I'm new to using xfs, so please forgive me if I'm missing somehting
obvious. I've heard good things about xfs' performance, and I have to
admit I'm quite impressed with what I've seen so far. However, I've run
into some confusing problems.
With all of the XFS kernels I've build (2.4.24-pre1, 2.4.25-pre4) I've had
trouble creating large files.
dd if=/dev/zero of=/mnt/blah bs=1024k count=10000
locks up at app 6gigs with bdflush and kswapd consumming all available CPU.
Once this occurs any thing attempting to access the filesystem gets stuck
waiting on the kernel.
Creating many 1 gig files at once doesn't exhibit the trouble.
while this in 1 2 4 5 6 ... 20 do; dd if=/dev/zero of=/mnt/blah/file.${this} &
done
The volume is not going to be used for such largefiles but it makes me
considerably uneasy to have such a clear problem.
As a side note, I'm using a new Promise SATA->SCSI raid box (not official
released yet) with dual U160 interfaces. With the naive tuing that I've
done so far I'm able to get ~120MB through a single channel. It's a 820gb 14
disk raid 10 set, with a 64k stripe.
What is a good recomened journal size? There's some suggestion that a
bigger one could be better, but how big? Maximum size? Likewise,
increasing the logbufs and logbufsize. What settings people having success at?
This box going to be a NFS server for small file IO and lots of dbfiles at
about 5-10 mb a piece. I expect a pretty high transactional IO load.
--
Kelsey Cummings - kgc@xxxxxxxxx sonic.net, inc.
System Administrator 2260 Apollo Way
707.522.1000 (Voice) Santa Rosa, CA 95407
707.547.2199 (Fax) http://www.sonic.net/
Fingerprint = D5F9 667F 5D32 7347 0B79 8DB7 2B42 86B6 4E2C 3896
|