Quote:
Originally Posted by hussar
But I was wondering, can you use a journaling filesystem on a RAID array? Does it make any sense to?
|
Absolutely yes! You really should use a journaling fs. fsck'ing after a crash will take a very long time on a large array. Since you've already settled on ext2 going with ext3 is probably easiest since you can do that without recreating the filesystem. Just unmount it, tune2fs -j /dev/md0, change fstab entry to ext3. Remount. Done.
I used xfs for my raid (4x320GB raid-5) because it mainly holds large files.
ext3 filesystem:
dirac:/raid # dd if=/dev/zero of=./test bs=1024k count=2048
86154529 bytes/s
dirac:/raid # dd if=./test of=/dev/null bs=1024k count=2048
174263984 bytes/s
write performance: 82.2MB/s
read performance: 166.2MB/s
xfs filesystem:
dirac:/raid # dd if=/dev/zero of=./test bs=1024k count=2048
185052817 bytes/s
dirac:/raid # dd if=./test of=/dev/null bs=1024k count=2048
180374561 bytes/s
write performance: 176.5MB/s
read performance: 172.0MB/s
These tests were done on a completely empty filesystem, and have little to do with real world throughput. However, the clients connect to this server via gigabit ethernet and the bottleneck is the network and not the disk i/o.