Start-up promises unlimited computing power

A handful of former Nortel executives have launched a company they say combines telecommunications-class networking with high-performance computing to create a nearly unlimited pool of server resources that can grow and shrink in response to application demands.

Called Liquid Computing, the company is making a modular server system that uses a high-bandwidth fiber interconnect to let users shift workloads and reconfigure systems on demand. The company is targeting the system, which is scheduled to begin trials this summer, at high-performance computing customers in areas such as biosciences and government. It is expected to be generally available next year.

The system is made up of a 36.75-inch-high chassis that holds 20 modules - each consisting of four dual-core AMD Opteron processors, memory and I/O - that plug into a common midplane. Multiple chassis can be tied together via fiber-optics cables, and the initial release will scale to 12 chassis, or 960 processors, says Brian Hurley, Liquid Computing’s CEO.

To continue reading, register here and become an Insider. You'll get free access to premium content from CIO, Computerworld, CSO, InfoWorld, and Network World. See more Insider content or sign in.

A handful of former Nortel executives have launched a company they say combines telecommunications-class networking with high-performance computing to create a nearly unlimited pool of server resources that can grow and shrink in response to application demands.

Called Liquid Computing, the company is making a modular server system that uses a high-bandwidth fiber interconnect to let users shift workloads and reconfigure systems on demand. The company is targeting the system, which is scheduled to begin trials this summer, at high-performance computing customers in areas such as biosciences and government. It is expected to be generally available next year.

The system is made up of a 36.75-inch-high chassis that holds 20 modules - each consisting of four dual-core AMD Opteron processors, memory and I/O - that plug into a common midplane. Multiple chassis can be tied together via fiber-optics cables, and the initial release will scale to 12 chassis, or 960 processors, says Brian Hurley, Liquid Computing’s CEO.

The interconnect is Liquid Computing’s key feature. It is a combination of commodity components and proprietary devices that enable it to have latency of  “less than 2 microseconds from processor to processor” and throughput of up to 6G bytes per second over fiber-optic lines, Hurley says. He stresses that the proprietary nature of the interconnect is “transparent” to applications. “It’s all standard interfaces,” he says.

“With a simple software command, we can configure the processors to look like a bunch of two-ways, a bunch of four-ways, a bunch of eight-ways - with any combination of memory and I/O that’s required by the application at that point in time,” Hurley says.

While the company is targeting the HPC market initially, Hurley says the technology also will be useful in enterprise data centers.

“We’re really doing for computing what SAN and NAS have done for storage in context of allowing an aggregation of resources that gives you economies of scale relative to manageability, high availability services and flexibility in how the resources are used,” Hurley says.

While the debut systems will be built on Opteron processors, Hurley says it is really “processor agnostic.” Intel Xeon processors could also be used, for example. “We can mix and match processor types within the platform without issue,” Hurley says.

Analysts say the idea is similar to today’s blade servers but takes things a step further.

“They’re including things like broadband connectivity as well as a little bit more advanced networking capabilities than your run-of-the-mill server blade,” says Alan Freedman, research manager of infrastructure hardware at IDC.

Digital Globe, a multimedia firm that sells high-resolution satellite images of Earth, says Liquid Computing will enable it to get more efficient use of data center resources.

Today, the Longmont, Colo., company has idle hardware standing by to provide the huge amount of computing power it needs for just 10 or 15 minutes every hour and a half, the time it takes for its satellite to circle the Earth.