The days when only the largest manufacturers could afford supercomputers based on proprietary hardware are giving way to an era of lower-cost high performance computing (HPC) clusters built on x86 processors. Suffice to say, it isn't quite as simple as buying standard hardware and plugging it in. With clusters, a linked group of computers share processing horsepower.

By Roberto Michel, senior contributing editor (robertomichel@charter.net)

Share

The days when only the largest manufacturers could afford supercomputers based on proprietary hardware are giving way to an era of lower-cost high performance computing (HPC) clusters built on x86 processors. Suffice to say, it isn't quite as simple as buying standard hardware and plugging it in.

With clusters, a linked group of computers share processing horsepower. In manufacturing, product development teams use HPC clusters to model the effects of design variations. But increasingly, these solutions run on standard platforms. According to Framingham, Mass.-based analyst firm IDC , 65 percent of HPC systems sold in the first half of 2006 ran on Linux, while during the last few years, x86 processors have come to dominate the market.

“It's really these microprocessors that are driving the whole market right now,” says Dave Morton, CTO with Linux Networx , a vendor of Linux-based supercomputer clusters that run on x86 chips. “They have such a high level of performance, and they are cheap. So by starting with these low-cost computing elements, we have the basis for powerful systems.”

The trick is arriving at a complete solution. According to Morton, Linux Networx competes with more generic “white box” cluster vendors by bundling in a complete software stack, including cluster management software.

Morton cites these other value-adds:

• A proven implementation of standard interconnect protocols such as Infiniband and Myrinet.

• Math and communication software “libraries” that accelerate cluster performance.

• Expertise in optimizing clusters to run third-party applications, including computational fluid dynamics (CFD) and finite element analysis packages.

Digital Technology Laboratory (DTL), a provider of engineering services to its parent company, machine tools manufacturer Mori Seiki , is using a Linux-based HPC cluster from Linux Networx. DTL will use the supercomputer to accelerate the performance of engineering software, including the Fluent CFD package from Ansys .

The 13-node system, which features dual-core Intel processors, is expected to significantly outperform the previous platform DTL used for CFD, according to Zach Piner, director of mechanical engineering with DTL. In initial testing of the 13-node supercomputer, its performance was up to 30 times faster than the previous PC platform, says Piner.

The increased performance comes from the HPC solution's ability to make use of clustering and parallel processing between nodes, though the new platform's 64-byte capabilities also is a factor with larger simulations, says Piner. For example, the supercomputer is well suited to simulating the effect of room-temperature changes on machine tool performance.

Besides engineering software, other supercomputing uses in manufacturing include financial risk analysis and simulations related to business process optimization, says Bjorn Andersson, director of HPC for Sun Microsystems . Sun's HPC clusters can be configured with x86 processors from Intel or AMD, and support Linux or Sun's Solaris operating systems.

Andersson also believes HPC is more accessible thanks to commodity components, and he too sees a need for bundling the necessary hardware, software, and support.

“Many companies tried putting together clustered HPC systems on their own at least once, but they don't want to try it again because of the hidden costs of figuring out how to put it all together and manage it.”