Use cases
Common reasons teams put BlixtFS in front of a bucket.
AI and machine learning training
Training jobs read the same large dataset from many nodes, often with frameworks that expect a local path. Mount the dataset bucket read-only on every node. BlixtFS reads large ranges in parallel, and Enterprise and High Performance deployments can warm the cache before a run.
Relevant pages: Read-only buckets, Scale-out.
AI inference
Inference jobs typically read a smaller number of large model files. BlixtFS was specifically designed to handle these workloads:
- High throughput: BlixtFS can read large files in parallel from object stores.
- Lower costs: BlixtFS can re-use the local SSDs that typically come bundled with GPUs.
- Redundancy and high performance: Files in BlixtFS can be replicated to any number of disks. Reads use all disks in parallel.
- Caching guarantees: Deployments can warm the cache before starting.
Relevant pages: Scale-out.
Media and video production
BlixtFS can serve object storage video archives straight to editing suites on MacOS or Windows. The High Performance Edition enables parallel reads from object stores, as well as reading from multiple BlixtFS servers in parallel.
Office file shares
A bucket costs far less than a file server and never runs out of space. BlixtFS presents it as an NFS or SMB share with ordinary folders and permissions. People keep working the way they do today.
Backups
- Existing backups on object stores are easy to access.
- Desktop computers can back up directly to object stores.
Relevant page: Connecting clients.
Multi-cloud and hybrid access
One BlixtFS deployment can serve buckets from several providers side by side, for example an S3 bucket next to a GCS bucket. Applications see one directory tree. Moving data between clouds becomes a file copy.
Relevant page: Cloud providers.
Inference and model serving
Model weights live in a bucket, but inference servers load them from a path. BlixtFS caches them close to the servers after the first load.
Sharing public datasets
A public dataset bucket can be served read-only and anonymously, without cloud credentials of your own, and without BlixtFS ever writing to it.
Relevant page: Read-only buckets.