Small- to medium-sized manufacturers can realize major benefits from HPC modeling, simulation and analysis, but are hesitant to take the plunge
Folk wisdom can sometimes be right on target. For example, there’s that old bromide about leading a horse to water. In this case, the water is high performance computing (HPC) and the reluctant equine is the huge base of small- to medium-sized manufacturers (SMMs) in the U.S.
According to the National Center for Manufacturing Sciences (NCMS), there are approximately 300,000 manufacturers in the United States. Over 95 percent of them can be characterized as SMMs. These companies, usually with less than 2,000 employees, may be small when compared to the Boeings, Dow Chemicals and General Electrics of the world, but they are mighty. SMMs account for more than twice the total employment of all the Tier One manufacturers combined and generate about three quarters of the total global R&D output.
In addition to the SMM tag, these manufacturers have been called the “missing middle” — catchy, but inaccurate. NCMS has coined a new term — “backbone manufacturers,” pointing out in its recently released Voice of the Customer (VOC) survey1 “Without the backbone, the biggest manufacturers simply couldn’t operate.”
The survey, which was a joint effort between NCMS and the Michigan Manufacturing Technology Center (MMTC), was distributed to about 10,000 Michigan-based manufacturers with between 10 and 2,000 employees. According to Danielle Jones, NCMS business development manager and one of the survey’s authors, they not only gathered a wealth of data from the 350 respondents, but also conducted one-on-one interviews. The same survey is underway in Ohio and, eventually, will be rolled out to other manufacturing centers in the country.
NCMS is not the only organization attempting to bring HPC to the backbone manufacturers — particularly modeling, simulation and analysis (MS&A). Hardware and software vendors; federal, state and local governments; academic institutions; and creative new initiatives like the HPC UberCloud Experiment — all are looking to tap in to this underserved market.
Easier said than done. Despite the promise of digital manufacturing, the main response from the backbone has been lukewarm. “HPC is well-established in the Tier One manufacturing companies,” says Jon Riley, vice president of digital manufacturing at NCMS. “But the technology has barely made a dent in the SMM supply chain.”
Part of the problem is that the backbone manufacturers are under constant pressure to do more with less, to operate with razor-thin margins, and cut costs and staff while delivering superior quality. You would think these SMMs would be lining up to incorporate the latest HPC-based manufacturing tools and processes. And you would be wrong.
Although the backbone may recognize the need for these advanced tools, they are reluctant to adopt them. One of the biggest barriers is cost: the cost of hardware; software with its tangled and expensive licensing requirements; workforce issues; maintenance, data center power, cooling and floor space costs — the list goes on.
NCMS’s Jones points out that these backbone companies are very lean. They have neither the time, money, personnel nor the inclination to take on HPC, no matter how innovative the solutions might be. They are focused on getting through the day, rather than taking a long view that might include planning to incorporate innovative HPC tools and processes into their IT ecosystem. However, this is slowly changing. Backbone manufacturers are increasingly being asked by their customers to come up with innovative solutions and will need to adopt new technologies — including HPC.
HPC on the Factory Floor
There also may be a misunderstanding at play. Ed Turkel, group manager, worldwide HPC business development at HP, has taken a close look at the backbone’s needs. Conversations with SMM customers have revealed that they are looking for help not so much with product design problems, but with the manufacturing process itself — the CAM (computer-aided manufacturing) side of the equation. In many instances, the parts these companies design are relatively simple and can be handled with CAD on a workstation — no HPC required. Instead, they are concerned with developing faster and more cost-effective tooling while still delivering high quality.
There is a generational factor, Turkel adds. Employees, who have successfully been using 2D-CAD on PCs and workstations for the past 25 to 30 years, are less likely to be open to new HPC-based tools. The younger generation, just coming out of school, will have learned about these advanced MS&A solutions and are more inclined to adopt HPC-based solutions once they become part of a backbone organization.
One of the biggest roadblocks to HPC adoption among the backbones is risk. These companies are focused on day-to-day operations and anything — such as introducing complex and expensive HPC-based technology that could interfere with production and negatively impact the company’s CAPEX and OPEX — is to be scrupulously avoided.
Jon Riley notes that, according to the NCMS survey, return on investment (ROI) is a major consideration as to whether or not a backbone adopts HPC. “Small companies work on a day-to-day, week-to-week basis,” he says. “If the ROI from adopting this new technology is longer than six weeks or so, the investment becomes problematic.”
ISVs and VARs attempting to penetrate this market are running up against another problem. They usually present their prospects with general use cases that highlight their products and services. But, backbone manufacturers are not interested. According to the survey, “Nearly every backbone manufacturer we surveyed indicated that little or nothing they had seen so far showed how digital manufacturing could be of benefit to them, and many were dubious about semi-generic examples and case studies that didn’t directly reflect their own needs and processes.”
Many of the backbone companies view MS&A as far more highly advanced and idiosyncratic than it is. HPC advocates, whether in the vendor community or with government, academia or non-profit organizations, have to do a better job of communicating what HPC is and why it can bring major benefits to these beleaguered backbone manufacturers.
Despite the difficulties, there are some encouraging signs that the backbone manufacturers are willing to examine working with HPC, especially when it comes wrapped in the cloud. One of the more innovative initiatives attracting attention among the SMM community is the UberCloud Experiment.2 Initiated by Wolfgang Gentzsch and Burak Yenier in 2012, the experiment is a free, community-driven effort that helps SMMs explore the use of remote HPC resources. Over 2,000 organizations and individuals — many of them SMMs — from more than 70 countries have participated so far. Some 155 teams consisting of end users, resource providers, software providers and HPC experts have been involved to date, and more are signing up all the time.
