XRAID Explained: A New Declustered Parity RAID for Linux and XFS with Up to 20x Faster Writes
Linux kernel developer Christoph Hellwig has publicly released XRAID, a new parity RAID implementation built into the XFS filesystem. It targets faster writes, shorter rebuilds and support for zoned storage. It is also explicitly experimental. Here is what it does, why it is designed this way, and what it means for storage administrators.

What Is XRAID?
XRAID is a new declustered parity RAID implementation that works together with the XFS filesystem. Hellwig announced its initial public release on the Linux kernel mailing list, saying it has been in development for about 18 months with two main goals:
- Enable parity RAID on zoned devices. Zoned storage requires data to be written sequentially within zones, which does not mix well with the in-place updates that classic parity RAID relies on.
- Deliver high performance without the data integrity problems that affect some traditional RAID designs.
Although XRAID is designed primarily for zoned storage, it also works with conventional hard drives and SSDs. Early benchmarks show write performance up to 20 times faster than the existing Linux mdraid implementation, even though SSD optimization was not a priority for this release.
How XRAID Integrates with XFS
XRAID’s main architectural difference is that it lives inside the filesystem stack rather than underneath it as a generic block layer. It builds on XFS’s zoned allocator, which separates file data from metadata using the XFS realtime subvolume:
Placed in sequential-write zones. XRAID applies parity protection to this data, where it is most efficient.
Stored on conventional devices or zones. Its redundancy is handled by existing Linux mechanisms such as MD RAID mirroring.
The reasoning is that metadata consists of smaller objects that need frequent in-place updates, which suits mirroring better than parity. Large file data is written sequentially, which suits parity on zoned layouts. Splitting the two lets each use the protection method that fits it.
What “Declustered” Means for Rebuilds
In a conventional RAID configuration, each stripe spans a fixed set of disks. When one fails, the rebuild is concentrated on the same small group, and the replacement disk becomes a write bottleneck. A declustered design distributes data and parity across a larger pool of storage instead.
According to the announcement, this can greatly reduce rebuild times after disk failures. XRAID’s rebuild behavior includes:
- Reconstruction across multiple devices that have available capacity, rather than funneling everything onto one replacement disk.
- Prioritizing zones that lost multiple levels of redundancy, so the most at-risk data is protected first.
- Rebuilding data zones before parity zones, which reduces the performance impact of degraded reads.
Shorter rebuilds matter because the window during which an array is degraded is exactly when a second failure causes data loss. Faster recovery narrows that window.
Parity Levels and Current Limits
| Capability | Status in the initial release |
|---|---|
| Zero, one or two parity blocks | Supported |
| Three or more parity blocks | Not included yet; developers experimented with the SnapRAID parity library and prepared the code for future expansion |
| Zoned storage devices | Primary design target |
| Conventional HDDs and SSDs | Work, though SSD optimization was not a priority |
| Mainline Linux kernel | Not ready for submission |
Tooling
- A dedicated branch of xfsprogs adds filesystem creation support for XRAID.
- A new
xraidutility manages devices and inspects XRAID filesystems. - Zonar, a zoned filesystem visualization tool, has been extended to show how logical zones map onto physical storage devices.
Current Status: What Administrators Should Know
For practical planning, a sensible approach looks like this:
- Treat it as a research preview. Test only in a lab or on disposable data, using a custom kernel and the dedicated xfsprogs branch.
- Keep production on proven stacks. Continue using mdraid, hardware RAID, ZFS or Btrfs depending on your requirements until XRAID matures and lands upstream.
- Watch for independent benchmarks that compare it with mdraid on your kind of hardware and workload.
- Follow the upstream discussion. Feedback on the mailing list is exactly what the project is asking for, and the review process will show how stable the design is.
If you run Linux storage servers or NAS builds alongside the rest of your infrastructure, see our other guides on VMoreCloud .
Frequently Asked Questions
What is XRAID?
A new declustered parity RAID implementation for Linux that is integrated with the XFS filesystem, announced by kernel developer Christoph Hellwig after about 18 months of development.
Is XRAID really 20x faster than mdraid?
Early benchmarks from the developer show write performance up to 20 times faster than mdraid. That is a best-case figure from an experimental release, so independent testing on your own workload is needed before relying on it.
Can I use XRAID in production?
No. It is experimental and the developer states it is not ready for submission to the mainline Linux kernel.
Does XRAID only work with zoned devices?
No. It is designed primarily for zoned storage but also works with conventional hard drives and SSDs.
How many parity blocks does XRAID support?
Zero, one or two in the initial release. Support for three or more has been prepared in the code but is not yet included.
Where can I read the technical details?
The XRAID kernel source, the matching xfsprogs and test repositories, and Hellwig’s FOSSY 2026 presentation are linked in the sources below.
Sources and Further Reading
- Christoph Hellwig’s XRAID announcement (Linux kernel mailing list)
- XRAID kernel source (xfs-raid branch)
- Hellwig’s FOSSY 2026 presentation: architecture and benchmarks (PDF)
- Linuxiac: Linux Gets a New XRAID Implementation with Up to 20x Faster Writes
Details in this article are based on the developer’s announcement as reported by Linuxiac. Because XRAID is under active development, verify current behavior against the project repositories before testing.
