Make your RAID5 chunk size larger than 4kB. I'd suggest you start with
32kB chunks; I chose 256kB for my RAID5 over four Serial ATA disks.
# time tar jxf linux-2.4.24.tar.bz2
real 0m35.627s
user 0m27.480s
sys 0m5.170s
# time rm -rf linux-2.4.24
real 0m5.761s
user 0m0.000s
sys 0m2.440s
# xfs_info /home
meta-data=/home isize=256 agcount=25, agsize=1048576 blks
= sectsz=4096
data = bsize=4096 blocks=26214400, imaxpct=25
= sunit=64 swidth=192 blks, unwritten=1
naming =version 2 bsize=4096
log =internal bsize=4096 blocks=12800, version=2
= sectsz=4096 sunit=1 blks
realtime =none extsz=65536 blocks=0, rtextents=0
--
Matt Stegman
On 2 Apr 2004, Jens Schmalzing wrote:
> Hi,
>
> I'm currently testing a small fileserver, using an array of SCA disks
> with xfs on top of software raid5. The chunk size of the raid is set
> to 4k, as is the sector size for the xfs filesystem. To be more
> precise, the file system was created with
>
> | # mkfs -t xfs -f -s size=4096 -l size=16384b /dev/md0
> | meta-data=/dev/md0 isize=256 agcount=32, agsize=5608052
> blks
> | = sectsz=4096
> | data = bsize=4096 blocks=179457664, imaxpct=25
> | = sunit=1 swidth=6 blks, unwritten=1
> | naming =version 2 bsize=4096
> | log =internal log bsize=4096 blocks=16384, version=2
> | = sectsz=4096 sunit=1 blks
> | realtime =none extsz=24576 blocks=0, rtextents=0
>
> While the setup performs well in general, I am a bit disappointed at
> the file creation speed - for example, unpacking a Linux kernel tree
> (a bit more than 200MB in a bit less than 20k files and directories)
> takes more than a minute and a half. The same is true for removing
> the whole thing again. On the other hand, reading is really fast.
>
> Any suggestions on tuning the file system or the software raid to
> obtain better performance? I'm also open to suggestions involving a
> different file system altogether.
>
> Regards, Jens.
>
>
|