Myth #4: Deploying HPC in the Cloud is Difficult
Editor’s note: This is the 4th blog post of the ebook Dispelling the Myths of Cloud HPC. Read the full piece here.
Historically, many HPC applications have been challenging to deploy because they depended on custom-built and maintained infrastructure. Each application required a specific environment and dedicated tooling that needed to be planned, procured, and then learned by end users and administrators.
Modern HPC applications today make use of HPC-as-a-service (HPCaaS) and automation frameworks to simplify application and infrastructure deployments. IT and engineering tasks that once required weeks, even months, can now be completed in a few days using pre-configured compute environments, utilities, and workflows. Cloud enablement solutions like Rescale work with customers to automate connections between hybrid and multicloud resources as well as complex workflows. With thousands of automations already deployed for various customers, other organizations can realize immediate benefit with Rescale’s turnkey solution.
Previously, IT organizations would have needed to hire a team of specialists to integrate the disparate components that make up an HPC platform, for example: schedulers, license servers, specialized chip architectures, file system, operating system, compilers, and libraries.
Of course, the most expensive element of an IT solution is talent, which, in the HPC industry, is becoming increasingly difficult to find. HPCaaS platforms reduce the need for significant personnel overhead by pre-packaging components in a way that makes it easier for end users of HPC applications to programmatically invoke HPC services. There’s still a need for some level of HPC expertise but now that there are platforms that abstract away the underlying complexity of the HPC environment there are also going to be more developers than ever building HPC applications.
In effect, the cloud and cloud enablement solutions make it possible to shift more time to realizing the benefits of an HPC system instead of setting up and managing it.