High-Performance Profits

The cluster enables Merlin to cross-correlate about a quarter-million stock exchange price points each day. The company is pinning its reputation on being able to provide its clients with faster trading analysis than competitors. Merlin sends out daily reports at 8 p.m., while its competitors generally provide them at 6 a.m. the next day.

"There's an advantage to being able to let [hedge-fund managers] plan their strategies and have more time to make decisions prior to the next day's trade floor opening," Mettke says. "We started essentially on a greenfield with no legacy technology to worry about, but we couldn't just go with whatever the latest technology was to hit the market because it could fail, and we can't have any downtime."

Improvements in the ability to run parallel complex computing problems on cost-effective server platforms let Merlin get into the market at a relatively low technology threshold. And Dell has provided support infrastructure that fixes any problems within four hours. As the company grows, Merlin easily can add cluster nodes, Mettke says.

Seismic solution

Chevron Energy Technology Co. has evolved its energy-exploration operations in supercomputing environments since it began with an IBM mainframe in the '70s, says Peter Breunig, general manager of technical computing. Later it used Cray and SGI symmetric multiprocessing computing platforms, before moving to x86-based clusters in the past few years.

"You can't ignore the business model," Breunig says. "They're just too cheap relative to the power and performance you get out of traditional high-performance computing platforms. For the same dollar value, we can now create much more throughput from our cluster and get that much resolution in our seismic imaging."

Chevron uses internally developed software to record seismic data in areas it's considering for oil and gas drilling operations. The software uses algorithms that automatically provide acoustic wave corrections based on known velocity through depths of the earth to find viable reserves of oil and gas.

Moving its proprietary software onto clustered x86 servers running Linux has let Chevron put about twice as many processing nodes into algorithmic calculations than previously was possible, Breunig says. "With more systems running, we can increase the resolution of the output and reduce the number of approximations in the algorithms."

Chevron has 700 IBM eServer 326 systems, including a recent installation of dual-core-enabled systems using Operton processors from AMD. That's a total of more than 1,700 processing cores, all aimed at what Breunig calls the "one-trick pony seismic-imaging program."

Illustration by Stone