Amazon FSx for Lustre

• Fully managed parallel distributed file system for high-performance computing.

• Name derives from Linux + Cluster.

• Ideal for workloads requiring massive throughput and low latency.

Use Cases

• Machine Learning training pipelines.

• HPC (High Performance Computing) simulations.

• Video rendering and processing.

• Financial modeling, electronic design automation.

Performance & Storage

• Scales to hundreds of GB/s throughput, millions of IOPS, and sub-millisecond latency.

• SSD – Low-latency, IOPS-heavy, small/random file access.

• HDD – High-throughput, large/sequential file operations.

S3 Integration

• Can mount S3 data as a file system.

• Can write computation results back to S3.

• Accessible from on-premises via VPN or Direct Connect.

Deployment Options

• Scratch:
• Temporary, non-replicated storage.
• Very high burst performance (up to 200 MB/s per TiB).
• Use for short-term or disposable workloads.

• Persistent:
• Long-term storage, replicated within the same AZ.
• Automatic recovery from hardware failures.
• Use for sensitive or long-lived workloads.