Ted Ts'o: EXT4 Within Striking Distance Of XFS

This round of EXT4 file-system testing against XFS was not like any of our many file-system benchmarks from a desktop or netbook / notebook perspective, but was done with a 48-core AMD system with a 24 SAS-disk hardware RAID array providing a seven Terabyte volume and 256GB of system memory. While the HP Linux file-system test results still show the XFS file-system as leading, when using EXT4 with no journaling and the out-of-tree "use the bio layer directly" patches, the results for this evolutionary file-system are quite close. The XFS file-system generally has performed well on hardware like this where the EXT file-systems have not scaled quite as well, but with EXT4 things are changing.
There's also results from the EXT3 file-system, which trail EXT4 in all tested combinations (vanilla EXT4, EXT4 patched, EXT4 no-journal, and then the EXT4 no-journal patched that did the best). File-system benchmark results were shown for large file creates, random writes, and running a mail server with each test being done under 1, 48, and 192 threads. No results for Btrfs or any other file-system were included on this very powerful system. These tests were also done against the Linux 2.6.36-rc6 kernel.
From reading Ted's blog post, "We're now within striking distance of XFS, and there’s more improvements to ext4 which I have planned that should help its performance even further. This is the kind of performance improvement that I’m totally psyched to see!"
Our most recent desktop testing of the EXT4 file-system (along with Btrfs) indicate performance regressions in the Linux 2.6.36 kernel, while previous to that we also compared these two latest Linux file-systems to the ZFS-FUSE file-system, and when using these file-systems on a solid-state drive. Benchmarks of both EXT4 and Btrfs atop the latest Linux 2.6.37 kernel development code will be available in the coming weeks.
66 Comments