Running Simulations in the Cloud (AWS)

To accelerate simulations, the model can be executed on cloud virtual machines. For example, as of this writing, a 96 Core machine (c5.24xlarge) can be rented from Amazon Web Services for ~$1.5/hr (spot price). The basic usage cycle involves the following steps:

• Request a ‘spot’ instance from AWS (spot instances are ~70% less expensive)

• SSH into the instance, download the model docker image, run the simulation

• Upload simulation results for later analysis

• Shut down instance (per-second billing)

A similar process can be followed with other cloud providers (GCP, Azure, etc…).

Private Cloud MPI Clusters

Similarly, multiple instances could be started to form a private MPI cluster, and the simulation could be distributed across multiple instances. See AWS ParallelCluster.