EROFS aims to form a generic read-only file system solution for various read-only use cases (embedded devices, containers and more) instead of just focusing on storage space saving without considering any side effects of runtime performance. [1]
added to Linux 5.4
adding support for file-backed mounts so that file-system image files can be used directly without needing a loopback block device or similar. This file-backed mount support is intended for use with container images and sandboxes. Using file-backed mounts is intended to avoid loopback devices that can behave intricately and can simplify error-prone lifetime management of virtual block devices.
TLDR mount an image from the FS without having to make some brand new thing in memory.[2]
As the FS is being used for massive dataset for AI training, they added 48bit addressing support
from a 32bit that was limited to 16TB, should now land us to 1,024 Petabytes.[4]
Can now use QuickAssist Technology accelerators (QAT) for improving DEFLATE decompression performance.[5]
After it was request by a user, EROFS now supports metadata compression. Now the resulting FS will be way slower and not be much smaller but it was a simple thing to implement and it was requested.(the request: https://issues.redhat.com/browse/RHEL-75783)
Also support for readahead for directories in the readdir function now improves drastically performance. In a readdir test on a large directory with 12,000 sub-files went from handling 926,385 files per second to 2,380,435... Or about a 2.5x improvement. The maintainer, Gao Xiang, also state(speaking about readahead): "We may consider further improvements later to align with ext4's s_inode_readahead_blks behavior for slow devices too." [6]