Core concept

Compute environment (spot/on-demand/Fargate mix). Job queue. Job definition (container + command + resources). Submit jobs → auto-scale + execute.

Advertisement

How it works

Managed CE simpler but less control. Unmanaged you provide EC2. Multi-node jobs for MPI workloads. Array jobs for parametric.

Advertisement

Trade-offs + gotchas

Genomics pipelines. Financial risk models. ML training. Video encoding. Any compute-heavy job with clear start/end.