EXT4 Lands Optional Buffered IOmap For Significant Performance Win With Large Writes

The EXT4 file-system changes were merged Sunday night for the ongoing Linux 7.4 merge window. The headline feature of EXT4 in this new kernel version is opt-in support for buffered IOmap usage that can lead to "significant" performance improvements for large writes. A few weeks back I wrote about EXT4's bufered IOmap support coming about with some significant performance potential by replacing the file-system driver's buffer head usage with the IOmap framework. For large write some very sizable performance improvements while for read performance it's more modest. Ted Ts'o wrote of the buffered IOmap support in the EXT4 pull request: "Add support for buffered iomap support in ext4. This is not enabled by default, but requires the buffered_iomap mount option, as this mode does not support online defragmentation, inline data, fsverify, fscrypt, non-extent inodes, and data=journal mode. If buffered_iomap is enabled, ext4 will fall back to the original I/O code path automatically if necessary. Large writes show significant performance improvements when buffered iomap is enabled. There are also miscellaneous bug fixes and improvements, including for fast commit and avoiding deadlocks and out-of-bound reads when handling corner cases involving corrupted file systems." Zhang Yi of Huawei led the work on this buffered IOmap support and showed the huge gains from its usage: So be aware of the "buffered_iomap" mount option for EXT4 when moving to Linux 7.4.