[Gluster-users] Quick question regarding xfs_repair

[Joe Landman]

On 03/14/2011 01:02 PM, Mohit Anchlia wrote:
> What RAID level is this? Would RAID 10 or RAID 6 help in such scenarios?

Not so much ... if the underlying block device gets corruption, the file 
system atop it might not be able to survive writing to or reading from a 
corrupt region.  RAID6 could work if you have a read after write 
function built into the RAID to guarantee that what gets written is what 
was committed to disk.  Most (actually all) hardware on the market at 
the moment doesn't have this.  Some do talk about capability like this, 
but the real test is to take the write, cache it, compute md5 or similar 
sums, perform the write, then force a non-cached re-read of the data, 
and compute then compare the md5 or similar sums.  If the sums are 
incorrect, then reallocating the write to another location and iterating 
this.

Technically, GlusterFS could do this in a translator.  Realistically, 
this would kill performance.

RAID10 doesn't compute parity, so you need checksums to see if the data 
is correct.  RAID6 does compute parity (2 different mechanisms), but the 
implementations can't really identify a failed write/read, as the parity 
is recomputed at write time, so if corruption occurred in the RAID card 
(say a failing cache chip, or data cable, or ...), its possible bad data 
was committed, and the parity data wouldn't reflect a problem.

Regards,

Joe


-- 
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics Inc.
email: landman at scalableinformatics.com
web  : http://scalableinformatics.com
        http://scalableinformatics.com/sicluster
phone: +1 734 786 8423 x121
fax  : +1 866 888 3112
cell : +1 734 612 4615