Deployment of AWS cluster with custom odyhpc.com AMIs   (ii – cluster design)

Case 2

For arm-powered clusters, both master and compute instances must be AArch64 instances. For example, the script for a cluster with 4 c5.12xlarge compute instances would read:
        [cluster wrfarmcluster]
        …
        master_instance_type = a1.large
        compute_instance_type = c6g.12xlarge
        cluster_type = ondemand
        initial_queue_size = 4
        max_queue_size = 4
        maintain_initial_size = true
        …
or
        [cluster wrfarmcluster]
        …
        master_instance_type = a1.large
        compute_instance_type = c6g.12xlarge
        cluster_type = ondemand
        min_vcpus = 192
        desired_vcpus = 192
        max_vcpus = 192

Case 3

If, instead of having a fixed number of instances, the objective is to have a variable number of instances controlled by the scheduler, the script would read:
        [cluster wrfcluster]
        …
        master_instance_type = c5.large
        compute_instance_type = c5.12xlarge
        cluster_type = spot
        disable_hyperthreading = true
        initial_queue_size = 2
        max_queue_size = 8
        maintain_initial_size = false
        …
With this script, the cluster launches with 2 instances, and will terminate or spin up more instances (up to 8) depending on the scheduler load.

Close Menu