Easing Windows into High Performance Clusters

High-performance computing clusters are largely the domain of Linux. Microsoft executives have designs on those server farms. They envision Windows software powering those multi-node clusters churning in parallel on a single task.

In figuring out how to work Windows into this market, which IDC currently values at a fast-growing $9.2 billion, Microsoft has chosen not to take Linux-based clusters head on. Instead, Microsoft aims to use traditional strengths, like broad industry familiarity with its platform, deployment features and a suite of management technologies to ease Windows into less technical markets than the ones where HPCC has traditionally thrived. At the same time, the company is trying to carve a new niche by broadening the appeal from clusters of hundreds or thousands of nodes to clusters of a few dozen nodes that work on smaller-scale problems.

After a few fits and starts, Microsoft is officially in the market. Windows Compute Cluster Server 2003 was released to manufacturing in early June and will be generally available next month. With an estimated volume licensing price per node of $469, which includes both the operating system and the system software needed to run and participate in a cluster, Microsoft is hoping to have hit on a competitive price point against Linux.

For the RTM announcement, Bob Muglia, senior vice president of the Server and Tools Business at Microsoft, summed up Microsoft's business plan. “Microsoft is making HPC technology more mainstream by bringing the cost advantages, ease of use and partner ecosystem of the Windows Server platform to departments and divisions in commercial industry and the public sector. We want HPC technology to become a pervasive resource — something that’s as easy to locate and use as printers are today.”

Two of Microsoft's major early adopter customers are seizing on ease-of-use aspects of Computer Cluster Server in their shift to the product.

At Northrop Grumman's Space Technology division, the company recently upped a test network running Compute Cluster Server to 100 CPUs. The company had been running a large-scale Linux cluster, but it was maxed out on jobs and required specialized administration. While some engineers wanted to use the cluster for small- to medium-sized jobs, they would wind up hung up in complex internal scheduling procedures to get access to the cluster. Other engineers would try to set up their own clusters for advanced jobs. In either case, according to Thi Pham, a Ph.D and systems engineer for NGST, projects were taking five times as long as projected.

The company connected a Compute Cluster Server cluster to the Active Directory, and new clusters now take a week to set up rather than months, according to Pham.

Meanwhile, some ISVs are seeing value in Windows Compute Cluster Server 2003 for their applications. For example, Livermore Software Technology Corp. (LSTC) makes software that simulates the behavior of objects in complex situations. The software is used by automobile manufacturers to simulate car crashes, and also has applications for military contractors, aerospace companies and sheet metal manufacturers, among others.

The company's LS-DYNA application previously ran on 32-bit Windows Server 2003, and the company created a distributed memory version that will run on a cluster. LSTV recently ported the distributed memory version to Windows Compute Cluster Server 2003.

About the Author

Scott Bekker is editor in chief of Redmond Channel Partner magazine.