AWS Storage
FSx
Managed file systems for Windows, Lustre, NetApp ONTAP, and OpenZFS
Amazon FSx is a family of fully managed file systems that provide native compatibility with popular file system types: Windows File Server, Lustre (HPC), NetApp ONTAP, and OpenZFS. FSx handles the complexity of deploying and managing specialized file systems, enabling lift-and-shift migrations and high-performance computing workloads without operating the underlying infrastructure. Choosing the right FSx variant requires understanding the specific protocol requirements and performance characteristics of your workload.
The Four FSx Variants - When to Use Each
Each FSx variant is optimized for a specific use case and protocol. The wrong choice leads to either unnecessary cost or missing capabilities.
| FSx Variant | Protocol | Primary Use Case | Key Differentiator |
|---|---|---|---|
| FSx for Windows File Server | SMB | Windows apps, .NET, SharePoint, SQL Server user data | Active Directory integration, DFS Namespaces, shadow copies |
| FSx for Lustre | Lustre (POSIX) | HPC, machine learning, financial modeling, video processing | Sub-millisecond latency, hundreds of GB/s throughput, S3 integration |
| FSx for NetApp ONTAP | NFS, SMB, iSCSI | Enterprise storage migration, multi-protocol, SnapMirror DR | Familiar ONTAP features: deduplication, compression, snapshots, tiering |
| FSx for OpenZFS | NFS | Lift-and-shift ZFS workloads, Oracle databases, DevOps tools | ZFS data management features: clones, snapshots, compression, NFS v3/v4 |
FSx for ONTAP is the most feature-rich option and supports SMB, NFS, and iSCSI simultaneously - making it the best choice for mixed Windows/Linux environments or teams already familiar with NetApp.
FSx for Lustre - HPC and ML Workloads
FSx for Lustre is purpose-built for workloads that require fast access to large datasets. It can serve as a high-speed cache in front of S3, processing data at hundreds of GB/s.
| Deployment Type | Description | Use Case |
|---|---|---|
| Scratch | Temporary, non-persistent, higher throughput per TB | Short jobs, can tolerate data loss, HPC burst |
| Persistent 1 | Data replicated within single AZ, self-healing | Long-running workloads, tolerate single point of failure |
| Persistent 2 | Higher throughput, SSD-backed, multi-AZ option | Mission-critical, ML training pipelines |
The S3 integration is one of the most powerful FSx for Lustre features - datasets in S3 are imported into the file system on demand and results can be exported back:
# Create an FSx for Lustre file system linked to an S3 bucket
aws fsx create-file-system \
--file-system-type LUSTRE \
--storage-capacity 1200 \
--subnet-ids subnet-xxxxxxxxx \
--lustre-configuration \
ImportPath=s3://my-training-data/,\
ExportPath=s3://my-training-data/results/,\
DeploymentType=PERSISTENT_2,\
PerUnitStorageThroughput=500
# Export changed data back to S3
aws fsx create-data-repository-task \
--type EXPORT_TO_REPOSITORY \
--file-system-id fs-xxxxxxxxx \
--paths /results/FSx for Lustre integrates natively with SageMaker, AWS Batch, and EC2 HPC instances (C5n, P4d, etc.). For ML training, it provides the I/O throughput needed to keep GPUs fed with data.
FSx for Windows File Server - Enterprise Integration
FSx for Windows File Server provides a fully managed Windows-native file system with Active Directory integration, shadow copies, and SMB protocol support - enabling lift-and-shift of Windows workloads.
| Feature | Detail |
|---|---|
| Active Directory | Integrates with AWS Managed AD or self-managed on-premises AD |
| Deployment type | Single-AZ or Multi-AZ (synchronous replication, automatic failover) |
| Shadow copies | Automated user-accessible previous versions - hourly/daily schedule |
| DFS Namespaces | Consolidate file shares across multiple file systems into one namespace |
| Throughput | 8 MB/s to 2 GB/s depending on throughput capacity setting |
| Storage type | SSD (default) or HDD |
| Backup | Daily automatic backups to S3, configurable retention |
# Create Multi-AZ Windows File Server joined to AWS Managed AD
aws fsx create-file-system \
--file-system-type WINDOWS \
--storage-capacity 300 \
--storage-type SSD \
--subnet-ids subnet-aaaa subnet-bbbb \
--windows-configuration \
ActiveDirectoryId=d-xxxxxxxxx,\
ThroughputCapacity=64,\
DeploymentType=MULTI_AZ_1,\
PreferredSubnetId=subnet-aaaa,\
AutomaticBackupRetentionDays=30,\
CopyTagsToBackups=trueFSx for NetApp ONTAP - Enterprise Migration
FSx for NetApp ONTAP enables organizations running NetApp on-premises to migrate to AWS while retaining familiar management interfaces, tools, and ONTAP features.
| ONTAP Feature | AWS FSx for ONTAP Support |
|---|---|
| Protocols | NFS v3/v4, SMB 2.0/3.0, iSCSI - simultaneously on same volume |
| SnapMirror | Replicate to/from on-premises ONTAP or another FSx ONTAP |
| Snapshots | Space-efficient, point-in-time copies - up to 1,023 per volume |
| Deduplication and compression | Inline and background - significantly reduces storage costs |
| Tiering | Automatically tier cold data to S3 (capacity pool pricing) |
| FlexClone | Instantly create writable clones for dev/test |
| Multi-protocol | Same volume accessible via NFS from Linux and SMB from Windows |
The tiering capability is key for cost optimization - hot data stays on SSD-backed primary storage while cold data is automatically moved to S3-backed capacity pool storage at ~$0.019/GB vs ~$0.115/GB for primary.
FSx Pricing Comparison
FSx pricing varies significantly by variant and deployment type. Understanding the cost structure helps avoid surprises.
| FSx Variant | Storage Cost (approx) | Additional Costs |
|---|---|---|
| Windows FS (SSD) | $0.23/GB-month | Throughput capacity $2.20/MBps-month, backups $0.05/GB |
| Windows FS (HDD) | $0.025/GB-month | Throughput capacity $2.20/MBps-month |
| Lustre Scratch | $0.14/GB-month | No additional - simple pricing |
| Lustre Persistent 2 | $0.145/GB-month (500 MBps/TiB) | Backup $0.05/GB, data repo task requests |
| ONTAP (SSD) | $0.115/GB-month | Capacity pool (tiered to S3) $0.019/GB, throughput $0.10/MBps |
| OpenZFS (SSD) | $0.09/GB-month | Throughput capacity, backup storage |
FSx for Windows and FSx for ONTAP charge separately for throughput capacity (in MB/s) which can be a significant portion of the total cost. Always include throughput costs in your estimates.
Interview Focus Points
- 1A company is migrating a Windows HPC workload to AWS that uses SMB shares - which FSx variant would you recommend?
- 2How does FSx for Lustre integrate with S3 for ML training pipelines?
- 3When would you choose FSx for ONTAP over EFS for a Linux workload?
- 4What is the difference between FSx for Lustre Scratch and Persistent deployment types?
- 5A company running NetApp on-premises wants to migrate to AWS with minimal operational changes - what is your recommendation?
- 6How do you provide high availability for FSx for Windows File Server?
- 7Explain how FSx for ONTAP's tiering feature works and when you would use it.