Audi runs Red Hat Linux on an 18-system cluster from Linux Networx. Each box houses two 32-bit Intel 2.4-GHz processors. Sixteen of those systems are devoted to simulation calculations for castings. One system serves as a hot standby and the other system houses the management software. Audi uses LSF (Load Sharing Facility) from Platform Computing to manage the cluster. The migration was mostly plug-and-play, Saller says. Magma provided most of the updates, and Audi's internal developers tweaked the LSF user interface to work with the Linux Networx boxes in just a few weeks.
Audi leases the cluster for about half the price of the HP system, roughly 250,000 euros ($312,000) over three years. Audi did not solicit bids from HP because the software was not optimized for Unix. Thanks to clustering and optimization, the Magma application runs much faster on this configuration than on the Unix platform, Saller says. A simulation of a single component that used to take two weeks with the RISC system now takes less than two days.
"Optimization makes a huge difference," says Rob Enderle, principal analyst at the Enderle Group technology consultancy. "Think of a race car that is optimized for straightaways. Even if it's faster, a car optimized for an oval track will chew it up on the oval."
The Magma connection isn't unusual. Manufacturers devoted to a particular package of specialty software like Magma's often architect their systems around the application, Enderle says. U.S. automakers have been known to construct their systems around the requirements of CAD (computer-aided design) tools, for example. Audi had been using Linux systems for crash simulation, noise-vibration simulation and fluid dynamics, Saller says. The crash-simulation system alone requires 400 CPUs. Saller praises the cluster's performance and says it has run just as reliably and stably as the RISC system that preceded it. In fact, during a heat wave last summer, 29 legacy RISC systems failed, while the Linux Networx cluster stayed online, Saller says.