performance Computing for the Masses

By HENRIK STEEPLER, PH. D., PARTNER ACCOUNT MANAGER EMEA, MICROSOFT

Raw computing power is one of the most important assets for competitive organizations. If you can compute more efficiently than your competitors, you are likely to outcompete them by having faster access to more accurate information about the problem. In areas ranging from automotive aerodynamics to oil exploration reservoir simulation and stock portfolio management, we are witnessing a technological escalation where companies are buying more compute power to compete more effectively and efficiently — and this trend is spreading quickly across most industries. For example, I know a software company that had a turn-around time of 24 hours for its “daily” builds. By purchasing 100 compute servers, this company cut down its turnaround time to just 15 minutes — a dramatic shortening of its time-to-market.

The continual and exponential increase in CPU clock frequencies over the last 20 years has been a key contributor to the success of the IT industry because it enables software developers to leverage higher abstractions. But recently this performance increase has almost come to a stop due to limitations in physics, which is why we rarely see clock frequencies above 3 GHz. Today, the only way to increase compute power is to use parallel computing — distributing compute tasks over several CPUs, often spread over several physical computers — usually called High-Performance Computing (HPC).

The use of HPC in science and engineering is well established. Among the most widely deployed applications are simulations of different physical phenomena such as computational fluid dynamics, structural mechanics, and electromagnetics. A significant performance boost is achieved through parametric simulations, a form of simulation that allows for optimal scalability by solving the same problem in parallel for a wide range of parameter values. COMSOL Multiphysics version 4.0 supports all of these uses of HPC.

“For scientific and engineering software companies like COMSOL, the Windows HPC server represents an out-of-the-box solution.”

Microsoft entered the IT industry with the vision of one computer on every desk, a vision now more than fulfilled. In disciplines such as scientific computing or engineering, users are often double blessed by having two computers on their desk. The first computer is typically used for e-mail, reporting, and routine day-to-day work using Microsoft Office with Excel, Word, Outlook, and similar applications. The second computer is reserved for heavy computations, simulations, compilations, renderings, or similar time- and resource-intensive tasks.

There are many reasons for having two computers: You may need different operating systems for certain applications, and some applications require so many resources they cannot co-exist with other applications. Or maybe you simply want double the computing power. The key benefit of two computers is that it gives you the opportunity to work on parallel tasks, such as writing a report on your current experiment while simulating the next experiment.

From a productivity perspective however, there are no advantages to using several operating systems simultaneously — quite the opposite, in fact, because your workflow gets disturbed. While you may want to do things in parallel, you also need to be able to interact between tasks, say, cutting and pasting live data from a simulation into a report. But in this respect, each operating system places fairly strict borders around its workflow. Clearly, this model for scaling up computing power is not sustainable in the long term.

Microsoft entered the HPC market more than 4 years ago with the goal of enabling Windows desktops to offload compute-intensive tasks onto a cluster of Windows servers. This would enable engineers to stay within their Windows desktop at all times, while compute-intensive tasks like COMSOL sent all their heavy computational jobs to the cluster. The cluster, managed by the same IT staff maintaining the rest of the Windows infrastructure, would use the same identity and access technology as the email, file, and print servers. Microsoft’s vision was for all Windows desktops to be empowered by a few scalable, central resources for heavy computations, similar in concept to centralized storage systems.

For scientific and engineering software companies like COMSOL, the Windows HPC server represents an out-of-the- box solution containing everything their users need for cluster computations, including a server operating system, a job scheduler, MPI library, and management tools. Previously, such a solution had to be built with products from a number of different vendors and even open source communities, which often left your total cluster solution without complete support.

The Windows HPC server is completely based on Windows Server and, like all Windows products, comes with a 10-year support commitment. Together with the rich support for parallel development in Visual Studio, this makes the Windows platform the most efficient and economical solution for developers of compute-intense scientific software by far. For the end-user, this translates into having many affordable HPC applications such as COMSOL available in a familiar Windows desktop environment, along with an almost unlimited ability to scale-out on the commodity hardware and server operating system.

About the Author

Henrik Steepler earned his PhD in Computer Science in 1999 at Chalmers University, Sweden. Since 2003, he has been working at Microsoft on emerging markets like Security, Virtualization, and since 2007 on their High Performance Computing (HPC) initiative. He is now managing the partner network for Microsoft in Europe, the Middle East, and Africa around HPC.