|
Microsoft Back on the Top 500 List of Biggest HPC Systems
Published: June 27, 2007
by Alex Woodie
Microsoft has regained two places on the Top 500 list of the world's biggest supercomputers. According to the latest list, which was released this week in conjunction with the International Supercomputing Conference, Windows Compute Cluster Server (CCS) has regained two slots on the Top 500 ranking after falling off the list last November. The company also reported headway in Windows CCS deployments, primarily among smaller, departmental-level clusters, and reiterated its high performance computing (HPC) commitment to OEMs.
Windows Compute Cluster Server 2003, which became generally available last August at a cost of $469 per node, is a version of Windows Server 2003 that includes an implementation of Argonne National Labs' Message Parsing Interface (MS-MPI), a job scheduler, and support for standard Gigabit Ethernet, InfiniBand, and other WinSock Direct-enabled network architectures. The software will only run on X64-based architectures from Intel and AMD. CCS systems are available from most of the major OEMs, including Bull, Dell, Hewlett-Packard, IBM, NEC, SGI, and others.
When Microsoft announced CCS last June, Windows CCS owned two places on the Top 500 list. However, when the Top 500 group updated the list last fall, Windows CCS was nowhere to be found. As it turned out, one of the clusters Microsoft thought was running Windows CCS was actually a dual-boot machine that ran Linux, and the owners of the cluster, National Center for Supercomputing Applications (NCSA), decided to run Linux on the powerful cluster when it submitted its cluster for validation to the Top 500 group. The other cluster fell off the list due to natural causes, namely the validation of other, more powerful machines.
Following that setback, Microsoft is now back on the Top 500 list, and has the boasting rights that come with it. Placing at number 193 on the list is a new Windows CCS-based supercomputer cluster at Mitsubishi UFJ Securities that is running simulations for the group's risk-management practice. The system is based on an IBM BladeCenter HS21 cluster with 1,792 processors, generating 6.52 teraflops (or trillions of computations per second). The other listing is a 256-node, 2,048-core cluster of Dell PowerEdge 1955 blades running Xeon 5300 quad-core processors and Cisco InfiniBand switches that generates 8.99 teraflops. This cluster, which is owned by Microsoft and installed at its datacenter at Tukwila, Washington, earned a ranking of 106 on the Top 500 list.
Validating Windows CCS implementations for the Top 500 group should be easier following the release of a new add-in for Excel 2007 that Microsoft developed to drive all the parameters and results required of the LINPACK benchmark used by the Top 500 organization. Microsoft says the Excel 2007 workbook will contain all the necessary data needed to perform the LINPACK benchmark, and will be available within 90 days at windowshpc.net/default.aspx.
While listings on the Top 500 generate positive press for Microsoft, the software giant maintains that it didn't intend Windows CCS to compete for the biggest clusters on the planet. Instead, the company designed the operating system for use on smaller, departmental clusters, says Zane Adams, director of server marketing for Microsoft. "Our focus when we released the product was to bring HPC to the masses, rather than the limited use we see in the market today, and also to drive higher productivity in the market," he says.
Just the same, Microsoft has been pleasantly surprised by the use of Windows CCS in large nodes, Adams says. "We designed our solution to hit the departmental level, but it's scalable and open enough to do the large deployments of 7,000 to 10,000 nodes," he says.
Windows CCS has enjoyed particular success in the financial services market, which is evident by the large cluster at Mitsubishi UFJ Securities and other customers, including the Markets and Investment Banking division of UniCredit Group, one of the largest banking groups in Europe. According to Microsoft, the group has applied Windows CCS to its pricing and risk-management applications, and is finding the software fast and easy to use.
"We like the idea of Microsoft .NET as a full product family, the support for diverse programming languages, the excellent interoperability with Excel, and the higher developer productivity," says Andreas Kokott, the director responsible for forming the strategy for the group's structured derivatives platform. "The idea is to have the most standardized IT environment possible so that managing the cluster is just part of the normal workflow for IT personnel. The complexity should only be in the algorithms, not in the infrastructure."
Microsoft expects Windows CCS to enjoy more success in the financial services division. Earlier this month, Microsoft released results of a survey that found about 78 percent of customers in financial services are planning to expand their clusters. What's more, 83 percent of the respondents said they were planning on evaluating Windows CSS.
Adams says these findings validate Microsoft's strategy to provide easy-to-use HPC products that will expand the use of HPC in targeted industries. "If you look at the design point for Compute Cluster Server--integration with existing environments, development tools, Active Directory, security--it just becomes a natural point that they want the HPC deployment to be managed exactly the same as the server, rather than having different islands of management," he says.
Other industries where Microsoft hopes to win Windows CCS converts include oil and gas and digital content creation (animation); drug discovery, computer-aided engineering, and fluid dynamics are also areas the company hopes to play. All of the major developers of HPC applications are now writing to Windows CCS, which is critical to the operating system making inroads into these industries, Adams says. The fact that Microsoft offers customers choice in which MPI and job schedulers to use is another critical factor, he says.
In the long-term, Microsoft's belief is that HPC computing technologies and techniques will trickle down from the large scientific, academic, and commercial applications to the desktops and laptops of billions of ordinary people, transforming the whole of computing along the way. We are on our way toward that eventuality, claimed Dr. Burton Smith, a Microsoft technical fellow, during a keynote address at the International Supercomputing Conference this week.
"Our industry and the universities must work together to reinvent not only computing, but also the computing profession," Dr. Smith said. "The coming years will fundamentally reshape software and transform the way people use and interact with computers. In order for consumers to enjoy performance improvements in the future, mass-market technology providers will have to embrace parallel computing to differentiate and compete. It's vital that software and hardware adapt to new models of computing."
In other HPC news, Microsoft is extending partnerships with two of its OEMs, including HP and IBM. The partnership with HP will see HP selling Windows CCS direct and via reseller channels as part of the HP Unified Cluster Portfolio. The portfolio is supported on HP's ProLiant, BladeSystem, and HP Cluster platforms. Details on the new partnership with IBM are expected to be released Thursday.
RELATED STORIES
SGI and Microsoft Partner on Windows Supercomputer Clusters
Details Emerge in Microsoft's Top 500 HPC Setback
Windows CCS Falls Off Top 500 Supercomputer List
Windows Compute Cluster Server 2003 Goes GA
Windows Compute Cluster Server is Ready, Microsoft Says
Post this story to del.icio.us
Post this story to Digg
Post this story to Slashdot
|