Current development, research ideas, and completed milestones in the OrangeFS ecosystem.
A living list of work recently finished, currently underway, and planned next. Use it to see where help is useful and what is already shipping.
Majority of active work targets Version 3: a new data model for file and metadata resiliency, more flexible server membership, and background maintenance processes.
Helpers to stage research data to and from OrangeFS from disparate sources so computation starts with data already where the jobs run.
Visualize OrangeFS utilization and performance by leveraging stat counters built into OrangeFS.
Most users encode metadata in directory hierarchies, which limits discovery of extended attributes. This idea targets search over standard and extended attributes without brute-force scans, using distributed metadata.
Optional client-side caching middleware with configurable consistency, plus design work on storing smaller files in distributed metadata so file data can return with metadata for small files.
Upstream kernel attribute caching significantly reduces stats-over-the-wire activity. Linux 4.9 and newer benefit automatically.
LMDB is now the default metadata option and shows significant performance improvement over BDB. Note: there is no in-place upgrade path from BDB; new filesystems are required.
POSIX-style access control with signed credentials and capabilities for distributed environments, addressing user-level identity spoofing risks in the original PVFS design.
Infrastructure for duplicating physical data across servers with automatic failover when marking files immutable via chattr or pvfs2-xattr.
Multi-layer direct interface (POSIX, stdio, MPI-IO) bypassing the kernel for maximum performance, under src/client/usrint in the source tree.
Directory entries spread across multiple metadata servers via hash-based distribution (GIGA+ approach), enabling parallel access to very large directories.