Digipede Technologies is a Microsoft Gold Certified Partner with the only grid computing solution built on .NET. We recently caught up with their president, John Powers, at the 5th Annual Financial Developers Conference. Because financial services is Digipede’s number one vertical market, we asked John for an update on the state of the market for HPC and grid computing in finance.

WFS: What’s driving adoption of grid computing today?

JP: There are both new and old trends at work here. For more than three decades, financial services companies sought competitive advantage through the application of high-performance computing to computational finance problems. As technology and skill sets improved, the application of advanced computational finance algorithms moved from research to risk management to near-real-time trading analytics. Moving complex calculations closer to each trade, at the same time that the number of instruments and volume of trades are rapidly increasing, has resulted in a rate of increase in computing requirements that far exceeds Moore’s Law.

WFS: So HPC has increasingly meant not just faster processors…

JP: …but more processors put to work simultaneously on hard problems.

WFS: Is that unique to finance, or are you seeing this in other fields?

JP: Financial services stand out with the most immediate link between computing horsepower and profitability.

WFS: Because?

JP: Superior pricing analysis and risk management, delivered faster than the competition, leads to superior trades.

In the heart of every asset management company is someone with a better idea of what makes a good investment, and what makes a good trade. Turning those ideas into actionable algorithms while managing exposure to risk is a core technology challenge for thousands of firms today. This challenge increases exponentially as the number of assets traded, the volume of trading and the volume of accessible market information grows at the same time as latency between analysis and trading shrinks.

WFS: How do you see financial services companies addressing those needs?

JP: For at least the past decade, they’ve turned to distributed computing, combining the power of many computers to increase the performance of critical applications. High-performance computing with clusters and grids has been an important initiative at many leading firms, although these first-generation distributed computing systems have been quite difficult to deploy and use. In particular, it’s been very difficult to adapt applications to grids and clusters, requiring development approaches that recall the days of low-productivity procedural programming.

For high-value applications, the benefits of increased performance are so great that they outweigh the costs and pain and suffering involved in adapting applications to these platforms.

WFS: Have there been successes?

JP: A few. Some of these grids have grown to many thousands of compute nodes. However, no financial services grid project has yet reached the promised potential, and the primary reason is the difficulty in adapting applications to the grid.

WFS: That’s not a very rosy assessment of the current state of HPC and grid offerings. What else are you seeing?

JP: Actually, it’s worse than that. The problems with adapting applications to a cluster or grid are now beginning to appear even at the level of applications written for a single machine. While demand for computing power is increasing rapidly, the supply of computing power is changing rapidly. In 2005, increasing power requirements and consequent heat challenges led chip manufacturers to stop increasing clock speeds – actually, in some cases, reducing them, focusing instead on packing more processors or cores on a single chip. Dual core and quad core chips are now mainstream. Intel demonstrated working prototypes of an 80-core chip. Within a few years, we can expect hundreds or thousands of cores on a single chip.

While total processing power of a single chip continues to increase at approximately the same rate as for the past 40 years, doubling every 12 to 18 months, accessing that power now requires applications that are aware of multiple processors. This is a huge change. In the past, a program written two or three years earlier would just automatically run faster when installed on a new computer. Last year, for the first time, a program written in 2004 installed on a new computer might very well run more slowly. Only applications written specifically to take advantage of multiple cores would run faster.

So the big trends, from the datacenter to the desktop, point to more cores per chip, more chips per machine and more machines per grid.

WFS: How are users adapting to these trends?

JP: Unfortunately, this is where it gets worse. As the need for distributed computing has increased and the difficulty of adapting applications to first-generation distributed computing products has become evident, demand for developers trained in obscure distributed computing paradigms has outstripped the supply. One recent estimate is that a programmer trained in implementing applications on commercial grid products costs about £1000 per day in London.

WFS: And in New York?

JP: It’s comparable, and purely a function of the needless complexity embedded in distributed computing programming models required to date.

WFS: Are Microsoft’s efforts having an impact?

JP: Mainstream developers of financial analysis applications are skilled in object-oriented programming techniques. Over the past several years, Microsoft has done a great job of improving the development environment and tools via .NET and Visual Studio. The transition from object-oriented enterprise programming in the familiar Microsoft tools to parallel programming for most grid offerings is a high hurdle indeed.

WFS: Where does that leave Microsoft and Digipede in this market?

JP: The clear need in the market today is for distributed computing solutions that take advantage of the skill sets already in place in the developer community. Vastly more cost-effective than retraining hundreds of thousands of developers to learn exotic, low-productivity parallel programming techniques is to create distributed computing tools and techniques leveraging the object-oriented development paradigm already familiar to mainstream financial developers.

With the release of Microsoft Windows Compute Cluster Server 2003 last year, Microsoft has overcome competitive issues in license cost and deployment. Now it’s time to capitalize on key advantages on the Microsoft platform. Microsoft and its partners have the unique advantage of a huge developer base already trained in .NET and Visual Studio. We’ve focused on bringing the benefits of distributed computing to mainstream enterprise developers through the familiar object-oriented paradigm in the familiar Microsoft tools.

Our financial services customers tell us that the value in such solutions is in simplicity, not merely scalability. Banks, asset managers, hedge funds and insurance companies can now achieve the benefits of distributed computing without the wholesale retraining of their entire development staff to roll out new strategies and investment products and begin trading new asset classes in days or weeks, instead of months or years.

WFS: Can you cite an example?

JP: A large investment advisor managing direct investments called us. Their internal tools required an increasing amount of developer and IT staff time to operate and maintain. Adapting new apps to these distributed tools was a laborious process. We started with a single app POC study on a 10-machine network. Their lead solutions architect budgeted a month for the analysis. It was done in under a week. Installation was straightforward. Our Framework SDK made grid enabling our applications far simpler than anticipated. We demonstrated near-linear scalability on a critical application with just a few lines of code. They’ve now implemented a much larger grid, adding new applications every few weeks.

WFS: Another example?

JP: We’re working with one of the country’s largest mortgage originators who started a new commercial desk last year, and in pricing and trading loans they do a lot of heavy lifting. One of their quants had a 17-hour-long spreadsheet as part of his analysis. They got a proof-of-concept system deployed quickly, running on just 10 CPUs. With just a few lines of grid-specific code, they got near-linear performance improvements, reducing that 17-hour job to two hours. They’re running their ‘more than overnight’ jobs every day, then several times intraday, literally changing the way they work and trade. They recently began a much larger deployment.

WFS: The future looks good?

JP: When it’s distributed.