Sysbench file benchmark
Filesystem I/O performances are a difficult thing to profile. For this reason, I run another set of sequential and random I/O transfer benchmarks using the sysbench utility. Sequential speed tests were run with 2 MB big blocks, while random speed with 4 KB blocks.
Let's start with sequential speed:
We don't have a clear winner in this test.
Now, random speed:
This test constantly give us very strange results when paired with a XFS filesystem, so take these scores with a (very big) grant of salt.
Sysbench file I/O benchmark give us the opportunity to use DIRECT I/O also. How EXT4 and XFS fare in sequential, direct I/O test?
EXT4 in synchronous, direct mode seems penalized in the write test, while in the others we see a substantial parity.
And in random, directed I/O pattern?
Again, we see very strange XFS results...
All considered, while sequential I/O scores are interesting and in line to what I expected, I tend to not consider Sysbench random test as a representative one. So, Sysbench file I/O shows us a virtual tie between EXT4 and XFS.