informa
/
2 MIN READ
News

Clustered Linux Cuts Down On Crash Tests

System lets DaimlerChrysler run more virtual-impact simulations, reducing demand for real-life studies.
As if the economy wasn't bad enough, more crash-test dummies will be looking for work now that DaimlerChrysler AG has implemented a clustered Linux system that can run more virtual-impact simulations than the company's current Unix systems. That means less of a demand for real-life impact studies and, hopefully, more five-star ratings from the National Highway Traffic Safety Administration.

The world's third-largest carmaker, with more than $136 billion in sales last year, revealed Monday at the Convergence 2002 conference in Detroit that it has implemented a 108-node IBM IntelliStation Linux cluster running LS-DYNA software from Livermore Software Technology Corp. to perform crash-test simulations. The cluster, which can perform 18 concurrent simulation tests, has more capacity than the three high-end Unix servers it's replacing.

"We're getting 20% better performance out of the cluster than out of our newest (high-end Unix) servers," says John Picklo, manager for high-performance computing for the company's Chrysler Group. In all, DaimlerChrysler has 20 high-end servers performing crash-test simulations, including several SGI Origin, Hewlett-Packard Superdome and IBM RS/6000 SP servers. Although Picklo wouldn't say how much the cluster cost, he says it's about 40% less than what he would have spent on a new high-end Unix server.

There are two primary benefits to virtual crash testing--an automaker can do more simulations earlier in the design process, and it doesn't have to spend money on prototype parts that will eventually be destroyed. Crash-test simulations are compute-intensive applications, Picklo says. "We'll take a whole vehicle and break it down to 700,000 pieces, like a Lego version of a car. We break each second down to microseconds to look at how each piece bends and twists and interacts with its neighbors." The computation of a fifth of a second of simulation time could take 48 hours split up across 12 processors.

Editor's Choice
Brandon Taylor, Digital Editorial Program Manager
Jessica Davis, Senior Editor
Cynthia Harvey, Freelance Journalist, InformationWeek
Terry White, Associate Chief Analyst, Omdia
John Abel, Technical Director, Google Cloud
Richard Pallardy, Freelance Writer
Cynthia Harvey, Freelance Journalist, InformationWeek
Pam Baker, Contributing Writer