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

Optional changes
The following parameters allow users to modify the cluster size and choose spot instances.
[cluster] section
master_instance_type: The default instance for this parameter is c5.large. The master instance does not perform any computation, but it hosts the scheduler and other services. Hence, it can be a small instance (e.g. c5.large or m5.large) unless it is also being used for graphics processing (see advanced options).
compute_instance_type: These are the instances where the apps run on and should be chosen accordingly.
To set the cluster size, there are 2 groups of parameters. The first group uses initial_queue_size, max_queue_size & maintain_initial_size; the second group includes the min_vcpus, desired_vcpus and max_vcpus parameters. We prefer the first option, but the CLUSTERs subdirectory also has templates using the second group.

Case 1

Cluster with a constant number of 4 c5.12xlarge compute instances:
[cluster wrfcluster]
          …
          master_instance_type = c5.large
          compute_instance_type = c5.12xlarge
          cluster_type = spot
          disable_hyperthreading = true
          initial_queue_size = 4
          max_queue_size = 4
          maintain_initial_size = true
          …
This script would launch the cluster with spot instances. Alternatively, it could read:
          [cluster wrfcluster]
          …
          master_instance_type = c5.large
          compute_instance_type = c5.12xlarge
          cluster_type = spot
          disable_hyperthreading = true
          min_vcpus = 192
          desired_vcpus = 192
          max_vcpus = 192
          …

Close Menu