Storage
Amazon EFS
Amazon EFS provides elastic shared NFS file systems for Linux workloads across Availability Zones in a Region. Multiple inference or processing tasks can read common model artifacts without copying them into every container image.
Key points
- Regional file systems provide multi-AZ availability while One Zone lowers cost with narrower resilience
- Performance and throughput modes address different access patterns
- Access points enforce application-specific roots and POSIX identities
When to use it
- Share model weights among an autoscaled inference fleet
- Mount a common document-processing workspace from ECS tasks
Exam tips
- Choose EFS for shared files, EBS for block storage, and S3 for object-scale durable artifacts
- Measure model startup reads and provision throughput when burst credits would miss latency targets