CG Scooby stars seamlessly next to his human co-stars.
San Francisco, CA—Engineering software continues to migrate off the heavy workstation. Seeking larger markets, software providers have been writing "cheap and cheerful" versions of CAD, FEA, and CFD to run on lightweight desktop PCs.
In general, that means software has moved off the UNIX OS onto a user-friendly Windows interface, making it available to average design engineers, not just the specialists in the back room. And that in turn has lead to the rise of PLM (Product Lifecycle Management), a business philosophy that engineering software should act as an umbrella over all stages of the company, allowing design data to be used by management, sales, marketing, maintenance, assembly, and purchasing, as well as the engineers who created it.
Sounds great, but there's one drawback—many engineering problems still demand vast computing resources to solve. Problems in analysis, simulation, visualization, and animation can typically run all night, slowing the design cycle and making it tough for engineers to run alternate scenarios.
In this age of the Internet, the answer is obvious—spread the problem around. Whether you call it a compute-farm, enterprise, network, cluster, or grid, the general process is the same. Split your calculations into smaller chunks, then assign those jobs to a variety of linked processors. It's the same approach that SETI (the search for extraterrestrial intelligence) takes to filter through the gigabytes of radio signals it collects each day—then parcels out to unused PCs around the world as their owners sleep.
The most simple application of this principle is in a dual-processor server or workstation. But the concept quickly scales up. In Germany, DaimlerChrysler runs crash simulations for Mercedes-Benz cars on an enormous Linux cluster, marshalling the strength of several hundred AMD processors working in concert. "Shortened product development processes made possible by gains in computer aided engineering require increasingly demanding simulations that must be carried out in less time," says Johannes Lugisland, DaimlerChrysler's manager of IT infrastructure for safety and comfort simulation. "This demands the appropriate computer processing capability."
Another example is CEI (Apex, NC), which makes EnSight software for visualizing CAE results. The application creates graphics renderings of large engineering simulations, such as CFD, automotive and aircraft crash worthiness, tire hydroplaning, and occupant safety studies. To speed the process, CEI has partnered with MSC.Software (Santa Ana, CA) to use Beowulf clusters powered by MSC.Linux, a proprietary operating system. A Beowulf cluster uses several off-the-shelf PCs connected via Ethernet to solve problems that would normally be handled by a single supercomputer.
Sun Microsystems (Santa Clara, CA) makes Sun Grid Engine software, now in version 5.3, to parcel out the processing power in unused CPUs. That way a department, campus, or even international company can run its most intensive computer calculations on shared resources. Sun says there are over 5,000 such grids in operation today, and the number is growing since the company offers free downloads of the software on its site. "Basically it's a social problem more than a technical problem," says Sun's Miha Ahronovitz. So the new release can ensure that a group shares its computing resources in an equitable and enforceable way. First it uses a policy module to count the number of CPUs on a network, their cycles, licenses, I/O, and other variables. Then it prioritizes and schedules waiting jobs, and distributes them to any free nodes.
One user is Ford Motor Co.'s powertrain and transmission group, which has linked 500 dual-processor workstations. Engineers submit more MCAD and CFD work each day than the system can handle; so at night, Sun Grid Engine marshals the resources of all 1,000 CPUs and runs batch jobs, says John Tellefsrud, Sun's Grid marketing manager.
Another application for groups of processors is DCC (digital content creation), like the part-human, part-CG (computer-generated) movie Scooby-Doo. While human actors like Sarah Michelle Gellar and Freddie Prinze Jr. play the human characters, the eponymous crime-fighting hound is completely computer-generated. Warner Brothers contracted a studio called Giant Killer Robots (GKR, San Francisco, CA) to create the CG elements of the film, using software such as Maya, Softimage, After Effects, Shake, Commotion, and Photoshop. GKR used networked workstations from Boxx (Austin, TX) to handle the tremendous digital load.
For more information about Grid Engine software from Sun: Enter 535