Boeing's R&Ddivision is replacing expensive mainframe computing resources with Linux and HP -compute clusters to run compute-intensive simulations for designing aircraft more cost-effectively.

need to reduce computing costs

Michael Mills managed computing resources for Boeing's Phantom Works division, the centralized R&D group for the giant aerospace company that does contract work for the U.S. government and NASA. He was responsible for overseeing the division's $30 million computing budget. "Boeing is becoming increasingly focused on rate management-the dollar per hour value that you charge outside customers when you do a contract," explains Mills. "If you want to keep your rate stable or reduce it, you could either bring in more work with the same people or you could reduce heads. No one wants to do the latter so we tend to focus on reducing indirect costs. The largest controllable indirect expense is computing. There is a tremendous focus on that. We are constantly asked to reduce our computing budget."

A few years ago, the R&D group was examining its high-speed compute serving costs. They were using Cray supercomputers and IBM mainframes for complex computational jobs to support their aircraft design work. For example, to do a computation of fluid dynamics of a wing variation, engineers need to measure the structural performance at 60 million points on a wing. It can take a couple weeks to run this type of computation. "When you're paying for processing time, it gets very expensive. Running a Cray can be faster but it costs a lot of money. We thought there had to be a better way," explains Mills.

Beowulf clusters from MSC.Software

Mills had heard good things about Linux and Beowulf clusters. Beowulf is an approach to building a supercomputer as a cluster of off-the-shelf personal computers, interconnected with a technology like Ethernet, and running programs written for parallel processing. "It seemed like a good way to solve our problem," he says. "Instead of buying cycles or big processors that are outdated in a year, we could pick up a Beowulf cluster of high-speed workstations for a fraction of the cost. We could easily scale the cluster in the future to provide more processing power, or break it up and reallocate the resources in other ways if it didn't work out."

So Boeing R&D formed a project to evaluate PC clusters running Linux. They decided to buy a turnkey Beowulf cluster from MSC.Software, the software company that also provides the key engineering software Boeing uses for Computational Fluid Dynamics processing. MSC.Software and HP work in close partnership to offer turnkey solutions to their customers.

MSC had developed their own version of the Linux operating system- MSC.Linux- that is optimized for compute-intensive applications like Computational Fluid Dynamics processing and specific applications like Nastran. Boeing's initial cluster included ten HP C3600 workstations running MSC.Linux. HP provided the hardware, MSC configured the cluster and then they both delivered a turnkey system to Boeing. "The hardware people from HP and application people from MSC worked together very closely. They had us up and running in two hours," says Mills. "We've never had any downtime."

superior customer support reduces computing costs

Boeing has been very happy with both price-performance and customer support from HP. "The HP systems provide great price-performance. If I can get better performance, I need fewer machines. This also helps us reduce computing costs," says Mills.

Customer support is a critical differentiator for companies that want to keep their internal support headcount at a minimum. "We're very happy with the customer support we've received from both HP and MSC," says Mills. "From a financial standpoint, the more we get in hardware and software support from a vendor, the less we have to provide within Boeing. In a big company like Boeing, rates for resources are high. By the time you calculate the cost of a support person, it's expensive. That's why you see so much outsourcing, the more you get in customer service, the lower our computing costs are."

performance up by 600 percent at half the cost

The combined Linux HP compute performance cluster clearly enabled the R&D group to meet their financial and computing resource objectives. "The results of the Linux cluster running on HP systems have been tremendous. We've had a 600 percent increase in performance and reduced our costs by 50 percent," says Mills. "All our users are asking: 'Why were you so conservative? You should have done this a long time ago.'"

"The results of the Linux clusters running on HP systems have been tremendous. We've had a 600 percent increase in performance and reduced our costs by 50 percent. All our users are asking: 'Why were you so conservative? You should have done this a long time ago." — Michael Mills, Computing Resources Manager, Boeing.