SunGard

SunGard Supports 512-Core Computing Grid with SQL Server 2008 for Complex Modeling

SunGard creates solutions for 41 of the world’s top 50 banks, so it wants the best platform on the planet to support its massively scalable asset and liability management (ALM) solution—Ambit BancWare ALM. The company recently completed testing that found that the Microsoft® Application Platform, including Microsoft SQL Server® 2008 database software hosted on a server computer with four Intel® Quad-Core Xeon® 7300 series processor cores could support a computing grid created from 512 Intel Quad-Core Xeon 5400 series processor cores. Testing complex simulations with up to 20,000 variables found near-linear scalability in moving from a 128-core grid to the 512-core grid. This scalability benefits SunGard customers by enabling more complex simulation models to be processed, which helps enhance predictive accuracy. The scalability can also be used to generate predictions more quickly.

Our testing of SQL Server 2008 supporting a 512-core grid using Intel processors shows we can scale to meet our growing customer needs. Otis Bricker Senior Software Developer, SunGard Ambit BancWare

Situation

When it comes to running complex financial models, analysts will take as much computing power as they can get. To stay ahead of its customer needs, SunGard—whose solutions are used by more than 800 banks in 70 countries including 41 of the world’s top 50 banks—requires the best software running on the fastest and most robust hardware.

SunGard, which has annual revenue exceeding U.S.$5 billion, is a global leader in software and processing solutions for several industries, including financial services for which it provides a comprehensive core banking solution and best-of-breed point solutions for retail, commercial and private banking. The company’s asset and liability management (ALM) solution, Ambit BancWare ALM, provides financial institutions with modular, integrated solutions for asset/liability management, profitability, strategic planning, credit risk, operational risk, economic capital and regulatory compliance.

Ambit BancWare ALM solutions provide the analytic tools and scalable processing power to enable rapid and complex calculations for better risk management decisions. From predicting the payment performance of consumers purchasing an automobile, to calculating option-adjusted spreads, to analyzing the risk exposure of complicated financial instruments, analysts depend on Ambit BancWare ALM to help them strategically manage risk on the balance sheet.

Financial institutions are constantly modeling more intricate instruments, adopting more sophisticated balance sheet management strategies, and increasing internal reporting frequency. As a result, they regularly need their ALM systems to perform more complex calculations in less time.

Solution

During an in-depth 5-week test at Intel’s state-of-the-art data center lab in DuPont, WA, SunGard successfully completed a series of performance benchmarking tests using Microsoft® SQL Server® 2008 Enterprise (64-bit) database software running on a 512-core computing cluster, and using DataSynapse GridServer software. The testing project was a joint effort of the SunGard ALM development team, the Microsoft SQL Server Customer Advisory Team, and the Intel High Performance Team.

The lab testing, which benchmarked performance as the grid was scaled from 96 processor cores to 512 cores, was part of SunGard’s efforts to continually improve the scalability and performance of its products.

We feel we get the best price-performance benefit from basing our solutions on SQL Server running on the Windows Server operating system, and hosted on computers powered with Quad-Core Xeon processors. Jiaping Zhang Senior Software Engineer, SunGard Ambit BancWare

Increasing scalability and performance is of keen interest to SunGard and its customers because Ambit BancWare ALM provides multipath analysis of the balance sheet, incorporating interest rate risk, income simulation and market valuation using static and stochastic modeling. Generally speaking, the more variables (referred to as paths) that can be modeled, the more accurate are the predictive results. The downside is that the more paths a model includes, the longer it can take to process. Rather than wait for days or hours, analysts generally opt for running simpler models that require shorter computing time.

Previously, SunGard had worked with Intel and Microsoft in successful testing of a 128-core grid. That testing was of Ambit BancWare ALM running on the Microsoft Application Platform, including SQL Server 2005 Enterprise Edition (64-bit) and Windows Server® 2003 Enterprise Edition (64-bit), running on Intel® X5400/E7300 Quad-Core Xeon® processors.

For the new testing, using SQL Server 2008 running on Windows Server 2003 Enterprise (64-bit), the Intel lab provided a 512-core platform consisting of:

• 64 computational servers, each configured with two Quad-Core Intel Xeon 5400 series processors (2x4x64 = 512 cores). 
• 3 database servers, configured with four Intel Quad-Core Xeon 7300 series processor cores. Some testing was done on a server configured with 32 GB of RAM, other testing was done using a database server with 64 GB of RAM. Three SQL Server instances were used to isolate workloads between multiple databases used for testing and tuning. The bulk of the testing load was performed against a single instance of SQL Server 2008 (64-bit). 
• Intel 9501 10-gigabyte (GB) and Intel 82563 1-GB Ethernet adaptors with IOAT acceleration were also used to connect the computers into a grid. 
• DataSynapse GridServer software was used to optimize multi-machine scaling.