Over the past two years, Gentzsch and Yenier have had a hands-on acquaintance with the problems facing their participants. “The major challenge is accessibility to compute power, which today is still not user-friendly enough,” says Gentzsch. “To compete in a global marketplace, the SMMs are really under pressure. But, they cannot afford to buy an HPC cluster.” Total cost of ownership (TCO) is also a major limiting factor according to an IDC study3 cited on the UberCloud Web site.
Gentzsch points out that software licenses are a big problem — most ISVs do not have a clear HPC-in-the-cloud business model, even though the backbone engineers and scientists using this technology are a large potential market for additional temporary licenses.
And then there is job security. Server-hugging IT departments, no matter how small, do not look kindly on losing control of the computing infrastructure. Desktop workstations are one thing; HPC in the cloud is another.
Signs of Progress
Despite all these obstacles, there are indications that HPC, including MS&A, is beginning to make some inroads into the backbone manufacturer community. The enthusiastic response to the UberCloud Experiment, mentioned above, is one of those positive signs.
In addition, HP’s Turkel likes to point to the establishment in 2013 of the Center for Collision Safety and Analysis (CCSA) at George Mason University in partnership with NCMS. George Mason is the site of one of NCMS’s Grid Cells. CCSA is working with NCMS and HP to make HPC resources available to small-and medium-sized automobile parts manufacturers who typically can’t afford to buy and operate their own HPC clusters. The 12-node HPC cluster at CCSA was designed by HP engineers in collaboration with NCMS to meet its backbone users’ specific needs. It’s built around an HP ProLiant DL380p head node and powered by 12 HP ProLiant SL230s server trays with Intel Xeon processors in an HP ProLiant SL6500 scalable system. This setup provides SMM users with the horsepower they need to run complex crash testing simulations.
Commenting on the CCSA’s mission, Jon Riley described the goal of similar initiatives springing up in Michigan and other manufacturing centers. Says Riley, “Our mission is to look at avenues where we can introduce the technology, introduce the value, help them (the SMMs) explore the return on investment; and then give them this enabling mechanism to affordably access technologies where they can begin to explore the benefits in real-life applications.”
NCMS’s first Grid Cell devoted to advanced materials and composites was launched last year. Hosted at NCMS headquarters in Ann Arbor, MI, the Grid Cell provides backbone manufacturers with a virtualized approach that includes HPC-based MS&A, as well as data mining tools and the digitization of manufacturing processes to optimize speed, reliability and efficiency.
Gentzsch applauds the creation of the Grid Cells and is working with NCMS to make the UberCloud’s hands-on approach to HPC in the cloud available to backbone manufacturers. But Gentzsch also takes the long view, commenting that education is the key. “There are not enough universities that offer computational engineering as it relates to manufacturing,” he says. “We have to attract young students to the field by building excitement around engineering and the sciences early on — in the 5th, 6th or 7th grades.”
Adaptive Computing, provider of workload management software for data center and cloud environments, is one of the many ISVs that are exploring how to best serve backbone manufacturers.
Jill King, the company’s director of marketing, likes to point to SMM customers who have successfully made the leap into the world of HPC. For example, one backbone customer that makes parts for Boeing recently added HPC and MS&A to their manufacturing process. The Tier 2 supplier is now able to move far faster than its competition — a benefit to Boeing as well.
And then there is the agricultural company whose scientists calculated it would take five months to analyze some critical data using their desktop systems. One of the company’s IT people said “Let’s try running this on our HPC cluster.” Two days later, the team had its results.
One of King’s colleagues, Paul Anderson, Adaptive’s director of professional services, comments that with today’s affordable commodity servers and accelerators, smaller companies can field a HPC cluster with peak performance for a relatively light investment. These systems handle everyday computational requirements, and overflow can be taken care of in the cloud or by colocation.
Gentzsch has added an interesting twist. The newly launched UberCloud Marketplace is using open source lightweight Linux containers to provide users with bundled package of containerized applications at their fingertips. This solution removes a lot of the problems around cloud accessibility, licensing and security that deviled many of the teams involved in the UberCloud HPC Experiment. “Now,” says Gentzsch, “we are applying this technique to teams participating in the Experiment. We are finding that what once took three months can now be accomplished in one or two weeks.”
A New Dawn
Grid Centers, HPC in the cloud, awakened interest on the part of HPC OEMs, ISVs and VARS, government and non-profit initiatives, outreach on the part of the supercomputer centers — all these developments point to a renaissance of computational capabilities among the hundreds of thousands of backbone manufacturers that can profit from the adoption of HPC.
But keep those recalcitrant horses in mind. Some have taken a sip, but most have yet to taste the benefits that HPC can bring.
As HP’s Ed Turkel notes, drawing on his long experience as a member of the HPC community, “We are just at the beginning.”
1. Voice of the Customer (VOC) survey: http://doitindigital.com/wp-content/documents/VoiceOfTheCustomer.pdf
2. UberCloud Experiment: www.theubercloud.com/hpc-experiment
3. IDC study: www.theubercloud.com/cost
John Kirkley, President of Kirkley Communications, is a writer and editor who specializes in HPC. He may be reached at editor@ScientificComputing.com.