Yusuf Goolamabbas wrote:
>Here's result of running akpm's synctest available at
>I tried to run synctest in a way which modelled Postfix's
><http://www.postfix.org/> queue writing behaviour.
>I ran it against ext3 (htree,ordered) and xfs. I used xfsprogs-2.6.9 to
>mkfs the xfs filesystem. I found that with the parameters I ran, xfs was
>very slow compared to ext3
>Hardware, P3-450 with 384MB RAM, filesystems on a 9GB SCSI disk
>Anticipatory IO scheduler with global_tag_depth of 4
>Tests performed after machine was cleanly booted.
>Multiple tests done, box rebooted between tests.
>numbers are more or less the same
>kernel version: 2.6.5-rc3-mm4
>/usr/bin/time -p ./synctest -fu -t 100 -p1 -n1 /path/to/appropiate-dir
>htree,ordered : real 60.16 user 1.64 sys 17.81
>xfs : real 880.71 user 2.93 sys 48.09
Did you use default mkfs parameters on xfs? Try upping the log size:
mkfs -t xfs -f -l size=32768b .....
mount -o logbufs=8 ....
The multiple test dirs will make xfs spread out all over the disk,
not sure how ext3 will place them. Probably seek heaven, what
does the disk sound like on ext3, it chatters a lot for xfs.