A collection of measurement tools was used during testing, including Performance Data Warehouse, a new feature of SQL Server 2008 that captures query performance of ALM tests against SQL Server instances. The team used SQL Server Dynamic Management Views to analyze query performance, SQL profiler traces, and Intel’s vTune to gather data from perfmon counters.

“Test results showed that SQL Server 2008 scaled to easily meet the demands of supporting the 512-core computing grid, and that scaling Ambit BancWare ALM to 512 cores resulted in sharply improved scalability, reduced load times, and increased overall performance,” says Jiaping Zhang, Senior Software Engineer for SunGard’s Banks and Corporations business. “We’re excited about our continuing work with Intel and Microsoft to provide our customers with highly scalable multipath solutions.”

Benefits

Running Ambit BancWare ALM on the Microsoft Application Platform, hosted on Intel Quad-Core Xeon processors, SunGard found near-linear scalability in going from a 128-core grid to 512 cores. The company found enhanced performance, and its ability to support up to 20,000 paths of simulation should help provide greater modeling accuracy. SQL Server 2008 data collector tools helped SunGard identify opportunities that enabled it to reduce memory use by 80 percent, and the company has found that working with the Microsoft Application Platform and Intel processors helps provide a low total cost of ownership.

Near-Linear Scalability

SunGard engineers were delighted that testing at the labs found near-linear scalability for Ambit BancWare ALM running on SQL Server 2008 and the Windows Server 2003 operating system hosted on servers using Intel processors. Linear scalability is important because with some solutions the additional overhead of managing extra resources diminishes scaling efforts.

As shown in Table 1, running a 5,000-path simulation on a 512-core grid requires only 4.26 minutes longer to startup, compared to a 128-core grid, while the 512-core grid reduces computation time to about 70 minutes, compared to 239 minutes on the 128-core grid.

When scaling from 5,000 paths to 20,000 paths, as shown in Table 2, the 512-core grid handles the 400 percent increase in computation with core compute time per path decreasing slightly from 7.24 minutes for 5,000 paths to 7.14 minutes for 20,000 paths.

“Testing included processing complex ALM models, including full balance sheet simulations, on 512 processing cores,” says Otis Bricker, Senior Software Developer for SunGard’s Banks and Corporations business. “Our scalability tests found we could easily scale from performing 5,000 interest rate simulation paths to 10,000 paths, and then to 20,000 interest rate simulation paths while enjoying near-linear scalability. The tests demonstrated that our newest version of Ambit BancWare ALM is five times more scalable than its previous release.”

Bricker adds: “Scaling from 128 cores to 512 cores was impressive, because we were loading data into four times as many machines, but it only took an extra minute or so. I’m confident we will be able to scale to 1,024 computing cores, which is what we will be testing next.”

80 Percent Reduction in Memory Use
Whenever SunGard engineers found performance had slowed because of resource constraints, they used data collector and other tools in SQL Server 2008 to identify and remedy the problem in the application code. This approach helped the company achieve a significant reduction in the amount of memory required on the database computer.

“One of the factors that enabled us to scale to 20,000-path solutions was that we were able to achieve an 80 percent reduction in our memory use,” says Bricker. “The reduction in memory use wasn’t directly due to SQL Server, but it was using SQL Server 2008 data collector tools that enabled us to identify some inefficiencies and change our application code to remedy them.”

“SQL Server was never the problem,” says Bricker. “We used data collector tools within SQL Server to identify, for example, when our queries were generating excessive traffic. We were able to identify how to better structure our queries to minimize traffic and CPU utilization.

Whenever SQL Server responses slowed, we used SQL Server data collector and other tools to gain the performance insights we needed to resolve the problem. SQL Server 2008 never let us down.”

Enhanced Performance
The near-linear scalability of Ambit BancWare ALM running on the Microsoft Application Platform and the 512 processing cores has dropped model processing time significantly. “Computing a 5,000-path model using 96 cores took 480 minutes,” says Bricker. “The same 5,000-path model running on 512 cores can be processed in 77 minutes. Customers who need the highest possible performance can simply add computing cores to our ALM solution.”

