Daimler Maps Grid Savings

PHILADELPHIA -- Daimler-Chrysler is testing a new workstation-based grid system for running crash tests, which could result in massive savings compared to using an expensive cluster of servers.

Speaking at the Gt04 conference here today, Steffen Neumann, project manager for consulting and partnerships at Daimler Chrysler Research, explained that the company has harnessed workstation processing power to support its crash test research in Palo Alto, Calif. The workstations are being used to run the car manufacturer’s key LS-DYNA crash test application.

Currently, crash tests consume up to 70 percent of the computing resources in Daimler-Chrysler’s high-performance computing data center which uses a range of HP/UX server clusters.

Such is the pressure on car manufacturers to boost both safety and productivity that General Motors Corp. recently installed a gigantic IBM Corp. (NYSE: IBM)supercomputer at its Michigan data center to slash the time spent processing data from expensive crash tests (see IBM Speeds GM Crash Tests).

It’s not surprising, then, that Daimler-Chrysler was so keen to find a cost-effective way of running these tests. But why build a workstation-based grid? Well, the IBM ThinkPads deployed in the grid were already on-site, and, according to Neumann, were not being used on weekends and evenings.Neumann says that this contrasts with the traditional approach of employing a number of servers in a cluster, and expects to see some real cost savings. “It will help us avoid other investments in high-performance computing data servers.”

But just how much is Daimler-Chrysler expecting to save? Neumann was unwilling to provide a specific ROI figure at this early stage of the project, only saying he expects “significant cost savings,” although other sources close to the project told us that this could potentially run into millions of dollars. Neumann did confirm that a traditional server-based cluster for crash testing could cost up to $5 million.

The 6-gigaflop grid consists of two 12-node clusters based around Intel Centrino processors. One cluster is Unix-based and the other Windows-based, and 100-Mbit/s Ethernet is used to connect the grid. This is then supported by middleware from United Devices and a single dedicated server.

— James Rogers, Site Editor, Next-gen Data Center Forum