Making High-Performance Computing Work for Industry
We have all seen the stories of companies using High- Performance Computing (HPC) technologies to give them a competitive advantage in the market place. However, just how do these companies accomplish getting to the point where HPC gives them a competitive advantage? The key ingredients to get to that point include (i) making the business case for the time and cost required, (ii) pulling together the people, software, and computing resources needed, and (iii) making the resulting HPC technologies a normal part of your processes.
As an academic that creates specific HPC and spends some time working on supporting industry in the effective use of these technologies, I am not qualified to tell companies how to make the business case analysis for using HPC technologies. However, I can indicate that the range and types of companies that are able to make the case is broad as is the role that HPC plays in their business. Nearly everyone has heard of cases how aerospace companies have saved millions of gallons of fuel per year by using super computer simulations to squeeze another percent of performance from aircraft, and how automotive companies rely on crash analysis shave months and millions of dollars off the process of designing safer cars. However, even in those cases, it is less well know how advanced supercomputer based simulations have allowed an aerospace company to introduce new technologies such as active flow control into next generation designs, or how the automotive companies can trust a HPC based, simulation-based design system to design the interior comfort control systems that skip prototype testing go directly to acceptance testing on the first version produced. The use of advanced computing and HPC in the design and manufacture of the full range of consumer products by P&G is also becoming more widely known, as well as the fact that designing a superior baby diaper is not rocket science, the simulations involved are more complex!
The capabilities of available HPC simulation software and the ease to integrate it into useful industrial simulation workflow is improving
Companies we have been involved with developing simulation workflows for range from three person start-ups to some of the world’s largest companies. In some cases the application of advanced simulation is at the very core of their business. For example, HeartFlow provides an HPC based service that accepts a patient’s arterial system image data and creates a personalized 3D model of their coronary arteries and analyzes the blood flow in their arteries to evaluate the impact of blockages, so physicians can determine the best treatment plan for that patient. Other companies have used HPC technologies to improve their design procedures. In one case, a company wanted to determine why a new product they introduced was so effective in reducing industrial emissions and to improve their design. Detailed flow simulations demonstrated how to obtain superior mixing and lead to improved designs. However, it turned out that one of the major benefits to the company was the use of the flow simulation visualizations in showing customers what their device would do as part of their sales processes. In another example a 170 year old pump company wanted to increase their competitive position in large scale custom pump designs by being able to more quickly design pumps that they were confident would meet the customer specifications without the need to sell them a more expensive overly conservative design.
Even in cases where companies have an idea where HPC technologies can be used to provide them a competitive advantage, there is typically substantial effort required to develop the needed simulation workflows and to integrate them into their every business practices. This is particularly true if the company does not have the resident HPC simulation experts in place. It is not true that by national laboratories and universities making the HPC software and hardware they have available to industry, the industry will be able to take full advantage of HPC. It is not a “field of dreams” where “if you build it they will come.” It is usually a far more complex process where HPC experts must work directly with company representatives that understand their technical needs to define potential HPC simulation workflows that will address their needs. Once the basics of these workflows have been defined, this team must evaluate the potential software and computer hardware that can effectively address those needs and to carefully evaluate which of these various components make sense for use by that company considering both technical and business factors. From there the team must work together to put the selected workflow(s) in place, including filling in any missing components or interfaces. Finally these workflow(s) must be integrated into the company’s business practice, a process that typically requires additional refinements of the developed workflows.
The time and effort required to create a useful HPC workflow for industry can vary dramatically. In the best of cases that we have worked up to date, it takes months of effort. The good news is that, there are now some groups that have gained experience in being able to work with the industry to develop and support them in making HPC work for them. This has led to the ability to determine and create useful HPC workflows more effectively, and to be able to better estimate the efforts involved and inform interested industrial parties as to the expected levels of effort required. In addition, the capabilities of available HPC simulation software and the ease to integrate it into useful industrial simulation workflow is improving. However, there is considerable room for improvement across the board, but that is a subject of another day.