The requirements for HPC in the cloud are really no different that the requirements for HPC on-prem or hybrid HPC: Do you have a problem that requires accelerated compute or do you have a computing problem that scales up and/or out? If so, then you need to address the question of cloud, hybrid, or on-prem and in many cases, the decision to go to the cloud is based upon availability of resources with the customer and the time horizon for the project being executed.
Let’s look at a few scenarios and see where cloud might make sense to an organization:
There is an on-premises HPC cluster that runs at 80% capacity and the organization gets a large project that would out-strip their current compute capacity. This is a case of work outstripping capacity and a textbook case of when an organization should look to the cloud to cope with their occasional HPC projects that are too large for their current platform.
Proof of Concept
When looking to buy an HPC platform that will reside on-premises and they are trying to choose hardware. A cloud provider that has multiple platforms can serve as a proof of concept center and allow them to run benchmarks on a variety of hardware configurations, ensuring that they are making the best possible purchase for their current and anticipated needs.
When an organization is burdened by lack of real estate to devote to an on-premises system and the ever increasing operational costs of maintaining older systems, cloud systems make sense. A cloud solution solves two issues; alleviating their space constraints and their staffing issues by outsourcing the labor and real estate issues to the cloud provider.
When HPC is viewed as a utility rather than as an investment, cloud becomes very attractive. Just like a power company, the HPC cloud is there when you need it. If the organization only does sporadic HPC jobs, or, has determined that owning hardware is not in the best investment, it makes more sense to rent time. The primary benefit being that the cloud provider will be running the newest and most up to date systems, essentially allowing the organization’s computing dollar to go farther.