[Btrfs-users] btrfs v0.11 & btrfs v0.12 benchmark results

[Gabor MICSKO]

On Fri, 2008-02-15 at 08:49 -0500, Chris Mason wrote:
> On Friday 15 February 2008, Gabor MICSKO wrote:
> > On Thu, 2008-02-14 at 19:54 -0500, Chris Mason wrote:
> > > On Thursday 14 February 2008, Gabor MICSKO wrote:
> > > > Hi,
> > > >
> > > > I've recently benchmarked btrfs v0.11 & v0.12 against ext2, ext3, ext4,
> > > > jfs, reiserfs and xfs.
> > >
> > > Interesting, at least here I see pretty much the same speeds for all the
> > > filesystems on sequential IO.  But, I'll test again with more spindles.
> > >
> > > Could I trouble you to run one more btrfs run?  Please mount v0.12 with
> > > -o nodatasum
> >
> > Unfortunately these test boxes are not available anymore for me to
> > conduct further tests, but i plan to prepare even more btrfs related
> > tests in the future.
> 
> Ok, I'll try to reproduce your results on some machines here.  I see both 
> machines had more than one spindle, how were the drives configured? (striped, 
> mirrored, raid?)
> 
> -chris

I've created two "simple as possible" test environments: 

Fu-Si Primergy RX330 with 3 SAS drives:
---------------------------------------

- drive0 \
            HW RAID1 - /dev/sda1  - root fs, /dev/sda2 - swap
- drive1 /

- drive2 - /dev/sdb1

During tests i used /dev/sdb1 (~ 70GB) for btrfs, ext2, etc.
filesystems. 

For example: 

mkfs.btrfs /dev/sdb1
mount /dev/sdb1 /data


Fu-Si Econel 200 with 2 SATA drives:
------------------------------------

- drive0 - /dev/sda1 - root fs, /dev/sda2 - swap

- drive1 - /dev/sdb1 

During tests i used /dev/sdb1 (~ 160GB) for btrfs, ext2, etc.
filesystems. 

For example: 

mkfs.btrfs /dev/sdb1
mount /dev/sdb1 /data

Principles followed under the tests:

- single user mode
- echo "3" > /proc/sys/vm/drop_caches
- cron, etc. disabled
- before each tests i've created filesystems from scratch using default
options:

e.g.: 

umount /data
mkfs.btrfs /dev/sdb1
mount /dev/sdb1 /data
sync
sleep 10 
tiobench --dir /data --size 2048 --threads 2 --block 4096

- all filesystems were mounted using default mount options except in
"ext4 wb" tests where i used:

mount /dev/sdb1 /data -t ext4dev -o extents -o data=writeback

- all tests were run 3 times and the average was taken

Tests:

1.) tiobench --dir /data --size 2048 --threads 2 --block 4096
2.) tiobench --dir /data --size 2048 --threads 2 --block 16384
3.) bonnie++ -d /data -u root -s 2048 -m {btrfs, ext2, ex3, ...}
4.) dbench -D /data -t 60 50

Regards,

-- 
Gabor MICSKO - http://hup.hu