Since 1986 - Covering the Fastest Computers in the World and the People Who Run Them

Language Flags
September 23, 2008

Microsoft Aims Newest HPC Offering at Wall Street

by Dennis Barker

NEW YORK CITY — When profits drop, businesses look to boost productivity and performance — and nowhere is that demand more urgent right now than on Wall Street. Yesterday, about 60 blocks north of the scene of the recent financial meltdown, Microsoft announced it has released its latest product to provide that boost: Windows HPC Server 2008. The company says the new version will give firms in the financial services business a way to easily and cost-effectively deploy scalable high performance systems.

“Companies have to be more efficient than ever with IT resources, but they need to maintain their position in a competitive marketplace,” said Bill Laing, VP of Microsoft’s Windows Server and Solutions Division, during a speech at the HPC on Wall Street conference yesterday. Financial organizations are relying more and more on high performance systems for routine but critical operations such as real-time risk analysis, Laing said, but “they require HPC solutions that deploy quickly, integrate in a heterogeneous environment, and scale from workstation to cluster.”

Built on 64-bit Windows Server, the new platform essentially puts the Windows interface on top of high-speed compute clusters. “We’re providing supercomputing to the desktop guys, the financial analysts and the ones developing models — the guys doing real-time market calculations,” said Vince Mendillo, director of Microsoft’s Server & Tools Business Group, during a briefing. “Our goal is to accelerate the time to insight.”

Microsoft worked with dozens of companies in the financial industry to get their feedback to the server system, Laing said. One such company is Lloyds TSB, one of the largest banking groups in the UK. The IT director at Lloyds TSB Corporate Markets group in London, Ricky Higgins, reported that his team was able to stand up a new 64-node cluster in a very short time. “From bare hardware to first job submitted took barely overnight,” he said. “Due to market turmoil, we need to process data much more quickly than ever before.” Higgins said processing time has been cut by about 50 percent. “We’ve been able to greatly increase the number of transactions,” he said.

To speed up processing of financial workloads, Microsoft made a series of major changes to HPC Server. Instead of using Remote Installation Service to set up a cluster, HPC Server uses Windows Deployment Service, which the company says makes scaling much faster because it uses image multicasting to deploy nodes in parallel, but it’s also easier because a wizard system guides administrators through node configuration. A “to-do list” page walks the admin through the steps needed to configure a cluster, such as defining the network topology and setting up automatic deployment.

In fact, Microsoft lists simplified administration as one of the key benefits of the new platform. “Our goal was to provide efficient, scalable management tools for setting up and deploying a cluster. Reporting capabilities are very easy to use,” Mendillo said. Along with reporting, monitoring and diagnostic tools are all built into the new management console. A good example is the Heat Map, which gives an at-a-glance look at the health of the cluster.

Remote Direct Memory Access enables process-to-process communications, so “there’s very low latency when sending a process from one machine to another,” said Mendillo. Processes can write directly to the address space in another machine. The new job scheduler has been upgraded to work better with large clusters, handle more simultaneous chores, and be used in service-oriented applications, he said. The software now follows the Open Grid Forum’s HPC Basic Profile interface for interoperability with other schedulers.

Reliability is strengthened with advanced failover capabilities. “Redundancy on head nodes guarantees that the cluster will keep running, and job scheduler clients won’t see any change in the head node during the failover process,” explained Mendillo.

Microsoft has been working with independent software developers to build scalable applications for risk analysis, modeling trading, and other financial operations, as well as compilers, debuggers, performance optimizers, libraries, and other essential tools. Through integration with Visual Studio 2008, with its parallel programming environment, the new platform makes it easier to build software that takes advantage of distributed processing power, explained Medillo.

It’s not as if HPC is new in the financial district. Like Geno Valente, a VP for database analytics device developer XtremeData, said during a session yesterday, “Most people on Wall Street have some kind of HPC in the back room.”

But that system out back is often a Linux cluster. If Microsoft can’t replace that, it’ll work with it. “Interoperability with Linux systems, mixed cluster support, are essential,” Mendillo stated. “All the major file system vendors are supporting us, and our management package will allow administrators to manage Windows and Linux systems from a single console.”

Although its focus at this time of accelerated jitters is on the financial services segment, Microsoft is looking to bring high performance clusters to other industries with HPC Server. “We’re making high-end computing affordable, increasing productivity, and making it easier for companies to scale up to meet demand,” Laing said. “HPC Server will bring high performance to a much wider audience. It will eventually make it mainstream.”

You can download an evaluation copy at www.microsoft.com/hpc. Price of ownership is $475 per node.