Ingo Molnar wrote:
> just to make sure wether this is a RAID5 problem, could you test a RAID0
> setup with the same number and physical layout of disks? That way you can
> simulate similar IO load, without the complexity of RAID5. RAID0 is simple
> and long-tested enough to be trusted 100%.
Certainly.
I made a new /etc/raidtab that built a RAID 0 raid set using the same disks
and partitions. I made it have 9 disks, excluding the 10th disk that I had
removed from my raid5 set prior to testing. Note that I did not mke2fs the
new RAID0 set, as my RAID5 set still has valid data on it. However, I
believe this test is still valid, as it doesn't use the file system. [My
tape subsystem takes ~24 hours to restore ;(]
It doesn't work correctly with RAID0 either. Note that the ide code for the
hpt-366 has specific for "buggy" maxtor drives like mine, so it would seem
that the strongest likelihood is either in the hpt driver or a hardware
compatibility issue, and not with the software raid. I am purely
speculating, but it's possible that the bug only shows up under heavy io
load or something, which might cause it to not work with raid but work on
it's own?
new test RAID0 set:
[root@music /root]# for i in 1 2 3 4 5
> do
> dd if=/dev/md0 count=500000 2> /dev/null | md5sum
> done
e4e661d1bbfa23ddbf3e326b18a3312e -
143f82fcc0634affdf3d05966e785eed -
b31f38842f4b17c2e29a013447eeb582 -
c8a7ca85b71f2fe7c98d9c56042bd2a1 -
8a199e0c95849d69698ea36dc864e61f -
Thanks for your help!
Tom
