For those who want to run high performance computing applications in the public cloud, there are limited options, but not for the reasons one might expect. Cloud providers like Amazon, for instance, have gone to great lengths to kick users the right hardware and networks for these applications and the business model for spotty HPC use has been widely discussed as a viable, more desirable alternative to the massive in-house cluster if demand is occasional.
The real problem with bringing HPC applications to the cloud has little to do with roadblocks technology-wise. Rather, it is the thorny problem of software licensing that has stalled HPC cloud adoption with few companies offering HPC software taking the first steps to find a flexible model for pricing that rolls with the on-demand, scalable nature of the cloud.
This is problematic for smaller companies who either want to pop off the workstation for more horsepower or who simply don’t specialize in HPC and need to run a quick set of demanding HPC tasks without bothering with on-site hardware. In other words, the ability for smaller companies to quickly onboard with high-end engineering packages they’re used to using in a pay-as-you go sort of cloud environment is struck down—leaving smaller companies who have hit a workstation wall but that have large-scale simulations to run with the decision to either procure and maintain a power-hungry cluster or use an HPC on demand service, which is less nimble scalability and options-wise.