Amazon FSx
Amazon FSx is a family of managed file systems for the cases Amazon EFS does not cover. Where EFS is NFS for Linux, FSx runs full-featured third-party file systems as a service, with AWS handling provisioning, patching, replication and backup.
The four variants
Section titled “The four variants”| Variant | Protocol | Built for |
|---|---|---|
| FSx for Windows File Server | SMB | Windows applications, Active Directory-integrated file shares |
| FSx for Lustre | Lustre | High-performance and parallel computing, analytics, machine learning |
| FSx for NetApp ONTAP | NFS, SMB, iSCSI | Teams that need ONTAP features — snapshots, clones, SnapMirror, tiering |
| FSx for OpenZFS | NFS | Teams that need ZFS semantics: snapshots, clones, compression |
How an FSx file system is put together
Section titled “How an FSx file system is put together”Every variant follows the same shape. AWS runs the file servers and their attached storage; the file system is reached from a VPC through an elastic network interface, and clients — typically EC2 instances, often in an Auto Scaling group — mount it over the relevant protocol. Access control depends on the variant: FSx for Windows File Server authenticates against AWS Managed Microsoft AD or a self-managed Active Directory, while administration of the file system itself is governed by IAM.
Persistent deployments store redundant copies of the data and replace failed infrastructure automatically, so a file server that fails is replaced within minutes while client requests retry.
FSx for Windows File Server
Section titled “FSx for Windows File Server”A fully managed Windows file system, backed by Windows Server and reachable over SMB, including DFS namespaces for spanning shares.
Typical uses:
- Windows line-of-business applications, including SharePoint, SQL Server and IIS
- Home directories and departmental file shares
- Windows content management and media processing
- Analytics on Windows machines
- Any SMB share that needs Active Directory identities and NTFS ACLs
This is the answer whenever an application expects a Windows file share. EFS cannot serve SMB.
FSx for Lustre
Section titled “FSx for Lustre”Lustre is the parallel file system used widely in high-performance computing, and FSx runs it as a managed service. It is built for throughput measured in gigabytes per second across many clients at once.
Typical uses:
- HPC clusters and distributed simulation
- Financial modelling and seismic or genomic analysis
- Machine learning training over large datasets
- Shared storage for thousands of EC2 instances
Deployment types
Section titled “Deployment types”Scratch file systems are for short-lived, processing-heavy work: data is not replicated and does not survive a file server failure, in exchange for high burst throughput and a lower price. Persistent file systems replicate data and replace failed servers automatically, and are the choice for anything longer-lived.
Storage classes
Section titled “Storage classes”FSx for Lustre offers three, and the description of Lustre as an all-SSD service is now out of date:
- SSD provides sub-millisecond access to the entire dataset. Capacity is provisioned and billed whether used or not. Choose it for latency-sensitive work that needs flash performance across all data.
- Intelligent-Tiering is fully elastic — no file system size is specified and billing follows what is stored — and moves data automatically between Frequent Access, Infrequent Access (after 30 days) and Archive Instant Access (after 90 days) tiers, with no retrieval or transition fees. An optional SSD read cache gives sub-millisecond access to hot data. AWS positions it as “the best balance of price and performance for most workloads”, and unlike the provisioned classes it replicates data across multiple Availability Zones.
- HDD suits workloads that need consistent single-digit millisecond latency across all data at lower cost, with an optional SSD read cache sized at 20% of capacity.
SSD and HDD persistent file systems replicate within a single Availability Zone; Intelligent-Tiering replicates across several.
The relationship with S3
Section titled “The relationship with S3”FSx for Lustre links to an S3 bucket through a data repository association. Objects in the linked bucket appear in the file system, and their contents are loaded on first access rather than copied up front; changes written in the file system can be exported back to the bucket. The pattern is to keep the durable copy of a dataset in S3 and stand up a Lustre file system over it for the duration of a compute job. It is a live linkage, not a snapshot or archive destination.
FSx for NetApp ONTAP and FSx for OpenZFS
Section titled “FSx for NetApp ONTAP and FSx for OpenZFS”Both exist for teams that already depend on a specific storage platform’s behaviour rather than merely on a protocol.
FSx for NetApp ONTAP runs ONTAP as a managed service, serving NFS, SMB and iSCSI from the same file system, with ONTAP’s snapshots, thin-provisioned clones, storage efficiency features and SnapMirror replication. It is the usual landing place for a NetApp estate moving to AWS.
FSx for OpenZFS runs OpenZFS over NFS, bringing ZFS snapshots, instantaneous clones and in-line compression. It suits Linux workloads that want ZFS behaviour with managed infrastructure.
Choosing
Section titled “Choosing”The protocol usually decides it. SMB means FSx for Windows File Server. NFS on Linux, with no special requirements, means EFS. Sustained parallel throughput means FSx for Lustre. An existing operational dependency on ONTAP or ZFS features means the matching variant.