Supercomputers: Windows New Best Friend

 
 
By Peter Galli  |  Posted 2005-11-15
 
 
 

Supercomputers: Windows New Best Friend


Microsoft Corp. chairman and chief software architect Bill Gates will use his keynote address at the Supercomputing show, being held in Seattle this week, to announce the second beta for Windows Compute Cluster Server 2003, a 64-bit operating system for industry-standard x64 processors.

The first, more limited beta, was announced at Microsofts Professional Developers Conference in Los Angeles in September.

"We expect our customers to start pre-production deployment evaluations of the product with this beta," Kyril Faenov, Microsofts director for High Performance Computing, told eWEEK, adding that the team is not planning for a third beta, and a release candidate will be next.

Asked what feedback Microsoft has received from the developers who tested beta one, Faenov said it had become clear just how important it is to provide them with diagnostic tools.

"To a large degree, it is not as much the software challenges but the hardware challenges that come into play when some cluster configurations are deployed.

"So diagnostics become really important, and we have done some good thinking and made some improvements in that area. There was also a lot of focus during beta two development on optimizing the performance," he said.

It was also important for ISVs to be able to rapidly migrate their code, and Microsoft will be showcasing about 19 different applications, such as a car crash simulation guide, a weather modeling guide, as well as a demonstration of the BioTeams iNquiry software—which includes 160 open source Bio applications—on a Windows CCS cluster at its booth at the Supercomputing show this week, Faenov said.

"These companies were all able to move to Windows 64-bit if they were not already there, pick up our FDI and just run with it, and that has been very encouraging," he said, adding that the product is still on track to ship in the first half of 2006.

Asked what its strategy is towards the Linux and open-source community, given its dominance in the HPC field and the fact that Microsoft is including the Message Passing Interface—a library specification for message passing proposed as a standard by a broad-based committee of vendors, implementers and users—in the product, Faenov said Microsoft had no such strategy per se.

"Our strategy is to deliver the best value and help grow the market, and we are making very pragmatic choices based on specific requirements we hear partners or customers want. MPI was an example of this and if there are other places where this makes sense for us, our customers and/or partners, well look at them," he said.

Click here to read more about open-source code finding its way into the product.

All of the major original equipment manufacturers, like IBM, HP, Dell, NEC; those developing more personal clusters, like Orion Multisystems; as well as all the major interconnect vendors, have also announced support for the product, Faenov said.

Microsoft will release a single 64-bit only version of Compute Cluster Server 2003, which would also run on all the hardware platforms supported by Windows Server 2003 SP1, on which it is based.

To read more about Windows Server 2003 SP1, click here.

"We are going to work with our hardware partners to tailor the systems to particular market scenarios and then to do benchmarks and evidence generation for key applications," he said.

Next Page: Microsofts milestone.

Microsofts Milestone


The announcement of beta two marks a milestone for the Redmond, Wash., software maker, which will be underscored in Gates keynote, Faenov said, where he will stress the companys long-term commitment to high-performance computing as well as to a broader technical and scientific computing area.

There are three main pillars for Microsoft: business computing, consumer computing and now technical and scientific computing, an area in which there is room for many software advances, Faenov said.

Gates will also point to the need for software to help scientists and researchers focus on their work and not on the underlying IT, he said.

Microsoft officials first announced its plans to enter the HPC software market in June 2004 and said at that time that the product would be specifically designed for customers running scalable, parallel computing workloads in vertical market segments such as engineering, life sciences and finance.

Company officials such as CEO Steve Ballmer have also admitted that Linux is ahead in the area of high-performance computing, but have said Microsofts staff comes to work every day looking at how to offer customers an even better solution.

Click here to read Ballmers remarks on HPC and his vision for the midmarket segment.

Gates will also talk about Microsofts multimillion-dollar investments in, and multi-year commitment to, 10 Institutes for High Performance Computing around the globe, where it is creating funded research positions at existing facilities such as the University of Utah, the Cornell Theory Center, the University of Stuttgart, Germany, and the Tokyo Institute of Technology.

"The centers we chose have a unique blend of expertise as they do research in HPC systems and, in many cases, are closely affiliated with a HPC center, giving them a unique insight from both a scientific and industry perspective on what will actually help solve real problems," Faenov said.

The agreements with Microsoft will see the staff at these centers deploy its HPC and other products and then work with Microsoft staff to explore and maximize its product roadmap going forward, he said.

Looking to the next release of the product, Faenov said those plans remain on track.

He previously told eWEEK that the development team will also continue to invest in making clusters easier to manage, particularly as their use moves away from centralized resources and down to the workgroup level.

Microsoft also planned to work on improving the user experience through better power management and use of the IPMI (Intelligent Platform Management Interface) or Web services management to reboot clusters without having to power down, Faenov said.

"[The second version] will look at how we can control the hardware using the standard ways of rebooting. We will also look at how we can make the integration of clusters into the applications, including workstation applications, even easier. We are also looking at performance monitoring tools for the cluster space as a whole, which is a big area for development," he said previously.

Check out eWEEK.coms for Microsoft and Windows news, views and analysis.

Rocket Fuel