Startup Lightfleet Claims To Increase Computing Speed In Servers

Startup Lightfleet Inc. says it has developed laser-based technology that moves data faster between large numbers of microprocessors in a server, boosting the processing speed of heavy-duty computing tasks found in scientific research and the financial industry.

The closely held Camas, Wash., company plans to send servers with its Corowave technology to beta testers in July, with general availability expected by the end of the year. Each server will be powered by 32 dual-core processors from Intel.

Lightfleet's technology, developed over the last four years, is attacking the delays that are inherent in the parallel processing used in today's supercomputing, which divides workloads among large numbers of processors. While the work is done simultaneously, microprocessors often have to wait for the data they need from other processors to go through switches that direct traffic. Because of the amount of data moving through the system, switches can get backed up. "Corowave does a way with the bottlenecks and congestion," Chris Kruell, vice president of marketing, said.

The technology does that by enabling microprocessors to transmit light containing data packets directly to all the processors in the system. Each processor has a receiver that reads an address contained in the packet, keeps the data the processor needs, and discards the rest. Light travels throughout the system through mirrors.

Because beams of light can travel through each other, and still reach their destination unchanged, the Lightfleet system can reflect many beams at the same time without interference. "That's a fundamental property of light that Lightfleet is taking advantage of," Kruell said.

The company didn't have any performance tests to show how much faster its technology would be than what's currently on the market. The problem, Kruell said, is the lack of benchmarks for parallel processing that Lightfleet could use for comparison. "We expect to see significant performance gains, which vary by application type," Kruell said. "Some will see 10s of percent improvement; others will see orders of magnitude."

Besides selling servers, Lightfleet hopes to license its technology to other companies. Storage vendors, for example, could find the product useful, as well as companies building devices for the medical or defense industries, Kruell said. Lightfleet servers would be targeted, for example, at organizations conducting scientific research, or financial organizations with intensive transaction processing systems. Pricing for the machines was not disclosed.

Among the future beta testers is the College of Oceanic and Atmospheric Sciences at Oregon State University. COAS's research includes data-intensive computing for modeling and simulations.

Carl Greiner, analyst for Ovum, said applications optimized for parallel computing would run best on Lightfleet's first generation servers. For the developer, there would be no change in building applications for the system. "It's agnostic to the application and protocol," Greiner said. "It's hardware."

Of course, the big question is whether Lightfleet's technology would work as advertised. "If it can do what they claim, than it's something to watch," Greiner said. "It could be fairly disruptive."