XRAID declustered parity RAID released for zoned XFS storage

Longtime Linux storage developer Christoph Hellwig has published the initial public release of XRAID, a new declustered parity RAID implementation built natively into the XFS file system. The code targets zoned XFS deployments, particularly SMR hard drives, but also functions on conventional HDDs and SSDs. XRAID has been under development for approximately 18 months and is now available as kernel patches for review along with corresponding xfsprogs user-space utilities including a new mkfs implementation and dedicated xraid management tool.

The architecture separates file data from metadata using the RT subvolume, placing only file data on sequential-write-required zones while metadata resides on conventional devices or zones. Parity RAID applies exclusively to file data, with metadata redundancy handled by existing stacked Linux drivers. The implementation sits below the Zoned XFS I/O path, aggregating physical zones into logical zones presented as realtime groups to XFS.

XRAID architecture and declustered parity design

• Zoned allocator integration: XRAID leverages the RT subvolume to isolate file data from metadata, ensuring only file data occupies sequential-write-required zones while metadata uses conventional storage, matching the underlying zoned device constraints.
• Declustered parity distribution: Data and parity blocks in each stripe spread across a pool larger than D + P disks, enabling all disks to participate in reads, writes, and reconstruction while distributing spare capacity throughout the pool rather than dedicating hot spares.
• Metadata redundancy delegation: Small in-place-updated metadata objects bypass parity RAID entirely, instead using existing stacked Linux drivers such as MD RAID mirroring, avoiding the inefficiency parity RAID imposes on small random writes.
• Dynamic rebuild prioritization: Reconstruction targets all devices with spare capacity rather than fixed replacement drives, with zone ordering that prioritizes multiple-redundancy-loss zones and rebuilds data zones before parity zones to minimize degraded read overhead.
• Performance positioning: Early benchmarks indicate write throughput up to 20 times faster than mdraid, though SSD-specific optimization remains a future focus rather than current priority.

Why it matters

XRAID addresses a longstanding gap in Linux storage by providing native RAID capabilities within XFS rather than relying solely on stacked block-layer solutions, potentially simplifying deployment and management for high-capacity zoned storage arrays. For data center operators and storage vendors adopting SMR drives, the declustered architecture offers substantially faster rebuild times and better utilization of large disk pools compared to conventional RAID approaches. The accompanying xfsprogs branch and xraid utility indicate Hellwig is building complete user-space support, though the explicit note that patches are not yet upstream-ready means production adoption will require additional testing cycles and kernel integration work.