>Paul Breen wrote:
>Both boxes are raided (Raid
>5), don't know if this is significant.
Not sure about the data loss problem, but from a performance perspective, RAID
5 doesn't sound to me like the way to go for a high performance db file
system. You are dividing up writes (and reads) between disks plus calculating
checksums during writes. I think that a better idea would be to use RAID
1+0. Take a few drives and stripe then in a RAID 0 config to get the storage
space you need, then duplicate the set and mirror the RAID 0 (hence RAID
1+0.) Your reads will go a lot faster that way, and although you have to
double up on writes, at least the blocks are contiguous on the drives, and
your controller doesn't have to calculate checksums.
Of course, if someone sees a reason that RAID 5 would be better than RAID 1+0,
I'd appreciate an explanation!
-Serge