Containers
Amazon EKS
Amazon EKS provides managed Kubernetes control planes for containerized applications. It supports GenAI platforms that need Kubernetes operators, scheduling controls, GPU nodes, or portable serving frameworks.
Key points
- Managed node groups and Fargate offer distinct worker models
- IAM roles for service accounts or EKS Pod Identity scope pod permissions
- Kubernetes autoscaling needs workload metrics and available node capacity
When to use it
- Operate a Kubernetes-native model-serving stack
- Schedule GPU inference and evaluation jobs with specialized operators
Exam tips
- Use EKS when Kubernetes capabilities are a requirement; ECS is simpler for AWS-native containers
- Coordinate pod and node scaling because pending accelerator pods cannot be solved by replicas alone