Open parallel storage for AI, HPC, and research

The OrangeFS project builds scale-out filesystem software for modern GPU clusters, simulation campaigns, and shared research data. Download releases, read the docs, or join development on GitHub.

Upstream Linux kernel client
In mainline since 4.6, so standard nodes can mount the filesystem from the kernel they already run

Open source releases
Source, tarballs, and Windows client on GitHub

Active development
v3 resiliency work, mainline client improvements, and project tooling

Researcher in a modern AI lab with GPU and storage infrastructure
Open source project

Filesystem research that ships

OrangeFS grew out of the PVFS lineage and remains both a research testbed and a production parallel filesystem. Servers and admin tools run in user space; the VFS path uses an upstream Linux kernel client (mainline since 4.6). A separate high-performance user-space interface is also available when you want to bypass the kernel path.

The community cares about real campaign I/O: multi-node training and fine-tuning, MPI-IO heavy simulation, and shared lab datasets that stress metadata as much as bandwidth. Releases target the I/O patterns AI and HPC sites run in production.

Where development is focused

Most active work lives on the projects page. Highlights today:

  • OrangeFS v3: new data model for file and metadata resiliency, flexible server membership, background maintenance
  • Upstream kernel client: mainline Linux integration and readahead improvements, distinct from out-of-tree client add-ons other parallel filesystems require
  • Community tooling: Grafana visualization from built-in stat counters; data staging helpers for research workflows
  • Research threads: searchable distributed metadata, small-file and small-I/O performance, optional client-side caching semantics

Want to propose a feature or report a bug? Use GitHub Discussions or open an issue on the repository.

Modern GPU rack in a contemporary AI research computing room
Multi-node research cluster with compute and storage for shared parallel workloads

Example workloads in the community

OrangeFS is not tied to one vertical, but sites commonly deploy it for:

  • Large-scale simulation: restart checkpoints and time-step output across MPI ranks
  • Shared research storage: a parallel mount for a lab or instrument pipeline
  • AI training and fine-tuning: read-heavy dataset and checkpoint I/O where aggregate bandwidth matters

Enterprise support, installers, and cloud marketplace deployments are on the OrangeFS support site.

OrangeFS 2.10.1

Latest open source release: compatibility updates for current Linux distributions, fixes, and ongoing 2.x maintenance while v3 work continues.

Download 2.10.1 Windows client (64-bit)

View ChangeLog · Documentation · Wikipedia article

OrangeFS is released under the LGPL License; some recent portions are also available under LGPL and Apache 2.

Explore the project

Docs, community, roadmap, and answers in one place.

FAQ

Install, clients, performance, operations, and architecture for OrangeFS today.

Community

Contributors, institutions, and how to get involved.

Projects

v3 roadmap, research ideas, and completed milestones.

Documentation

Build, install, configure, and operate OrangeFS.

Get involved