November 13, 2014
On Oct. 22, as the sun went down on Long Beach skyline, a small storm was brewing on the deck of The Queen Mary, now permanently anchored at the harbor as a 340-room hotel.
Aside from being a time capsule of the age of ocean liners, the hotel also happened to be hosting Siemens CAE & Test Symposium (October 22-23). On Wednesday, after the relative calm of customer presentations, the evening reached its peak in one of the staterooms during the Meet and Greet that brought together users and analysts.
The conversation topic was Computer-Aided Engineering (CAE) in the cloud. The inevitable question came up: Do current software licensing methods impede cloud-driven CAE? Roughly 30 minutes before cocktail reception, the exchanges showed no sign of dissipation. Those assembled were still busy trading arguments and counterarguments. Comments flew from Port to Starboard. The crowd was a microcosm of the CAE industry, and the debate seems long overdue.
Keith Meintjes, industry watcher CIMdata’s practice manager for simulation and analysis, sums up the evolution of simulation licensing: “In the old days, they were licensed per mainframe computer, with unlimited simultaneous users. Then, along came Unix workstations and PCs. Software was licensed by end user machine or, for server-based software, by named user. Then, along came serious workstations that were 100x more powerful than normal desktops. Licensing was by CPU power. Then, along came multi-processor Crays. Licensing started to be per CPU or parallel thread.”
The pay-per-core licensing model has its critics too. However, many enterprise users have learned to get their money’s worth from this model by investing in enough hardware (computing cores) to get justifiable speedup from the software.
But the prohibitive cost discourages those who want to pursue optimization or Design of Experiment (DOE) studies, which involve simulating multiple variations of a design on high-performance computing (HPC) hardware with hundreds of cores. For example, should a user be expected to purchase 25 licenses of a simulation software to simulate the crash worthiness of a new vehicle with 25 airbag deployment variations? A typical CAE license would let such a user run the simulation one variation at a time, but not all 25 in parallel—not unless the user pays for 24 more licenses.
Monica Schnitger, founder and president of analyst firm Schnitger Corporation, points out, “Buyers feel gouged when they want to run a complex simulation on multiple CPUs (theirs or someone else’s) but need to have that many licenses in their agreements.”
Meintjes says, “I counsel our solution-provider clients that pricing based on perceived customer value is toxic in terms of developing a long-term relationship. Some of them listen, many don’t.”
If existing CAE software vendors cannot accommodate DOE studies at a reasonable cost, users may migrate to start-ups that operate with a business model to cater to DOE champions. One such vendor is San Francisco-headquartered Rescale, which offers its computing infrastructure and job-management software as an on-demand bundle. Rescale has also struck up partnerships with existing CAE vendors—including Siemens PLM Software, ANSYS, Autodesk, Dassault Systemes SIMULIA, and CD-adapco—to negotiate and acquire additional licenses where needed.
Jim Rusk, a senior VP from Siemens’ digital factory division, says, “Our partnership with Rescale lets people use NX Nastran in a HPC environment to facilitate DOE-type work, so you can vary a set of parameters over a design to see what the design should look like. If the customer let us know they’re trying to do DOE, we have some approaches that let them accomplish that without paying for hundreds of licenses.”
Rusk refers to the two licensing modes available as DMP (distributed memory parallel, for HPC users) and SMP (shared memory parallel, for processing a single job on many computing cores). “In the DMP licensing, we have special configurations that let you run DOE,” he says.
Schnitger ponders on a few CAE providers’ policies. “Autodesk and CD-adapco—which price by the job—may be on to something,” she says.
Meintjes points to Altair‘s HPC business model as another DOE-friendly example. He notes, “Altair has a nice ‘any of the above’ attitude. You can get set up to run in their data center so it looks exactly like an internal HPC system, using your licenses or theirs. They will also host your licenses in the cloud so you can run anywhere. Finally, they will sell you an HPC server that has everything included. There are only two connections: Power and network. You can run whatever you want, as much as you want, in the pre-installed Hyperworks suite. No licensing. You are, of course, limited by the capacity of the box. Altair has their software ‘phone home’ to report license usage, so they get valuable information on what their customers are doing. Their token-based system also has a ‘leveling’ feature so customers are not penalized for running multiple applications together.” (For more on this, read “From Perpetual Ownership to On-Demand, On Lease, Augmented with Off-Premise Power,” Oct. 2013.)
Most CAE vendors may have licensing options to accommodate DOE at affordable cost, even if they don’t publicize the fact. The most advanced CAE users—aerospace and automotive manufacturers that make up a large portion of the CAE revenue streams—are using DEO studies for optimization and lightweighting. To have a licensing policy that hampers DOE would be a detriment to a CAE software vendor.
Meintjes says, “Most solution providers are pursuing added features for leading-edge users rather than a broader customer base. This runs counter to the desire of most end users to have simulation be democratized and pervasive.”
Schnitger points out, “No one really has it figured out.” That leaves the window of opportunity open for newcomers and adaptive old-timers who can come up with an attractive proposition for users with DOE problems.