Linux Filesystem Benches inside

[Ulukay]

hello

i've tested some linux filesystems (if someone is interested)

ext2, ext3, xfs, jfs, reiserfs and reiser4

Testsystem: Gentoo 1.4 RC3, gcc 3.2.2-r2, kernel 2.5.60 with reiser4 Patch, Barton@2220mhz, 1024MB Ram and an IBM DJNA 15GB 5400rpm HDD as test drive

i've tested:
copy of 50000 2kb files from a second partition to the test filesystem
copy of 50000 2kb files from the test filesystem to the test filesystem
delete of 50000 2kb files on the test filesystem

copy of 1 500mb file from a second partition to the test filesystem
copy of 1 500mb file from the test filesystem to the test filesystem
delete of 1 500mb file on the test filesystem

[img:a9dc34370d]http://www.ulukay.devisland.net/linux-fs-benches.gif[/img:a9dc34370d]

i hope the image works
if it does not, pls start a new IE and copy this url http://www.ulukay.devisland.net/linux-fs-benches.gif

[Jimbow]

Wow! Fabulous. Thanks.

Reiser4 looks like the wave of the future. The big deletes take a little longer but that graph is an order of magnitude smaller than the others.
_________________
After Perl everything else is just assembly language.

[geoffs]

so anyone want to fill me in with the easiest way to convert my HD to reiser4 I like gentoo...but I sure do waste a lot of time playing around with it.
_________________
CAT

[Malakin]

Nice graphs.

Reiser 4 is pretty new, I'd stick with Reiser 3 for now.

[sindre]

thank you for testing.

IE? no offence, but please don´t use this word in the gentoo forums.

Anyway, did you test ext3 in writeback mode? What about reiserfs with the -notail flag? I heard this should improve performance.

[Ulukay]

i simply benched all FS with the "out of the box" configs
mount /dev/hda4 /test
(always the same partition of course)

is someone want that i bench with special mount options i could rebench it

[Jimbow]

[magnet]

is there a way to convert a reiserfs3 fs to reiserfs4 without loosing data ?
I would like to bench it on my laptop.
nice graph btw.
_________________
every step aim at glory.

[Ulukay]

[magnet]

the braves,do no fear the grave.
_________________
every step aim at glory.

[Ari Rahikkala]

Hmm... I'd expect the time to delete files to be reduced as bugs are ironed out from ReiserFS - the time it takes to delete one large file is so long that it's unlikely to be caused by anything fundamental in the operation of the filesystem. And once whatever causes that is fixed, the performance might increase for deleting smaller files, too.

Not that these benchmarks would be really representative of real-world filesystem performance... I want to see a test where:

1) A partition is filled to 80% with files of varying sizes between zero bytes and three gigabytes (this phase can be timed although it's not likely to happen in real-world cases) in a random directory structure
2) 10% of those files are changed so that their size changes. The changes range from just a couple of bytes added or deleted to multi-gigabyte files being zeroed or small files being grown to several magnitudes of their old size. This phase is benchmarked.
2.1) The filesystem is aged by continuing phase two for a while. Adding and deleting random files may also be done. Assuming it's possible to calculate the contiguousness of a filesystem, you can see if it decreases over time here.
3) cp, mv, del and such benchmarks can be done at this phase as normal. Writing and reading large files at this part would give quite good results about how the filesystem handles fragmentation, I think.
4) The filesystem is filled to as close as 100% as possible with random files, preferably small ones to make it more representative. This phase is, of course, benchmarked. Also, the absolute amount of data is calculated to see how efficiently the filesystem can mash stuff onto a limited amount of space.

It would take a long while to run these tests, unfortunately... especially as you need to run them something like five times or more due to the randomness.

Well, that's what I think of the issue anyway. But then again, I'm no statistician...
_________________
<laurentius> gentoo linux?
<ari> Yesh.
<laurentius> they look horny

[XWRed1]

Copying files between hda3 and hda4?

Might be worthwhile to benchmark again with file operations between two different disks on two different ide channels (scsi too would be cool).

[sindre]

I would like to see how ext3 performs in writeback data mode, since another benchmark (can't find it, slashdot I think) found it to be faster than resierfs.

To boot an ext3 partition in writeback mode you must pass rootflags=data=writeback to the kernel at boot (grub.conf).

[mb4guns]

I see a bright light shining from reiserfs4 .... moment searching my sunglasses
_________________
sys0: Athlon64, Corsair TWINX DDR400 (1024mb), GeForce FX5600 256mb, 2x SATA MAxtor 120gig, Plextor 708a, Zalman cooling, Zalman powered, gnomed
sys1: Dell inspiron 4100 PentiumIII-mobile@1000 512mb ATI Radeon mobility m6 30gig, gnomed

[metacove]

Techinically it's called reiser4 and reiserfs. I tested the bitkeeper source of reiser4 a few days ago and it was quite nice. It still has a little bit to go before I'd run it full time. I am very excited for the stable release.