Issue link: https://read.uberflip.com/i/1187300
Amazon Web Services – HPC Lens AWS Well-Architected Framework Page 19 resources are provisioned and launched, data is placed on the nodes, jobs are run across multiple nodes, and the case output is retrieved automatically or sent to Amazon S3. Upon completion of the job, the infrastructure is terminated. Clusters designed this way treat infrastructure as code and allow for complete version control of infrastructure changes. Master nodes and job schedulers are less critical and often not used at all with an ephemeral cluster. The Auto Scaler, a mainstay of the traditional cloud cluster, is also not used because clusters are stood up once and then terminated. Reference Architecture Figure 6: Example ephemeral-deployed tightly coupled HPC cluster Workflow steps: 1. Users deploy each job or workload using a shell script using AWS CLI calls. 2. Each workload is launched onto its own customized cluster. 3. Output data is copied to Amazon S3 or securely copied back to the user's desktop. 4. Upon completion, the ephemeral cluster is terminated entirely.