Windows Ekes Out a Presence in Supercomputing

The bi-annual release of the Top 500
supercomputer list is an opportunity for a flurry of press releases as every company crows about its bragging rights: HP had the most machines on the list, IBM held most of the top spots, Intel chips were in almost 400 of the machines and AMD had the top two spots.

A quieter debut, without much fanfare, was the No. 10 spot, featuring the first appearance of a supercomputer from China’s Shanghai Supercomputer Center.

The Dawning 5000A is a Quad-Core Opteron machine with Infiniband connectors and 12 terabytes of memory, power by Microsoft Windows HPC 2008.

That makes for a total of four Windows HPC 2008 systems and one Windows Computer Cluster 2003 machine on the list. OK, so Microsoft (NASDAQ: MSFT) is only at one percent of the operating systems on the list, while Linux is on 389 (77 percent) of the machines.

But for Microsoft, recently entered high performance computing (HPC), it’s a start.

The company officially kicked off its efforts with a Bill Gates keynote at Supercomputer ’05, which was held in Seattle. Gates proved prescient once again, predicting that there would eventually be “supercomputers of all sizes,” while one of his slides referred to it as the “rise of the personal supercomputer.”

Cray
and nVidia have since proven him correct in that regard.

During his presentation, Gates announced a new version of Windows designed for HPC, and called Compute Cluster Server 2003. It shipped the following summer, and was targeted at modest clusters, from 32 to 64 nodes. But almost immediately, the company began aiming higher: “We found, in our first year, people wanted to go much bigger,” Jeff Wierer, group product manager for Windows HPC, told InternetNews.com.

“Our motivation is where requests have been coming from,” he said. “There’s a certain set of users with Windows base skill and want to be able to use it.”

For that reason, the company thinks it can get into HPC not as a displacement for Linux, but as a co-existent partner. “The reality will be people alternating between operating systems out there,” Wierer said. “Even if you’ve settled on using Linux, there will be times when those apps have overriding stacks. So this is a complement to existing strategies to run Windows for the code that can only run on Microsoft’s stack.”

Windows Computer Cluster 2003 had been derived from the Server 2003 line, so when Server 2008 was released, Microsoft was able to update its HPC offering as well. HPC Server 2008 is built on the basic Server technology with HPC-oriented technologies added to it — including components like message passing, scheduling and a high-speed network fabric.

Wierer said he would like to see parallel extensions used in HPC computing brought to the .NET Framework. Anyone who can program multi-core apps should be able to bring that to HPC, he argued.

“These allow you to do common patterns of parallelism across multiple cores. It would allow any
.NET developer to be able to take advantage of all those cores on a single workstation and then extend that to high-performance computing.”

The same applies for system management. Microsoft introduced with HPC 2008 the ability to manage an HPC system with many nodes the same way a single node computer would be managed. “We think the technology you use to manage a Windows server shouldn’t be any different for the tech you use to
manage a high performance cluster,” Wierer said.

Get the Free Newsletter!

Subscribe to our newsletter.

Subscribe to Daily Tech Insider for top news, trends & analysis

News Around the Web