FSx: Windows File Server and Lustre

Amazon FSx is a family of fully managed file systems on AWS built on industry-standard technologies. While EFS provides a POSIX NFS file system for Linux, FSx provides purpose-built file systems for specific workloads that need native protocol support or specialised performance characteristics. The four FSx flavours are: FSx for Windows File Server (SMB), FSx for Lustre (HPC), FSx for NetApp ONTAP, and FSx for OpenZFS. The SAA-C03 exam primarily tests Windows and Lustre.

Amazon FSx for Windows File Server is a fully managed, Windows-native SMB file system built on Windows Server. It supports all Windows-specific features: SMB protocol (versions 2.0-3.1.1), Active Directory integration, Distributed File System (DFS) namespaces, Windows ACLs, NTFS, user quotas, and shadow copies. It is the AWS-native replacement for on-premises Windows file servers and is specifically designed for Windows applications that require SMB access — including Microsoft SharePoint, SQL Server, home directories, and custom Windows .NET applications.

FSx for Windows integrates with AWS Managed Microsoft Active Directory or your self-managed AD (on-premises or on EC2). All users authenticate using their AD credentials, and you manage file permissions using Windows ACLs — exactly like an on-premises Windows file server. This makes FSx for Windows the most straightforward lift-and-shift target for Windows file shares. When connecting to a self-managed AD, FSx requires network connectivity to domain controllers via Direct Connect or VPN.

FSx for Windows has two deployment types: Single-AZ (one file server in one AZ — 99.9% availability SLA, cheaper) and Multi-AZ (synchronous replication to a standby file server in a second AZ with automatic failover — 99.99% availability SLA). Multi-AZ requires two subnets in different AZs and uses a floating IP that automatically migrates to the standby during failover. For production Windows workloads requiring HA, always choose Multi-AZ. Single-AZ is suitable for dev/test environments.

FSx for Windows supports automatic daily backups (retained 0-90 days) and on-demand manual backups stored in S3. Backups are incremental and application-consistent because FSx uses VSS (Volume Shadow Copy Service). Shadow copies provide previous-version access to files — end users can right-click a file in Windows Explorer and restore a previous version without contacting IT. Shadow copies are stored on the file system itself and are distinct from FSx backups.

Amazon FSx for Lustre is a fully managed, high-performance parallel file system based on the open-source Lustre file system. Lustre is the file system of choice for supercomputers and HPC clusters. FSx for Lustre delivers sub-millisecond latency, hundreds of GB/s throughput, and millions of IOPS — making it ideal for machine learning training, genomics, financial simulations, seismic analysis, and video rendering. It integrates natively with Amazon S3 as a backing data repository.

FSx for Lustre can be linked to an S3 bucket as a data repository. Files in S3 appear in the Lustre namespace immediately (lazy loading — actual data is copied into Lustre on first access). When your HPC job completes, you can export results back to S3 with a data repository task. This pattern allows you to store large datasets cost-effectively in S3, spin up a Lustre file system for the duration of a compute job, and then destroy the Lustre system — paying only for the compute time.

FSx for Lustre has two main deployment types: Scratch (temporary, non-replicated, highest throughput-per-price — for short-running batch jobs where data loss is acceptable because all data comes from S3) and Persistent (replicated within AZ, designed for long-running HPC workloads that need resilience). Persistent 2 is the current generation, offering PerUnitStorageThroughput of 125, 250, or 500 MB/s/TiB. For the SAA-C03 exam: Scratch = cost-optimised temporary jobs, Persistent = long-running or sensitive data.

FSx for Lustre is mounted on Linux instances using the Lustre client. Amazon Linux 2 and Amazon Linux 2023 include the Lustre kernel module. After creating mount targets and configuring the security group to allow Lustre protocol traffic (port 988 and port range 1018-1023), you mount using the lustre file system type. The mount point provides full POSIX semantics. For ML training, mount the file system on all training instances — they all see the same dataset without data copying.

FSx for NetApp ONTAP provides multi-protocol access (NFS, SMB, and iSCSI) with advanced ONTAP features including thin provisioning, data deduplication, compression, and SnapMirror replication — ideal for enterprises migrating NetApp storage to AWS without re-architecting. FSx for OpenZFS provides NFS-accessible storage with ZFS's built-in data integrity checking, compression, and snapshots. Both support Linux and Windows clients. For the SAA-C03 exam, ONTAP is the answer when the scenario mentions 'existing NetApp storage' or 'multi-protocol file access'.

Key selection criteria: EFS — Linux/POSIX NFS, elastic capacity, multi-instance sharing, Lambda support. FSx for Windows — Windows clients needing SMB, Active Directory integration, Windows ACLs, NTFS — the only option for Windows applications that require native Windows file server features. FSx for Lustre — HPC, ML training, genomics, video rendering — when you need hundreds of GB/s throughput and millions of IOPS with S3 integration. The SAA-C03 exam tests these distinctions through workload scenario questions.

Test your understanding of AWS Solutions Architect (SAA-C03) concepts from this lesson.

In this lesson you learned: FSx for Windows File Server provides native SMB with Active Directory integration for Windows workloads requiring NTFS, ACLs, and shadow copies, FSx for Lustre delivers millions of IOPS and hundreds of GB/s for HPC and ML training with direct S3 data repository integration, and Scratch deployments are cost-optimised for temporary jobs while Persistent deployments suit long-running resilient workloads. Next up we explore caching with ElastiCache.