Here are some timings on my new Linux system copying a few gigabytes of videos from one set of RAID disks to another set of RAID disks with three different filesystems: ext3, reiser3, and xfs:
ext3: root@mu:/data# time tar cf - videos | ( cd /data2 ; tar xf - ) real 20m37.590s user 0m5.898s sys 3m46.579s reiserfs: root@mu:/data# !time time tar cf - videos | ( cd /data2 ; tar xf - ) real 19m33.362s user 0m6.620s sys 3m35.222s xfs: root@mu:/data# !time time tar cf - videos | ( cd /data2 ; tar xf - ) real 17m24.317s user 0m6.296s sys 2m41.250s root@mu:/data#Holy cacophony, Batman! Look at how much faster XFS is in this test!
The question, then, is why is ext3 the "standard" filesystem for Linux, when it is the slowest by any measure of the word "slow"? (It's even worse for lots of small files as vs. lots of big files). I suppose it's because it's much simpler than ReiserFS and XFS. XFS also has a bad habit of filling out your files with lots of zeros if you lose power before it has managed to flush its buffers, because it journals its metadata but not its data writes (gah! I don't want my metadata committed until the data is on the disk!), but ext3 takes hours to fsck under that situation and generally ends up with lots of files in lost+found, so I'm not sure it's much better.
Still, it makes you wonder whether NIH (Not Invented Here) for SGI's XFS, and Hans Reiser's acerbic personality (I mean, the dude is on trial for *MURDER* because he's such a jerk that everybody believes he killed his wife, even though no body has ever been found and nobody has any idea how he did it) have as much to do with as technical merit. So much for that corny idea that Open Source would rule the world because decisions would be made due to technical merit, not infighting and politics and marketing reasons... sigh. Utopia still is just a fiction...
-- Badtux the Linux Penguin