Hello,
I did a very simple test sometimes ago on deleting files and found that it
is much slower on SCSI than on IDE drive !
Box 1: Bi-Pentium III 1Ghz, 256 MB Ram, hard drive SCSI FUJITSU UW2, 18 GB,
Adaptec AIC 7899, hdparm -t /dev/sda: 52.89 MB/sec
Test partition is /dev/sda3, 5.9 GB.
Software: RedHat 7.2, with vanilla kernel 2.4.16 + patches xfs-2.4.16-all-i386
and jfs-1.0.14, compiled with egcs 1.1.2. Test partition is /dev/sda3, 5.9 GB.
Box 2: Duron 700 Mhz, 128 Mo, hard drive IDE Western Digital Udma66, 20 GB
hdparm -t 20.98 MB/sec. Test partition is /dev/hda10, 3.6 GB.
Software: Slackware 8.0, vanilla kernel 2.4.16 + patches xfs-2.4.16-all-i386
and jfs-1.0.14, compiled with gcc-2.95.3.
The test consist in formatting the test partition, copying the source of the
kernel into the test partition, umounting the partition, and then
deleting the copied source tree.
Results of time rm -rf /testfs/LINUX; time sync
Box 1:
---------------------------------------------------------------------------------
ext2 ext3 reiserfs xfs jfs
---------------------------------------------------------------------------------
real 6.972s 7.245s 3.086s 1m17.333s 48.510s
user 0.020s 0.010s 0.050s 0.070s 0.030s
sys 0.180s 0.540s 1.210s 2.640s 0.800s
sync real 0.616s 0.569s 0.165s 0.014s 0.039s
---------------------------------------------------------------------------------
Box 2:
---------------------------------------------------------------------------------
ext2 ext3 reiserfs xfs jfs
---------------------------------------------------------------------------------
real 4.907s 7.416 4.242s 15.415s 15.005s
user 0.050s 0.020 0.060s 0.120s 0.110s
sys 0.230s 0.820 1.480s 3.610s 0.960s
----------------------------------------------------------------------------------
The test had been running several times with comparable output.
I known xfs is rather slow when deleting large number of files (above 11,000
here). But it is rather surprising that it performs *much worse* on a much
better machine. The problem is same with a mono processor kernel on Box 1.
Is there a known issue here ?
Liu
|