Hi,

since I had some chores to do and didn't want that my box got bored - also my
new replacement disk needed some stress testing, I did some simple and
probably very flawed benchmarking for the following fs:
ext3
jfs
xfs
reiserfs
reiser4.

What I did was very amateurish, but I like to share my results nonetheless.
You are free to critize. In fact, I would like to hear hints or tips for the
next time. What I don't want to read is 'you suck go away'. However 'You suck,
because,...' is acceptable.

This 'benchmark' consists of: creating two partitions with the fs to be
benchmarked - sdb1 and sdb4. sdb2 is /opt and sdb3 a backup partiton. Both
mounted, but with no access while I did the runs. I kept them mounted to
prevent accidental deletion by stupid fingers. sdb1 is ~45gb and sdb4 ~94gb
'huge'.

sda&sdb are Samsung HD502IJ 500gb, with NCQ turned on.
CPU: AMD X2 6000+ with 'performance' as scaling governor, 4GB of ram.
Nforce 520/MCP65 chipset. NCQ was on.

All filesystems were compiled into the kernel.

first step was mkfs, mounting sdb1 to 'source' and sdb4 to 'target'. After
that I copied a 'benchmark' directory consisting of ~9000pictures in one
directory with some,sub-directories, a maildir with ~160000 mails in several
folders, and a 'films' directory with some sub dirs and some more pictures in
sub-directories , all in all ~22GB of data, into the 'source' partition, in
the following called 'prepare'. This was not timed. The dataset consist of
what I would call 'my typical home' - I did not include my documents dir
because it is pretty much dwarfed by the other three (and I simply forgot.
Ahem).

From source I copied to target, followed by sync, this is called 'create' and
was timed with time.. I choose the same fs for both partitions for fairness
reasons. After that, umount, mount, echo 1 > /proc/sys/vm/drop_caches

Then I copied the benchmark-dir on the target. This is called 'copy' below.
'benchie' was copied to 'bencho' (I am great with names). cp, follwed by sync,
all of it timed, then umount, mount, echo 1 > ....

After that came 'move'. Moving the benchmark dir benchie into another
directory - called benchie2 - not renaming. Of course followed by sync and
timed.
umount/mount/drop_caches

rem1:time rm'ing the benchmark dir 'bencho' on target&sync, then umount,
mount, ...
rem2:time rm'ing benchie2 on targe&sync,umount, mount...

The results are by no means 'fair'.As you can see with the mount options. I
would have liked to turn on barriers for all fs. Same for reiser4 which also
used lzo compression (because that is what I am using and I was interested how
fast/slow it is compared to the rest).

One thing: mount option documentation sucks. man mount has some options,
/usr/src/linux/Documentation/filesystems has some options, and then there is
the stuff you have to google/grep sources for (reiserfs, I am looking at you).
This is something that IMHO needs improving. Even if most users use pre-tuned
distris.

I did no 'tuning'. If defaults suck, they suck. Sorry. But nobody can expect
08/15 home users to google for hours for the perfect set of settings.

X was not running
cron was not running
smartd, hddtemp, dbus, hald, udev, metalog WERE running. I tried to be close
to my 'everyday' setup - X was not running to prevent me from skewing the
results by clogging the cpu.

Results:

ext3:
mkfs.ext3 /dev/sdb1 0,02s user 0,99s system 6% cpu 14,816 total
mkfs.ext3 /dev/sdb4 0,03s user 3,63s system 9% cpu 39,030 total
mount /dev/sdb1 -t ext3 -o barrier=1,data=journal,noatime /mnt/source 0,00s
user 0,00s system 0% cpu 0,025 total
mount /dev/sdb4 -t ext3 -o barrier=1,data=journal,noatime /mnt/target 0,00s
user 0,00s system 11% cpu 0,030 total
#
disk usage after prepare:
/dev/sdb1 48062440 22160100 23460864 49% /mnt/source
#
create.sh
sh create.sh 1,08s user 105,19s system 3% cpu 45:17,78 total
/dev/sdb4 96221328 22167824 69165728 25% /mnt/target
#
copy.sh
sh copy.sh 1,09s user 114,32s system 3% cpu 50:15,30 total
/dev/sdb4 96221328 44147224 47186328 49% /mnt/target
#
move.sh
sh move.sh 0,00s user 0,01s system 4% cpu 0,218 total
#
rem1.sh:
rm -rf /mnt/target/bencho 0,04s user 5,12s system 6% cpu 1:17,75 total
rem2.sh:
sh rem2.sh 0,04s user 5,74s system 4% cpu 2:05,16 total
################################################## ######

JFS
mkfs.jfs -q /dev/sdb1 0,00s user 0,04s system 11% cpu 0,337 total
mkfs.jfs -q /dev/sdb4 0,00s user 0,09s system 12% cpu 0,731 total
#

mount /dev/sdb1 -t jfs -o noatime /mnt/source 0,00s user 0,00s system 121%
cpu 0,003 total
mount /dev/sdb4 -t jfs -o noatime /mnt/target 0,00s user 0,00s system 110%
cpu 0,003 total
#
disk usage after prepare.sh:
/dev/sdb1 48795072 22049220 26745852 46% /mnt/source
#
create.sh:
sh create.sh 0,79s user 64,93s system 6% cpu 16:14,70 total
/dev/sdb1 48795072 22049220 26745852 46% /mnt/source
/dev/sdb4 97719568 22046568 75673000 23% /mnt/target
#
copy.sh:
sh copy.sh 0,84s user 68,99s system 6% cpu 17:22,76 total
/dev/sdb4 97719568 44080448 53639120 46% /mnt/target
#
move.sh:
sh move.sh 0,00s user 0,00s system 0% cpu 0,353 total
#
rem1.sh:
sh rem1.sh 0,05s user 4,74s system 3% cpu 2:26,86 total
#
rem2.sh:
sh rem2.sh 0,05s user 4,55s system 3% cpu 2:27,94 total
################################################## ####

