Set up your execution environment#

Choose a system supported by the workload, then configure its resources and paths.

Standalone#

For local Fio execution, install fio and copy the system example:

cp conf/common/system/standalone_system.toml standalone.toml

Use Run your first experiment to verify this path. Standalone support varies by workload; check Workloads.

Slurm#

On the submit host, prepare Prepare workload prerequisites, then copy:

cp conf/common/system/example_slurm_cluster.toml slurm.toml

Edit slurm.toml for your cluster:

  • Set partition, node counts, and GPU counts.

  • Make output, install, model, and input paths accessible on the execution nodes.

  • Add container mounts for data outside the default paths.

Start with one case. Use Run on Slurm for allocation modes, scheduler arguments, and tracing.

Kubernetes#

Use Run on Kubernetes to configure cluster access and networking, then run Dynamo or NCCL. CloudAI generates the workload’s Kubernetes resources and collects results.

Check Workload and system support before choosing a system. Configuration fields for all systems are in Systems.

On this page