EXT4 vs XFS: large volumes with high-end RAID controller

Written by Gionatan Danti on . Posted in Linux & Unix

User Rating:  / 7

Bonnie++ results with DIRECT access

While applications normally access the disk by using host OS cache and buffers, sometime can be convenient to directly access the selected file. For example, think to audio/video editing programs: they access streams of data that will be read/written once, so it is not convenient to cache them. More importantly, direct file access is used by many virtual machine softwares (hypervisors) and by some database systems. So, I tested DIRECT file access also.

Let's see Bonnie++ sequential direct access results:


We have a tie here, with XFS using marginally lower CPU time.

Now, random seeks:


EXT4 continue to show a measurable advantage in random seeks.

What about file creation and deletion (metadata handling)?


EXT4 continue to be the fastest. File deletion is again so fast that Bonnie can not produce a valid result.

Now, synchronous operations:


XFS is faster in this mode.

So, from these Bonnie test it appear that XFS can be marginally faster than EXT4 in sequential operations, but it is generally slower in metadata handling. However, you can argue that file creation and deletion speed is rarely a deciding factor, at least when these operations are executed at acceptable levels (as we can see here).

A more interesting thing is that EXT4 appear faster then XFS in random seeks, as random read/write performances are quite important in a wide range of applications.


#1 Evgeny 2012-12-10 07:21
it's looks that you fsck time ext4 isn't true.
I think that you ran fsck.ext4 /dev/sd__something__
without "-f" key. It's means that fsck FS state
tune2fs -l
Filesystem state: clean
and if it clean do nothing

You have no rights to post comments