Choose the right FSx file system (Windows, Lustre, ONTAP, OpenZFS) for specialized workloads requiring high performance or enterprise storage features.
FSx — Specialized High-Performance File Systems
Amazon FSx provides fully managed versions of popular file systems for workloads needing specific file system capabilities beyond what EFS provides.
FSx for Windows File Server
Fully managed Windows SMB file shares with Active Directory integration. Perfect for Windows workloads that need NTFS file system features.
- Supports Windows ACLs, NTFS permissions, DFS (Distributed File System)
- Integrates with Active Directory for user authentication
- SMB protocol — works natively with Windows applications
- Multi-AZ deployment for high availability
- Use case: lift-and-shift of Windows file servers to AWS without application changes
FSx for Lustre — HPC and ML
Lustre is a high-performance parallel file system used for machine learning training and HPC workloads requiring massive throughput.
- Delivers hundreds of GB/s throughput with sub-millisecond latency
- Natively integrates with S3 — lazy-loads files from S3 on first access
- SageMaker uses FSx for Lustre for ML training datasets (much faster than S3 direct)
- Use case: ML training that needs fast local file access to TB of training data
FSx for NetApp ONTAP
Enterprise storage platform with advanced data management capabilities.
- Multi-protocol: NFS, SMB, iSCSI simultaneously
- Advanced features: snapshots, clones, deduplication, compression, SnapMirror replication
- Lift-and-shift: applications using NetApp on-premises work unchanged on AWS
| FSx Type | Protocol | Best For |
|---|
| Windows File Server | SMB | Windows apps, Active Directory integration |
| Lustre | POSIX/NFS | HPC, ML training, video processing |
| NetApp ONTAP | NFS+SMB+iSCSI | Enterprise storage, lift-and-shift from NetApp |
| OpenZFS | NFS | Linux workloads needing ZFS snapshots and clones |
Exam Tip: For the exam: FSx for Windows = Windows SMB + Active Directory. FSx for Lustre = ML + HPC + S3 integration. These are the two most tested. Know that Lustre integrates with S3 — data is lazily loaded from S3, making it much faster for ML training than reading from S3 directly.