Andrew Carr from supercomputing vendor Bull weighed in on the benefits of harnessing HPC on demand services versus investing in clusters. His primary reasoning is that since high performance computing requires significant up front and recurring investment, capturing a pay as you go set of cycles can open doors to efficient and flexible computing that shifts with demand.
Carr makes it clear that HPC on demand should not be confused with the more general purpose term “cloud” which he equates to public cloud services from the likes of Amazon or others. He says that “unlike commercial clouds, HPC on demand services address the requirements of HPC customers y providing on-demand remote compute facilities access with a pre-installed and configured environment where ISV applications and open source codes are installed and available.”
In Carr’s view, HPC on demand services are the only way that those with complex modeling and simulation needs can tap into the advanced environments they need. He claims that public clouds are too one-size-fits-all for HPC users and cannot keep pace with the types of applications many who are using HPC (whether they realize it or not) require.
As far as the adoption of HPC on-demand resources, Carr says that there should be broad appeal since many large organizations want extra burst capacity with adding more nodes and overhead simply to accommodate projected peaks in need. He also contends that HPC on demand services are likely to appeal to “SMEs and design offices that have basic computing resources but cannot invest further in these resources and would rather seize the opportunity to outsource hardware and technical support so they can concentrate their own resources on their core business.”
These benefits, in addition to the possibilities this kind of on-demand, pay-as-you-go access offers for the short bursty needs of those in 3D rendering and special effects, make HPC on demand appropriate for a broad base—so broad that Bull has opened its own on-demand services.