Intel trains SunGard’s software engineers to develop parallel programs and multi-threaded applications on multi-core, multi-processor platforms

Joe Sass, director of product strategy for SunGard BancWare ALM* and Ken Roller, global account manager at Intel for the SunGard Group, talk to Intelligence in Finance about recent collaboration.

There’s nothing quite like a financial crisis for driving banks to look for ultra performance from their systems, and America’s sub-prime meltdown has done just that.

SunGard’s BancWare ALM* (asset and liability management) application is used by the world’s largest commercial banks to manage their balance sheets, forecast net interest income, determine the economic value of their institution, and run sensitivity and scenario analyses. SunGard’s clients typically want to run a number of “what-if” analytics, says Joe Sass, director of product strategy for SunGard Bancware ALM. What if the Fed changes rates on this date, or what if interest rates change and spreads compress in this manner?

By working closely with Intel, SunGard has harnessed the power of new multi-core processors to not just meet, but leap ahead of the fast-growing demands of its most sophisticated customers.

“The client might say they will choose a simulation model and generate 1,000, 10,000, or 20,000 possible interest rate paths and then look at the resulting distribution of income or return measures and see what their risk exposure is,” explains Sass. “When you start talking about thousands of simulations, that is extremely computationally intensive.”

Meanwhile, the data isn’t standing still. A single simulation will model every asset class - loan, deposit, bond, and so on.

“Then we are going to age those positions through time and layer on new business because the bank is a living, breathing entity. It’s not closing up shop once we run the simulation, but it is making new loans while its customers are withdrawing money and opening new checking accounts. The bank’s balance sheet is dynamic, and simulating all of those positions requires millions of calculations. Then multiply it through thousands of interest rate simulations. That’s a fairly arduous activity and performing it within a timeframe so as to provide actionable intelligence is a pretty daunting—but readily accepted—challenge for us as a vendor.”

The difference between half a second and a quarter of a second for a calculation becomes pretty significant when you’re running millions of them, he adds.

“We need to get as many horses on the problem as we possibly can, and that is at the heart of what we are doing with Intel—to scale ALM as linearly as possible with no upper limit. If our customers want faster performance, they add more cores with each added-core being as efficient as the last one.”

Multi-core processors present a challenge to developers who have grown up in a world of single cores where operations run in sequence, says Ken Roller, Intel’s account manager for SunGard. Over the last two years Intel has held two-day lab sessions for SunGard software engineers in the U.S. and Europe to train them on Intel’s Software Development Suite using optimizations tools such as VTune™ and showing how to thread their applications to take advantage of multi-core and 64-bit processors.

“The SunGard developers’ coding education occurred years ago when multi-core processors and grid/HPC implementations didn’t exist,” says Roller. They are picking up the new skills quickly, he adds, and significantly reducing the time it takes to run their key processes and operations.

Achieving high speed and throughput requires more than just processor power. Intel and SunGard have also worked with Microsoft to speed the flow of data in and out of the Microsoft SQL Server* database and with DataSynapse* on optimizing parallel processing for a grid/HPC implementation.

Expanding the number of cores on which to run the SunGard ALM* application has been an iterative process, he adds. “Last year, we ran the software across 256 cores for the first time, which directly led to banks purchasing the software for their balance sheet risk management solution. Recently, SunGard has concluded a five-week test at Intel’s lab in Dupont, WA, where it ran across 512 cores. This software will be available to SunGard’s customers in October 2008. And our work is not stopping there: a 1,024-core test is on the horizon.”

Behind the technical accomplishment is a new usage model that provides major business value to the financial services industry. Banks that use BancWare’s ALM* application can now target a specific computation time. With more cores, a large bank can move from 1,000 paths to 5,000 or more in their scenario testing.

“What would happen if you could run 20,000 scenarios?” asks Roller. Running more scenarios in minutes or hours changes the way banks can use the software.

SunGard’s Sass says a large multi-national bank client is working to begin testing on a large-core system in a grid environment for its treasury operations.

“The typical arrangement is to run ALM in a batch overnight, but as we optimize our application for greater performance and scalability, we are shooting to get this as close to ‘real time’ as possible.

“Intel has provided technical expertise, labs, and early releases of their next-generation multi-core platforms so we can optimize our application to harness the power of the latest technologies, which will provide our customers with a competitive advantage.” he adds.

“Within our development process, we have Intel engineers as part of the team, teaching our engineers on how to use the Intel profiling tools and offering guidance.” One of the results of our latest collaboration has been an 1800% reduction in loading time for our application.

The scale of running value-at-risk (VAR) on a trading desk is comparatively simple, says Sass. “They use Monte Carlo* as well, but the context is usually a single position or a portfolio of them. We are working with the entire balance sheet. And while they might look at 10- or 20-day VAR, the simulation horizon is years and certainly the next 12 months for the entire balance sheet. These are big, complex problems we are tackling, and we are excited that we were able to run 20,000 full balance sheet simulations in just under five hours for one of the world largest multinational banks using the 512 cores.”

SunGard and Intel have signed an optimization licensing agreement (OLA) so that Intel engineers can study the source code and suggest ways to optimize ALM for greater performance and scalability and to realize new usage models for their customers. By working with the actual source code, Intel engineers can use its sophisticated software tools to zero in on snippets of code to locate delays in the processing.

“Once our engineers are educated to the business problems that SunGard BancWare is trying to solve, they work closely with BancWare’s software engineers to better understand their current architecture and benchmarks and identify their targeted future performance and benchmark milestones.

“I have to give credit to SunGard: they had the vision to start redesigning their applications for multi-core,” adds Roller. “A lot of companies don’t have the vision to work on both today and tomorrow. Companies like SunGard that allocated resources for people to work with Intel on tomorrow have to be applauded.”

Roller says SunGard is an example of what can be accomplished by a company that commits resources to multi-core and HPC architectures. Many firms dread dealing with their legacy code, which means they risk losing to competitors who write to multi-core processors. It’s like the television commercial that says you can “either pay me now or pay me later.” End users and ISVs will need to optimize their applications for multi-core processors and platforms or their applications performance will be impacted, which will lead to lost business.

*Other names and brands may be claimed as the property of others.

(1 votes, average: 5 out of 5)  Loading ...

Filed under: Issue 6 - Autumn 08