We’ve heard quite a bit about the “missing middle” in high performance computing, particularly in reference to the manufacturing sector. However, for those manufacturers who have already adopted HPC hardware and tools into their environments, there are still some gaps—despite the obvious benefits and increasing investments some companies at last week’s EnterpriseHPC event in Carlsbad described in great detail.
During a standout HPC in Engineering panel moderated by NCSA’s Merle Giles, we heard high performance computing leads at Rolls Royce, Cummins Inc., Mercury Marine, and others discuss the opportunities and pain points for their use of advanced modeling and simulation. Just as many might imagine, the skill gap in finding the right people to hop in and leverage HPC software (not to mention the hardware) was a theme, but more specifically, so were issues around process, workflow, integration, and toolsets that merged all of those together.
There were a number of themes that emerged and coalesced between the HPC modeling and simulation end users on the panel that were consistent despite organization size or workload differences. First, and perhaps least surprising, was the fervent agreement that the advanced capabilities from leveraging HPC hardware and software tools is critical to driving competitiveness and quality of end products. Second, with these tools, companies are able to achieve an ROI that’s a bit more difficult to quantify in numbers (i.e. we spent $1 million on infrastructure and modeling tools and saw x for a result). However, the ROI value in removing the expensive, inexact, time-consuming processes of physical testing is clear—and leads to better products. In short, HPC is the design and testing tool whereas validation is now the only defined step where prototypes are physically tested.
“We get through far more design iterations than we used to five or ten years ago since physical testing had to be built into everything we did,” explained Arden Anderson, Specialist in Engineering Simulation at Mercury Marine. He described HPC operations dedicated to their engine design business, noting that their investment in HPC is set to grow with the aim of doubling their core count every two years (they are still relatively small, operating 200-300 clusters for their designs) and growing their analyst workforce by 10-20% per year. With their current finite element and CFD applications, their challenge hasn’t just been around getting the right skilled people, but “integrating what comes out of the engineering simulations and integrating that into our design process.” Like other manufacturers on the panel, he said that HPC has already allowed them to use physical tests as a validation tool (versus strict design and testing—a long, expensive process) but with tools that were better integrated across the lines of business, they could do even more. “We can now design at a much faster pace but our real challenge is integrating at a pace that fits.”
“In the 80s, we would build and test components for compressors, combustion parts, and in engine design, but that cycle typically took many years. We would have to build a lot of different rigs, then study through repetition and analysis to find the right part. Today, when we do a rig test design we do one. And one alone. And this is for validation, mostly to appease other tests.” He said that instead of the emphasis being on these extensive physical test runs, his teams concentrate on getting the models right so that the first time it’s right. This is hugely cost and time saving—and leads to much faster time to market with more robust products.
Simons agrees that integrating these designs into larger workflows and internal processes is always a challenge and that people too are key to their success going forward. However, like others on the panel, they are reliant on some assistance from university partnerships, including their own at NCSA where they’re able to work with advanced teams to scale commercial code (including LS-DYNA, which we’ve covered in the past) and find more efficient ways to manage complex workflows.
The same challenges and opportunities for HPC in manufacturing that Rolls Royce and Mercury Marine face were persistent at Fortune 500 engine and power generation company, Cummins. According to the company’s Modeling and Simulation Services Leader, Michael Hughes, the advantages of advanced HPC are clear, but fitting them into the larger processes at the company is the real challenge. His team’s 2500 core facility is primed for 60% growth on the compute side, along the with the desire to hire more modeling and simulation experts. “our applications are becoming supercomputing capable,” he said, “we have a lot of engineers who were bound to the PC or workstation who now have a gleam in their eyes as they see the types of new problems they can solve.” With the ability to rapidly expand prototyping, design, and testing, however, comes something of a “Amdahl’s Law” of innovation—even with the fastest, most advanced tools to power the design side of the house, without the ability to integrate those into the larger pipeline and speed validation, the ROI of the HPC investment gets lost in the shuffle.
But for companies like Rolls Royce, Mercury Marine, and Cummins, if these challenges in terms of making the HPC investments pertinent along the other business-critical lines are all consistent (and at differing layers of the Fortune 500 stack) why aren’t more tools emerging to solve these problems?
First, it seems to be a matter of question about where the bottlenecks are. For some, the problem is integrating the many different toolsets within the HPC environment, which means it’s a matter of workload and workflow management. For others, that problem is compounded by the fact that even with such streamlined ways of handling those boulders in the HPC stream, the larger organization needs to be able to tune its processes to a jump in the pace of innovation by revising its validation and standards procedures.
The question is, what tools exist that do both of these things? And even with an answer to that question, how does on affect the cultural change required to bring it into practice. Further, even with both of those pieces of the puzzle clicked in, what if it’s too difficult to find the people to make it all come together?
The one key that seems to be working for all three of these companies is the valuable input and assistance received from public/private partnerships. From San Diego Supercomputer Center’s industrial outreach program (we heard about that during another panel at EnterpriseHPC we’ll touch on later), to the Ohio Supercomputer Center, NCSA, and others, it seems like the time is more ripe than ever to bring more funding and effort to bear. Not just to push potential new manufacturing users toward HPC, but to help fully onboard the whole organization as well as individual tool users who are already HPC users.
In short, for these end users, all with varying cluster sizes dedicated to a range of different modeling chores, it’s not a matter of access to high performance computing, it’s a matter of implementing and managing it in a way that makes sense throughout the organization—and with the right tooling to make that integration possible. While the vendor and ecosystem partner role is worth pointing out as well, for HPC to really go mainstream in manufacturing the push needs to be farther than just in the internal IT/modeling and simulations groups.