Issue link: https://read.uberflip.com/i/1187300
Amazon Web Services – HPC Lens AWS Well-Architected Framework Page 17 nodes to lessen the memory-per-core requirements and to target the best performing instance type. Tightly coupled applications require a homogenous cluster built from similar compute nodes. Targeting the largest instance size minimizes internode network latency while providing the maximum network performance when communicating between nodes. • Deployment: A variety of deployment options are available. End-to- end automation is achievable, as is launching simulations in a "traditional" cluster environment. Cloud scalability allows the opportunity to launch hundreds of large multi-process cases at once, so there is no need to wait in a queue as there is with an on-premises cluster. There are three example architectures to consider as starting points for design patterns for HPC applications: • Persistent cluster • Ephemeral cluster • HPC microservices (a variation of the ephemeral cluster) Persistent Cluster A persistent cluster mimics a traditional on-premises cluster or supercomputer experience. Clusters include a master instance with a scheduler that allows multiple users to submit jobs. The compute node fleet can be a fixed size or tied to an Auto Scaling group to increase and decrease the compute fleet depending on the number of submitted jobs. The cluster is almost identical to that of the HTC cluster except that the cluster requires a placement group and homogenous instance types. In addition to the scheduler and shared volume, other libraries such as MPI may be added to the compute nodes. CfnCluster is an example of a persistent cluster. Reference Architecture