Data Management: The Missing Link to the Missing Middle
As large manufacturers cut back on R&D funding, the “missing middle” of small-to-medium sized manufacturers (SMMs) represents a growing fount of innovation within the United States. Still, the very tools that enable this, remain beyond the grasp of many of the companies in this critical position.
Although computational fluid dynamics (CFD) has become a staple for large manufacturers, the fact remains that the missing middle continues to miss out on this advanced simulation tool. Beyond being able to afford the necessary software, CFD requires substantial investments in high performance computing (HPC) infrastructure as well as the talent that would meet both hardware and software demands. But according to Steve M. Legensky, founder and general manager of Intelligent Light, there’s more to CFD adoption than ROI.
“You have to get past the risk reduction argument before you can start to sell based on ROI,” Legensky said during a recent interview with Digital Manufacturing Report. “The small-to-medium guys aren’t seduced by CFD and it’s great mystery or promise. If they can’t cost-effectively predict the behavior of new products with low uncertainty, they’re not interested.”
He explained that because of limited resources that are found in smaller businesses, any disruption is often looked upon with a wary eye. And when someone proposes using CFD, it comes with it a number of disruptions from hiring additional talent to buying particular software and a cluster to run it on.
“Typically in our marketplace, the products we compete with focus on just taking the data files from simulation and allowing people to take pictures of them and make calculations with them,” Legensky said.But in focusing on data management over the past five years, Intelligent Light has taken this tool and made it the star of their CFD product, FieldView.
But what does that mean?
Legensky explained that, “From the HPC perspective, typically the world has divided up into two hardware camps within a company: the IT department that runs some cluster or gets space or compute time from an outside cluster[…], and then there is the desktop resource that the engineer has for reporting and analyzing. Clearly there is a mismatch between the two in terms of the disk-available and compute-available memory.”
What Intelligent Light found whether they looked at in-house clusters or HPC in the cloud, the typical result files spanned tens of gigs up to several terabytes of data, which presents unique challenges when it’s time to take a look at it using a desktop machine. As a result, Intelligent Light found three ways to bridge the gap and put them all in their standard CFD offering.
The first option is a client server split inside of FieldView where you can run a client process on your desktop or laptop and attach to server processes that run on the HPC resource. So the big data files stay on the HPC resource, the FieldView servers extract what you ask for and send it to your desktop.
To illustrate this, Legensky used the example of a car simulation, complete with a multitude of cells within the solution. “If you want to calculate the drag on that car you only need the shear velocities or momentum fluxes on the body of the car,” he explained. “The cells around the body of an object in an unstructured calculation (if you just look at the viscous boundary layer and what’s near to the car), that can be on the order of three percent of the whole solution space. You need all the cells in the distance around the car in order to do the calculation, but you don’t need all those cells when it’s time to post process.”
The second data management mechanism is what FieldView call true batch operation.
In Legensky’s car example, he explained that you’re never looking at just one car, or just one operating condition.
This tool begins with the same car simulation, but goes further to create an automated script using FieldView’s FVX post-processing language. Users can then take that script and run FieldView in batch mode entirely on the HPC resource, creating whatever pictures or numerical reports or spreadsheet outputs that they need.
To meet this demand, Legenskys said that Intelligent Light offers lower-priced, batch-only licenses to facilitate that kind of operation in a cost-effective way.
Finally, the third solution to the big data bottleneck sits somewhere in between FieldView’s first two data management options.
“All of those things that are created during a visualization session with FieldView can be saved out into what we call an extract database (XDB). The XDB is a full-fidelity, full-precision extract from your CFD results file that contains only what you ask for in a very compact binary file that can then be FTPed from the HPC resource to local resources for interactive post processing.”
And for the SMMs that are on the fence as far as adopting CFD, Legensky believes that this feature will be key.