cd /news/artificial-intelligence/erofs-with-linux-7-2-better-handles-… · home topics artificial-intelligence article
[ARTICLE · art-36526] src=phoronix.com ↗ pub= topic=artificial-intelligence verified=true sentiment=↑ positive

EROFS With Linux 7.2 Better Handles Large Sparse AI Datasets, More Efficient I/O

EROFS, the open-source read-only file-system, has been updated in Linux 7.2 with optimized chunk mapping for more efficient I/O and added sparse support for pcluster layout to better handle large sparse AI datasets. The FSCACHE back-end has been removed, with similar functionality now provided via file-backed mounts and fanotify pre-content hooks.

read2 min views5 publishedJun 23, 2026
EROFS With Linux 7.2 Better Handles Large Sparse AI Datasets, More Efficient I/O
Image: Phoronix (auto-discovered)

The EROFS open-source read-only file-system has some nice enhancements in place for the

First up, EROFS now has optimized mapping of requests for chunk-based inodes. The new EROFS chunk mapping code has been optimized for more efficient I/O performance. There are no performance numbers indicated with the erofs_map_chunks() patch but the I/O performance is simply reported to be more efficient without quantifying.

The other big EROFS change is that sparsee support has been added to the pcluster layout code. The motivation here is on helping large, sparse AI datasets. Alibaba engineer and EROFS maintainer Gao Xiang explained of the sparse support for pcluster layout in

Meanwhile EROFS previously marked its FSCACHE back-end as deprecated and it's been removed with Linux 7.2. EROFS with FSCACHE was originally intended to provide image lazy pulling functionality. But since FSCACHE later made NETFS a hard dependency, that's what led to EROFS deprecating the feature and now removing it. Similar functionality has since been implemented with file-backed mounts and fanotify pre-content hooks.

More details on these now-merged EROFS file-system updates for Linux 7.2 via

Linux 7.2kernel.First up, EROFS now has optimized mapping of requests for chunk-based inodes. The new EROFS chunk mapping code has been optimized for more efficient I/O performance. There are no performance numbers indicated with the erofs_map_chunks() patch but the I/O performance is simply reported to be more efficient without quantifying.

The other big EROFS change is that sparsee support has been added to the pcluster layout code. The motivation here is on helping large, sparse AI datasets. Alibaba engineer and EROFS maintainer Gao Xiang explained of the sparse support for pcluster layout in

the patches:"Although zeros can be compressed transparently on EROFS using fixed-size output compression so that it is never prioritized in the Android use cases, indicating entire pclusters as holes is still useful to preserve holes in the sparse datasets; otherwise overlayfs will allocate more space when copying up, and SEEK_HOLE won't report any hole.

This patch introduces two ways to mark a pcluster as a hole."

Meanwhile EROFS previously marked its FSCACHE back-end as deprecated and it's been removed with Linux 7.2. EROFS with FSCACHE was originally intended to provide image lazy pulling functionality. But since FSCACHE later made NETFS a hard dependency, that's what led to EROFS deprecating the feature and now removing it. Similar functionality has since been implemented with file-backed mounts and fanotify pre-content hooks.

More details on these now-merged EROFS file-system updates for Linux 7.2 via

── more in #artificial-intelligence 4 stories · sorted by recency
── more on @erofs 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/erofs-with-linux-7-2…] indexed:0 read:2min 2026-06-23 ·