EXT4 vs. XFS vs. Btrfs HDD Benchmarks On Linux 3.15

[phoronix]

Phoronix: EXT4 vs. XFS vs. Btrfs HDD Benchmarks On Linux 3.15

After earlier in the week delivering solid-state drive file-system benchmarks in comparing the Linux 3.15 FS performance to Linux 3.14 stable, now it's time to do a Linux 3.14 vs. 3.14 file-system performance comparison with a traditional hard drive. The file-systems being benchmarked here are EXT4, XFS, and Btrfs.

http://www.phoronix.com/vr.php?view=20368

[PreferLinux]

Running AIO-Stress on the Western Digital VelociRaptor was yielding speeds not technically possible for the Serial ATA 3.0 hard drive when using EXT4 and XFS. There was obviously some caching / lack of committing to disk involved that led to faster performance while the Btrfs file-system was in line with real-world expectations out of the random write disk test.

Um, no way. It is still over its real performance by probably at least 10 times, if not 500 times. That number (270 MB/s) would be about in line for an SSD, but is far too high for a mechanical HDD like this ? even though it is a 10K RPM drive. Check out the tests on page three ? sequential write (the first test) is about 135 MB/s, this is about twice that. HDDs have random speeds vastly lower than their sequential speeds. The random speeds shown on page three (second test) are only 0.6 MB/s.

[Imroy]

Why test with such a fast HDD? A slower HDD (even a regular 7200RPM one) would have revealed more about how efficiently each filesystem arranges data on disk.

I'll be getting a pair of WD Red's soon (5x00RPM) and would like to know how BTRFS compares to XFS in performance, especially in a RAID-1 setup. I'm still not sure about BTRFS's stability/reliability, but an overwhelming speed advantage might sway me.

[FRLH]

NTFS benchmark

Maybe a bit off-topic, but a comparison with NTFS on the same hardware would have been nice.
The kernel is competing against itself and against windows too.

[FRLH]

Since you often compare the linux graphics perfs with windows you could do the same for the kernel and the filesystems. Perhaps a new article?

[edoantonioco]

Maybe a bit off-topic, but a comparison with NTFS on the same hardware would have been nice.
The kernel is competing against itself and against windows too.

I would like to see that too, that would be interesting.

[oleid]

Since you often compare the linux graphics perfs with windows you could do the same for the kernel and the filesystems. Perhaps a new article?

A bit older, but maybe interesting for you:

http://www.phoronix.com/scan.php?page=article&item=ubuntu_win7_ws&num=4

[FRLH]

A bit older, but maybe interesting for you:
http://www.phoronix.com/scan.php?pag..._win7_ws&num=4

Interesting, but old and limited :
* Both systems (Ubuntu 10.04LTS and Windows 7) are now 4 years old
* kernel was 2.6.32
* only EXT4 vs NTFS
* only 2 filesystem tests, and with IOZone
* not a SSD

A newer and broader comparison would be realy interesting.

[rudregues]

Interesting, but old and limited :
* Both systems (Ubuntu 10.04LTS and Windows 7) are now 4 years old
* kernel was 2.6.32
* only EXT4 vs NTFS
* only 2 filesystem tests, and with IOZone
* not a SSD

A newer and broader comparison would be realy interesting.

100% agreed!
EXT4 vs. XFS vs. Btrfs vs. NTFS both in HDD and SDD disks would be great

[gadnet]

ext4 / btrfs ok but why not ext4+ soft raid1 versus Btrfs with raid option ?

Hi,

i think less and less people use use straight disk now. For security we use raid1 mostly so i think a test from EXT4+mdadm raid1 and Btrfs in raid1 like mode should be interesting no ?

regards,
Ghislain.