Bill Clark, executive VP of CD-adapco, remembered a typical conversation he had with customers eight or nine years ago. He recalled, “They’d come and say, ‘Hey, Bill, do you know your software costs more than the computers we’re using? It doesn’t make sense.’ And that got us thinking.”
The thinking, Clark said, led to an overhaul of the company’s computer-aided engineering (CAE) software licensing policy. It marked the introduction of two distinct licensing approaches: Power Session and Power Token. They’re specifically designed to address the customers’ increased use of high-performance computing (HPC) to run complex simulation jobs. CD-adapco introduced Power Session in 2010 and the Power Tokens in 2013. The company also offers Power on Demand, a SaaS-style usage-based model.
What is Traditional CAE Licensing?
Clark explained, “The traditional method of CAE licensing is, you buy a seat of the software, then you get access to HPC capacity by buying HPC licenses, albeit less expensive than buying additional seats. When you buy a seat of the software, which costs a good chunk of money, you can only run a single job on a single core. With the added HPC licensing, you can deploy additional cores, with a fee for each additional core you want to deploy.”
The method was sensible when most CAE software users were looking to tackle jobs one at a time on a single computer. But the rise of multicore machines, the increased complexity of simulation jobs, and the availability of relatively affordable HPC hardware demanded a new approach. What CAE users want today is to (1) run one large job on hardware with hundreds of computing cores; or (2) run multiple versions of the same design on hardware with hundreds of computing cores. With traditional licensing, a significant fee is required for each additional core delegated to solve the job. And to run ten derivative designs simultaneously would require ten seats of the same software. Therefore, the cost of HPC-powered simulation is too prohibitive to pursue, especially for small and midsized engineering firms.
Power Session (for One Large Job, Distributed over Multiple Cores)
The basic principle of CD-adapco’s Power Session is as follows, according to Clark: “We don’t care how big your simulation job is. If you have a million cores at your disposal, it should be relatively inexpensive to run your job.”
Clark estimated that, if the cost of a typical CAE seat is 1, the power session costs you roughly 2. In other words, for roughly the equivalent of one additional seat’s cost, you get infinite HPC scaling capacity in Power Session. That’s a reasonable burden for engineers who want to process a single complex job over multiple computing cores. The nominal additional fee is offset by the speed with which they can get answer back (in a few hours instead of days).
Power Token (for a Family of Design, Distributed over Hundreds of Cores)
The more ambitious type of HPC-powered simulation is commonly called design of experiments (DOE). With this approach, an engineer might use CAE to evaluate, for instance, ten variations of a new vehicle model, with ten cores devoted to each variation (that’s one hundred computing cores data-crunching on ten versions of the same design at the same time).
For DOE, CD-adapco suggests Power Tokens. Clark explained, “A token pays for either another simulation session or adding another core. So if you want to do ten simultaneous simulations on ten cores each, occupying a hundred cores, in our licensing model, you’d need to use a hundred Power Tokens. The cost of that would be roughly the equivalent of 4 or 5 traditional CAE software seats.”
One important distinction—this approach is permissible only with design variants, not multiple designs. In other words, the problem has to be set up parametrically in CD-adapco’s STAR CCM+ software as design variants (for example, ten variations of same aircraft door using different thicknesses). It cannot be used to address ten different designs (a car, a boat, an aircraft, a cellphone, and so on) running in parallel.
Changing Attitude in the Industry
Clark observed, “I think everybody recognizes that traditional licensing has to go away. It doesn’t make a lot of sense. A large majority of our customers have migrated away from that old single seat-plus-HPC setup.”
One impetus for CAE software vendors to come up with a more sensible approach is the competition from cloud-hosted SaaS-style CAE vendors. Newcomers and startups like Rescale make CAE users think of simulation as a service—a mix of software and backend computing hardware, both accessible from a browser window. The appetite for this simplified approach is increasing. If running CAE software on in-house HPC proves too costly, the users now have a cheaper alternative. That alone provides sound logic to altar the traditional CAE licensing.
For more on this topic, read “Time to Reengineer CAE Licensing: Pay Per Core, Per Job, or On Demand?“