Amazon FSx for Lustre
High-speed file system for HPC
A managed Lustre file system for very high throughput and low latency; it can link to an S3 bucket to read data as files and write results back to S3.
- Cost$$ Medium
- Difficulty●●●●●
- Costs money when idle?Yes
- Managed by AWS
- In examsSAA
Think of it asLike a multi-lane highway running straight from the raw materials warehouse to the production line.
When to use it
✓ Use it when
HPC, ML training, rendering and genomic analysis that need extremely fast parallel data reads.
✗ Avoid it when
Ordinary shared file storage → EFS; Windows file shares → FSx for Windows File Server.
Common uses
- ML training reading datasets from S3
- HPC simulation with ParallelCluster
- Film rendering, video processing
What does it cost?
- Pricing model: Charged by capacity and provisioned throughput level (different for scratch and persistent), plus backups.
⚠️ Cost traps
Scratch is cheaper but doesn't replicate data, so a failure means data loss; keeping a persistent file system around when unused still costs money.
