In rigorous testing for scalability, the combined platform performed 1.8 million reads per second and 1.1 million writes per second, with sub-one-millisecond latency, and also topped 16,000 page generations per second with six-millisecond latency over a local area network.

"With the GigaSpaces XAP running on Sun's Java HotSpot(TM) Virtual Machine (VM), application architects do not have to make a choice between distributed and vertically scaled architectures; that is, between scaling out and scaling up," said Ambreesh Khanna, Global Head of Financial Services for Sun Microsystems. GigaSpaces XAP enables applications to take advantage of the scalability of multiprocessing while realizing the benefits of a distributed architecture."

Achieved in testing with the full Monte Carlo Value at Risk method of risk simulations for trading portfolios, these results demonstrate that significant performance enhancements are possible using High Performance Computing (HPC) as the infrastructure is scaled. Traders and analysts can now cost-effectively perform full-risk simulations on their portfolios in mere minutes, instead of hours, and run pre-trade scenarios in milliseconds.

"This benchmark underlines how we can collaborate to innovate and drive real performance and total cost of ownership benefits for our clients," said Nigel Woodward, Global Director, Financial Services for Intel. "All banks are looking to understand their risks and exposures within a couple of hours, and real-time is the ultimate objective. These benchmarks demonstrate that customers can cost-effectively perform full risk simulations for their trading portfolios in minutes instead of hours, and pre-trade what-ifs in milli-seconds."

The tests were conducted on Sun Fire x4450 servers using Intel(R) Xeon(R) 7460 processors with six cores per CPU running Solaris(TM) 10 Operating System and Sun's Java HotSpot VM and using GigaSpaces XAP as the middleware software. This proved to be a scalable solution that provides both low latency and high throughput in HPC-based application environments. The tests confirmed the exceptional stability of the Solaris 10 OS and Java 6 Java Virtual Machine, even in the most demanding scenarios.

"The results are extraordinary," according to Shay Hassidim, Deputy Chief Technology Officer of GigaSpaces. "We found that the time to complete the analysis dropped linearly as new computation nodes were added, and we found evidence that this would continue up to extremely high numbers of nodes."

The mission-critical application benchmark measured the computational performance of a risk management system as processors and nodes were added to the grid. The load of performing the calculations was distributed across an increasing number of compute nodes, added dynamically without any modifications to the Gigaspaces application. The overall time to calculate the scenarios fell linearly as these additional nodes were added. In addition to this risk management benchmark, the tests also included a mix of high-throughput scenarios typical in real-time analytics and on-line financial services application. As the nodes were added, the time taken to recalculate the portfolio fell from 2560 seconds with one computation node to 70 seconds with 64 computation nodes.