XFS
mkfs.xfs -f /dev/sdb1 0,00s user 0,01s system 2% cpu 0,322 total
mkfs.xfs -f /dev/sdb4 0,00s user 0,01s system 0% cpu 0,907 total
#
mount /dev/sdb1 -t xfs -o barrier=1,noatime /mnt/source 0,00s user 0,00s
system 2% cpu 0,161 total
mount /dev/sdb4 -t xfs -o barrier=1,noatime /mnt/target 0,00s user 0,00s
system 1% cpu 0,168 total
/dev/sdb1 47G 4,2M 47G 1% /mnt/source
/dev/sdb4 94G 4,2M 94G 1% /mnt/target
#
disk usage after prepare.sh:
/dev/sdb1 48805696 21979288 26826408 46% /mnt/source
#
create.sh:
(mails extre slow, films fast, first 6gb were responsible for most of the
hour)
sh create.sh 1,09s user 82,65s system 2% cpu 1:04:30,78 total
/dev/sdb1 48805696 21979288 26826408 46% /mnt/source
/dev/sdb4 97707792 21979280 75728512 23% /mnt/target
#
copy.sh:
sh copy.sh 1,10s user 90,33s system 3% cpu 40:03,23 total
/dev/sdb4 97707792 43954232 53753560 45% /mnt/target
#
move.sh:
sh move.sh 0,00s user 0,01s system 2% cpu 0,540 total
#
rem1.sh:
sh rem1.sh 0,04s user 10,76s system 1% cpu 12:13,20 total
#
rem2.sh
sh rem2.sh 0,04s user 11,23s system 1% cpu 13:24,95 total
################################################## ####

reiserfs:
mkfs.reiserfs -q /dev/sdb1 0,01s user 0,04s system 0% cpu 10,455 total
mkfs.reiserfs -q /dev/sdb4 0,01s user 0,07s system 2% cpu 3,502 total
#
mount:
mount /dev/sdb1 -t reiserfs -o barrier=flush,data=journal,noatime /mnt/source
0,00s user 0,06s system 14% cpu 0,392 total
mount /dev/sdb4 -t reiserfs -o barrier=flush,data=journal,noatime /mnt/target
0,00s user 0,04s system 6% cpu 0,555 total
#
disk usage after prepare:
/dev/sdb1 48828008 21960572 26867436 45% /mnt/source
#
create:
sh create.sh 1,12s user 183,73s system 7% cpu 41:17,34 total
#
copy:
sh copy.sh 1,03s user 173,93s system 11% cpu 25:56,22 total
/dev/sdb4 97752532 43882824 53869708 45% /mnt/target
#
move:
sh move.sh 0,00s user 0,01s system 0% cpu 0,688 total
#
rem1.sh:
sh rem1.sh 0,04s user 16,59s system 30% cpu 53,700 total
#
rem2.sh
sh rem2.sh 0,06s user 16,57s system 29% cpu 56,769 total
################################################## #########

reiser4+lzo
mkfs.reiser4 -y -o create=ccreg40,compress=lzo1 /dev/sdb1 0,00s user 0,01s
system 2% cpu 0,423 total
mkfs.reiser4 -y -o create=ccreg40,compress=lzo1 /dev/sdb4 0,01s user 0,02s
system 1% cpu 2,619 total

mount /dev/sdb1 -t reiser4 -o noatime /mnt/source 0,00s user 0,01s system 0%
cpu 2,893 total
mount /dev/sdb4 -t reiser4 -o noatime /mnt/target 0,00s user 0,02s system 0%
cpu 4,717 total
#
disk usage after prepare.sh:
/dev/sdb1 46397568 21190652 25206916 46% /mnt/source
#
create.sh:
sh create.sh 0,88s user 123,96s system 15% cpu 13:04,77 total
/dev/sdb1 46397568 21190652 25206916 46% /mnt/source
/dev/sdb4 92886840 21192180 71694660 23% /mnt/target
#
copy.sh:
sh copy.sh 0,89s user 142,65s system 17% cpu 14:01,58 total
/dev/sdb4 92886840 42381216 50505624 46% /mnt/target
#
move.sh:
sh move.sh 0,00s user 0,01s system 1% cpu 0,602 total
#
rem1:
sh rem1.sh 0,07s user 23,64s system 22% cpu 1:47,22 total
rem2:
sh rem2.sh 0,06s user 20,18s system 19% cpu 1:41,35 total
################################################## ###########

the 'sh' are extremly simple. Just the operation (copy, move, rm) and sync.

I plan to do reiser4 with gzip and without compression in the next couple of
days. If you want me to try one of the other fs with different options, just
say so.

I was very surprised by jfs and xfs. The first was faster than expected (even
with the unfairness in its favour) and xfs was much, much slower than
expected. XFS was pretty fast with the films, but suffered a lot with the
emails, while reiserfs and reiser4 dealt very well with the emails.

Glück Auf,
Volker
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/