SunGard also saw performance increase dramatically for loading large models. “In testing we saw the graphical interface load times of large models drop from 90 seconds to 5 seconds,” says Bricker. “That’s a 1,800 percent performance improvement.”

This is why we use grid computing, Intel Quad-Core processors, and SQL Server. We need to provide the fastest computational environment we can because adding compute cycles improves model accuracy. Joe Sass Director of Product Strategy, ALM, SunGard’s Banks and Corporations

Greater Model Accuracy
The real-world benefit of near-linear scalability and enhanced performance comes down to giving customers the opportunity to run more complex models—up to 20,000 paths—using Ambit BancWare ALM. The more variables that an analyst can incorporate into a model, the more accurate will be the results.

“Complex financial models are stochastic by nature, because there are so many variables at work as our customers try to model what happens in the real world,” says Joe Sass, Director of Product Strategy, ALM, for SunGard’s Banks and Corporations business. “What if interest rates do this, what if they do that? Monte Carlo simulations are a technique that basically states ’Let’s look at a large number of possibilities,’ by estimating variables via random sampling. The price paid however is that they can require extensive processing power to derive results in a timely manner. This is why we use grid computing, Intel Quad-Core processors, and SQL Server. We need to provide the fastest computational environment we can because adding compute cycles improves model accuracy.”

“The variables can stack up in a hurry,” says Bricker. “If analysts are studying U.S. interest rates, they might have 50 variables they want to consider on the U.S. side, and another 50 variables to consider for how other currencies could impact U.S. interest rates. Multiply 50 variables by 50 variables and you have 2,500 paths to compute. Add another 10 variables and you are up to 25,000 paths. The degree to which you are able to compute what each of these paths would be, the more accurate your model will be. So there is a continuous demand for ever greater multipath modeling. Our testing of SQL Server 2008 supporting a 512-core grid using Intel processors shows we can scale to meet our growing customer needs.”

Low Total Cost of Ownership
The scalability tests showing that SQL Server 2008 running on Intel processors can scale to support 512 processing cores is good news for SunGard customers who can continue to enjoy a low total cost of ownership—from a licensing standpoint as well as from an operational one. Deploying multiple processing cores per processor slot reduces the number of computers that need to be managed and maintained.

“We feel we get the best price-performance benefit from basing our solutions on SQL Server running on the Windows Server operating system, and hosted on computers powered with Quad-Core Xeon processors,” says Zhang. “With multiple processing cores per server, it is easier to manage the computing grid. From a software perspective, we benefit and our customers benefit from Microsoft licensing agreements that are based on cost per processor—rather than processing core. Using quad-core technology we can achieve four times the processing for the same licensing cost as a single processing core. This provides a 4-to-1 benefit, which is significant when scaling up a grid computing deployment.”

 

Windows Server 2008, SQL Server 2008, and Visual Studio 2008
Windows Server 2008, SQL Server 2008, and Visual Studio 2008 provide a secure and trusted foundation for creating and running your most demanding applications. Combined, the products offer advanced security technology, developer support for the latest platforms, improved management and Web tools, flexible virtualization technology to optimize your infrastructure, and access to relevant information throughout your organization.

For more information about Windows Server 2008, go to: www.microsoft.com/windowsserver2008

For more information about SQL Server 2008, go to: www.microsoft.com/sql/2008/default.mspx

For more information about Visual Studio 2008, go to: www.microsoft.com/vstudio.

For More Information
For more information about Microsoft products and services, call the Microsoft Sales Information Center at (800) 426-9400. In Canada, call the Microsoft Canada Information Centre at (877) 568-2495. Customers who are deaf or hard-of-hearing can reach Microsoft text telephone (TTY/TDD) services at (800) 892-5234 in the United States or (905) 568-9641 in Canada. Outside the 50 United States and Canada, please contact your local Microsoft subsidiary. To access information using the World Wide Web, go to: www.microsoft.com

For more information about Intel products and services, visit the Web site at: www.intel.com

For more information about DataSynapse products and services, visit the Web site at: www.datasynapse.com

For more information about SunGard products and services, call (800) 825-2518 or visit the Web site at: www.sungard.com

 

This case study is for informational purposes only. MICROSOFT MAKES NO WARRANTIES, EXPRESS OR IMPLIED, IN THIS SUMMARY.

Document published February 2009