From the latest “filesystem test” of btrfs, nilfs, ext4 and xfs:
It took 20 seconds for this database test to complete under EXT3, 34 seconds under NILFS2, but 870 seconds for EXT4! XFS was at 1312 seconds and Btrfs was at 1472 seconds! These results are a bit shocking, but the Phoronix Test Suite does run these tests multiple times to ensure accuracy and statistical significance.
Ok, but the issue here is not statistical noise or averaging problems. ext4 & xfs & btrfs are flushing the drive’s write cache to be sure fsync is on disk; ext3 is not. This came up on their own list, and I explained that it was related to barriers etc. Short memories I guess. Phoronix could be useful if the tests were more meaningful & transparent, and if they engaged in a bit of communication with the experts in the subsystems they are testing. It could have been a teachable moment, but it’s just noise this way. :(