The barriers to entry for high performance computing (HPC) continue to fall, portending big benefits for engineers looking to do virtual reality studies on design concepts or to solve problems around computational fluid dynamics or structural mechanics prior to building physical prototypes.
The price/performance curve for powerful HPC hardware is accelerating at a rapid pace, opening up technology that was once accessible only to wealthy research and academic enclaves to mainstream manufacturers building everything from automotive components to high-end golf clubs. As little as five years ago, for example, a 100 million-degree-of-freedom structural analysis problem might require a supercomputer with a price tag of anywhere from $300,000 to several millions — hardly a bargain for most businesses. In contrast, that same simulation can run just as efficiently, if not more so, today on widely available 64-bit desktop or workgroup cluster offerings from Hewlett-Packard Co. or Silicon Graphics Inc. costing as little as $25,000.
Simultaneously, an increasing number of partnerships around packaging HPC solutions for such specific applications as computer-aided engineering (CAE) is also helping to push the technology within the reach of mid-size companies. On the heels of Microsoft Corp.’s release of Windows Compute Cluster Server 2003 in November 2006 and its predecessor, Windows HPC Server 2008, announced last November and due out mid-year, hardware vendors such as HP and modeling and simulation application providers like ANSYS and Presagis have partnered to roll out pre-packaged solutions that bundle up HPC hardware, operating systems and administrative and management tools along with specific CAE applications. The goal is being able to deliver systems that come as close to running out of the box like a traditional PC without requiring specialized programming and hard-to-come-by HPC administrative talent.
The confluence of both trends is creating a boom in the HPC market. According to market researcher International Data Corp., HPC server revenue jumped to $3 billion in the third quarter of 2007, an 18 percent jump over the same quarter last year and an 8.8 percent spike quarter-over-quarter. In 2006, HPC systems accounted for 26 percent of all processors sold in the server market, more than doubling the 2003 HPC share of around 12 percent, IDC officials say. As HPC systems become more affordable for smaller companies and business units and with the dramatic cost increases associated with live experiments compared to computer modeling and simulation, IDC expects HPC platforms to create a “sea change” in scientific and engineering research and development, a shift that will create a $15-plus billion market by 2011.
With all this high-end computing power at their fingertips and with many of the administrative and configuration challenges starting to be addressed, mainstream manufacturers are more apt to make HPC-driven simulation an integral part of the early design process. For CAE applications like finite element analysis (FEA) or computational fluid dynamics (CFD), in particular, the raw horsepower of an HPC environment delivers benefits on a number of fronts. On one hand, engineers are assured of significantly faster turnaround times on complex simulations, reducing tasks that previously might have taken weeks or even months on specialized systems to days, sometimes even hours. In addition, the extra processing muscle of HPC platforms means engineers can pursue much more detailed simulations and spend less time perfecting their models prior to running a simulation. Finally, because turnaround times are much less, engineers have more leeway and bandwidth to conduct simulation studies on multiple design points instead of conserving simulation horsepower and development time for one portion of the design puzzle.
“The expansion of computing capacity takes simulation from being a forensic study to figure out why you designed a function the way you did to becoming a true tool that influences the early design stage,” says Barbara Hutchings, who oversees strategic partnerships for ANSYS.
Cost-Effective Computing Horsepower
Indeed, HPC, especially spread more broadly within an organization, is a game changer for complex kinds of CAE simulations, according to Bruce Engelmann, chief technology officer for SIMULIA, the Dassault Systčmes brand that markets the Abaqus software. Additional, cost-effective computing horsepower enables complex kinds of modeling like non-linear analysis, which is oftentimes avoided because of the demands it makes on the system and the development schedule. “Non-linear simulation takes more computational work — it’s not as easy to solve,” Engelmann says. “You might need six hours to do a linear job, but the non-linear version might require days. When you’re running the non-linear job on a cluster computer, you bring it back into that five to six hour window.”
Cheaper cluster HPC hardware can also save engineers time since they don’t need to be as diligent about putting up-front work in to create a model that makes efficient use of compute time. “A big part of the simulation effort is building the model in the first place,” Engelmann says. “HPC doesn’t help optimize the model, but you can run a less optimized model from a computing efficiency point of view without having to do a lot of setup.”
Dana Corp., a supplier of axles and drive shafts to the automotive industry, can certainly attest to that. The company has been using SMP-based (Symmetric Multiprocessing) HPC systems for years and recently started working with HP blade servers and Linux clusters to up performance for its FEA and CFD simulations using Dassault’s Abaqus 6.7 software, which supports distributed computing. The new setup has drastically improved turnaround time by four or five times, explains Frank Popielas, Dana’s manager advanced engineering for its Sealing Products Div. “A 3-D power train simulation which took 17 hours before can now be done in three to four hours or less,” Popielas says. “A more detailed model that might have taken a couple of weeks is now done in two to three days.”
The real benefit behind the time savings is that Popielas’ group is more apt to include simulation as part of the design process, which is a must for getting a product design right the first time before going to prototype. “If you’re looking at weeks of simulation time and the time line calls for a couple of weeks to make a prototype, you’re likely to do it without simulation,” he says. “But if your goal is to get a product ‘first time right,’ you have to go through simulation.”
Along with SIMULIA, other simulation software vendors like ANSYS have spent years modifying their applications for a parallel architecture so they’re optimized to take advantage of distributed or cluster computing. To do so, the application splits the work load in pieces and manages the processing across an array of cores and processors. While in theory, the task sounds basic, it’s actually a tremendous amount of work. To help developers make the transition more readily, Microsoft, in November, announced its Parallel Computing Initiative, a program charged with creating a common set of development tools, programming models and libraries that can be used to build applications that span multicore desktops and clusters.
Presagis, which develops commercial, off-the-shelf (COTS) modeling, simulation and embedded display graphics software for the aerospace and defense industry, says Microsoft’s efforts are helping to spread its Terra Vista tools into the hands of a wider HPC audience. Presagis has entered into a three-way alliance with Microsoft and HP to deliver a bundle offering around Terre Vista that will provide simulation capabilities as an out-of-the-box solution. ANSYS also has a partnership with Microsoft to performance-tune its ANSYS 11 and Fluent 6.3 applications with Windows Compute Cluster Server 2003.
Windows HPC Server 2008, the successor to Microsoft’s server platform, is based on the Windows 2008 operating system and has features designed to address productivity, manageability and scalability issues. Specifically, the upgrade delivers high-speed networking capabilities, new scalable cluster management tools, advanced failover functionality, a Service Oriented Architecture (SOA) job scheduler and support for partners’ clustered file systems.
HP has a similar vision of cutting through the complexity to entice non-traditional audiences to adopt HPC platforms. Last November, HP announced the HP Cluster Platform Workgroup System based on its BladeSystem c3000 unit, which offers midsize businesses almost a teraflop per second of computing power while taking up only 2 sq ft of floor space. The system is even more mainstream-friendly thanks to such features as power supplies that plug into standard wall outlets, integrated network cables and management tools. Accompanying this system are Solution Blocks, which build integrated applications right onto the platform; ANSYS Fluent CFD configured on the HP Workgroup System with Windows Compute Cluster Server 2003 is one of the first bundled packages released on the HP platform.
“Most of the roadblocks to HPC have been around complexity,” says Ed Turkel, manager of product and technology marketing for HP’s High Performance Computing Div. “Being able to offer more or less a complete solution eliminates a lot of those major concerns.”