Sun, 11 Oct 2009
- Kernel: 126.96.36.199-56.fc12.i686.PAE
- I/O Scheduler: CFQ
- Encryption: LUKS/dm-crypt AES-XTS cipher 512 bit key
- Installation Media: rawhide boot.iso (20091010)
- Benchmark: bonnie++
- Graphing script: flotbonnie.py
- Hardware profile: 188.8.131.52-56.fc12.i686.PAE-hardinfo (note: ancient hardware)
- Raw data: 184.108.40.206-56.fc12.i686.PAE-btrfs, 220.127.116.11-56.fc12.i686.PAE-ext3, 18.104.22.168-56.fc12.i686.PAE-ext4, 22.214.171.124-56.fc12.i686.PAE-xfs
Posted by Luke at Mon Oct 12 02:31:36 2009
I should also note that these tests just run bonnie++ without any special arguments. I will most likely re-run these with -b to enable fsync() calls after writes, which will do a better job of simulating real-world loads.
Posted by Bernie at Mon Oct 12 06:40:45 2009
I'd be interested to see reiserfs (-o notail) included purely because I have an old squid box now straining under I/O load. Back in the day, reiserfs was the received wisdom for squid. No idea what my new squid box should run, maybe your graphs will give me a hint.
Posted by Luke at Mon Oct 12 16:35:16 2009
@Bernie, Yep I will definitely be bringing Reiserfs into the mix during my next round of testing.
However, I'm going to wait a few days for 2.6.23-rc3+ to hit rawhide, since it apparently contains some optimizations to the I/O layer.
Posted by Luke at Tue Oct 13 01:59:45 2009
I should also note that this was done on some OLD hardware. It's not meant to show how fast my machine is, but how the different filesystems compare to eachother.