◀ Download development team's most recent roadmap presentation.
A Brief History of PVFS
In 1993, with the help of a grant from NASA, the Parallel Virtual File System (PVFS) was born. Version "0" was based on Vesta, a parallel file system developed by IBM.
In 2003, a completely overhauled PVFS Version 2 (PVFS2) featured object servers, distributed metadata, file views based on the MPI protocol for high performance computing, and an architecture friendly to experimentation and extensibility.
In recent years, various contributors, charter developers still among them, continued to improve PVFS performance, and in 2007 Argonne National Laboratories chose the file system for its IBM Blue Gene/P, a super computer sponsored by the U.S. Department of Energy.
In 2008 a vision to address the growing needs of mainstream cluster computing spurred a new direction on an accelerated level.
Out of this evolution, the developers adopted OrangeFS as the new free and open brand for its next-generation virtual file system.
OrangeFS, the Next Evolution
What the PVFS2 foundation provided:
- An advanced parallel file system
- Built on a modular architecture
- Support for standard Linux kernels (no custom kernels needed)
- Distributed metadata
- Highly scalable storage, excellent for large data-sets and large files
- Performance significantly faster than NFS using MPI-IO interface
- Unified directory/namespace for unstructured data storage
What OrangeFS has already added:
- Commercial-grade support
- Professional development team focused on OrangeFS
- Supports metadata location for SSDs
- Improved metadata performance for small I/O
- Replicated data for immutable files
- Windows client (with certificate authentication ability)
- WebDAV Apache Module
What is currently being added:
- Distributed metadata for directories
- Enhanced client libraries
- Capability based security
- Client-side caching
OrangeFS Moves to Enterprise
OrangeFS ‘Next‘ (through 2012):
- Fully configurable, redundant data and metadata
- Seamless server-to-server migration
- Optimal data management through well-defined hiearchy of storage types
- Direct interface to system libraries and caching
- Enhanced management tools
- 128-bit object handles
- Peer server awareness for configuration
- Flow splitting for real-time redundancy
- Designed to meet Enterprise class requirements in security and stability while providing HPC level I/O
PFXS (Parallex FS) – Exascale research based on OrangeFS:
- Research work with Parallex (T. Sterling, W. Ligon)
- Structured similar to PGAS (Partitioned Global Address Space); highly object-oriented (C++)
- Use of active messages to invoke threads; futures synchronization
- Close integration, blurring the line between memory and file system