Financial firms can now deploy clusters of servers running Windows Server 2003 to benefit from supercomputing power at a fraction of the cost.  Kyril Faenov, Microsoft’s HPC product unit director explains how.

Some of the toughest computing challenges in the world are tackled and often solved by using high-performance computing (HPC) clusters. Companies can now cost efficiently deploy clusters of servers running Windows Server 2003 to benefit from supercomputing power.

To learn how, we talked to Kyril Faenov, director of the high-performance computing product unit in the Windows Server group, who leads the HPC product strategy and implementation at Microsoft.

Kyril, FSOs face enormous challenges as an industry. How do you see HPC helping?

Recent events have increased uncertainty, risk and regulatory requirements in many sectors, particularly the financial area. These challenges put conflicting pressures on FSOs to take more information into account, model more scenarios, look out further, disclose more information and potential problems, while reducing overall costs and offering higher quality service in a viciously competitive environment.  

This requires greater data storage capacity, faster and more complex analysis, speed of development for new analytical models and lower cost of management. HPC addresses these needs allowing you to tackle storage scalability, computational scalability, management scalability and development scalability.

What’s driving growth?

There are six driving forces, the first being integration with the desktop. Web services are becoming the standard way for integrating Excel, S+, Matlab and other desktop visualization and analysis tools with the back-end infrastructure. With Excel the primary way in which many analysts interact with their models and data, its integration with the computing infrastructure solves end-to-end customer problems.

The second driver is affordable 64-bit computing. Mainstream servers and workstations can be aggregated into powerful clusters to tackle problems that once required specialized supercomputers.

Third, inexpensive data storage.     

Number four, advances in latency and bandwidth performance, and most importantly price/performance. Cost-effective Gigabit Ethernet networks are becoming backbones of cluster computing and upcoming 10 Gigabit and RDMA technologies bringing system bus-like performance to distributed systems.

Number five, the emergence of a suite of well-established commercial applications.

And then there’s integrated distributed systems, now that enterprise software vendors like Microsoft solved problems, such as secure access, management and programming, and packaged them into integrated, industry-standard solutions.

How does Microsoft’s platform provide customers with value in managing large amounts of data?  

It’s an issue of integration. Most financial HPC starts with Excel. Microsoft protects those investments and skill sets while allowing you to scale up analysis from the PC to the Grid in minutes, not days or weeks. There’s integrated security through single and extensive security architecture and single sign-on, from the desktop to the Grid. There’s integrated Web services. Our tools expose and consume Web services natively.

For integrated manageability, tools like Software Update Server (SUS), SMS and MOM behave in a uniform way from the workstation, to the work group, the Grid and the data center. There’s also industry-leading value. Since so much infrastructure is built in, not added on, most benchmarks, Windows Grids will provide the best performance, price-performance and long term total value in the industry.

How would you characterize Microsoft’s commitment to HPC?

Since the late 90s Microsoft has invested in products and solutions that enable enterprise-scale distributed computing, including Active Directory, Web services, BizTalk, SQL Server, Datacenter Server, Network Load Balancing, Storage Server, Dynamic Data Center and MOM.
 
We’ve also made a strategic decision to invest in products and solutions that directly enable what has been a more traditional domain of HPC – parallel, data-intensive applications. The newly formed HPC product unit will work closely with the OEM, IHV and ISV partners and product teams within Microsoft to deliver a complete HPC platform featuring price/performance and ease of use.  

How do you feel Microsoft’s commitment stacks up with other major players?

We’ve pioneered work with IBM on driving Web services standards. Windows 2000 was the first platform to integrate Kerberos security mechanisms as its base infrastructure. Visual Studio is the most popular development environment for enterprise application development.  

As we engage with the HPC community on enabling parallel scientific applications, we’ll learn from customers and partners about the requirements and pain points. We’ll support de-facto standard parallel programming models and runtimes, such as OpenMP and MPI, and offer value propositions with our development tools and distributed security infrastructure.

Is there real synergy between Web services and HPC?

As I said earlier, HPC’s about achieving scale as cost-effectively and efficiently as possible. Web services is the answer to the complexity of enterprise distributed computing that arose from the lessons all of us learned by watching the Internet scale to adapt to the exponential growth.

What areas will you focus on?

We’ll use Web services as the standard mechanism for discovery and communication with the computational resources available to the HPC applications. This includes Excel clients using Web services to discover and submit computational jobs to cluster server farms, job schedulers discovering server and utilizing workstation resources that can be used for processing incoming job requests’ and job schedulers exposing cluster resources to the enterprise-wide and Internet-wide Grids.  Another scenario is using Web services as front ends for very large data repositories, and providing intelligent query services to the clients instead of shipping terabytes of raw data across WANs.

You’re conducting Web services pilots that speak to the issue of ease in HPC deployments.  What can you share?

An excellent example of the use of Windows-based Grids for financial analytics is our work with Société Générale and Platform Software to make advanced analytics part of the ‘production’ system for SG's commodities trading business. Their objective is to address increasingly strategic and time-critical business issues in trading and risk management. Over 800,000 data points are included in this application.  Results to-date include improved P&L reporting, improved real-time risk management and excellent operational uptime.

Editor’s Note: Please see www.microsoft.com/windowsserver2003/hpc and go to the Case Studies section.

What does the future look like for HPC?

HPC applications and scenarios are becoming more automated, easier to use, integrated into the analyst workflow, with performance that allows for true interactivity.  In the time it takes to refresh a screen, geographically distributed data sources will be securely queried for results without users having to be aware of their location or supply their credentials. Clusters of servers and desktops will be used to process the results seamlessly as if they were a single computing resource. And results will be visualized or inserted into a model ready for interaction with the user. You’ll also be able to stitch best-of-breed computational models residing anywhere within an enterprise or the Internet into complex, multi-stage simulations. Complex parallel, distributed and data-driven computations that underlie these applications will be easy to develop, test and deploy in production.  

For more information, please visit www.microsoft.com/windowsserver2003/hpc.