RAID 1 & 5 and corrupted files: resilience?

[Elledan]

If due to some cause the data on one of the drives in a RAID 1 or 5 array became corrupted (pesky cosmic radiation =) ), what effect would this have on the following read action involving this data?

With RAID 1 you've suddenly got two copies of a file which are just slightly different. With RAID 5 you've got either a corrupted file or corrupted parity data. This would suggest that RAID 5 has at least some chance of repaing this damage, while with RAID 1 it seems only one of the two files can be selected as the 'real' file (50/50 chance).

What do you think, Ice Czar?

 

[Valdain]

That's a good question! I guess the answer would have to lie in how the corruption occured. If it was due to bad data sent to the drive to be written (bad memory maybe?), you're out of luck becuase the array controller (hardware or software) will just see it as a normal write and sync up the array. If the contents of the disk just "changed" (pesky cosmic radiation) then it would greatly depend on how or when your controller does a consistancy check. Most of my experiences are with Compaq array controllers, so I'll go off of that. When the system is restarted, the controller will run a check verifiying the integrity of the array. The mystery corruption would get repaired there, hopefully. For a running system, I think the controller will do a spot check from time to time. My guess is along the same lines as yours. RAID 1, 50/50. RAID 5, it should be checking the parity and see that things aren't quite right on one of the disks. I think that a RAID 0+1 setup might also have some checks, but that could be a Compaq specific implimentation.

Just my 2 cents. Curious as to what others think.

 

[Ice Czar]

cosmic radiation borks RAM not bits on the platter
and you running ECC anyway?

If the memory is borked, its toast for it all, mirror or parity info
if on the other hand the HDD gets a bad sector, it should map it and realocate it if possible
same with filesystem recovery, rebuilding where possible

so a mirror need to be compared, and a RAID 5 synchd
Im goin to sleep so this is off the top of my head aand its very sleepy so....

but read this regarding RAID 5
Write Hole

and about soft bit error as well
Error Correcting Memory - Part I
Error Correcting Memory - Part II: Myths and Realities

I'll dig up info on the HDD error mapping, and filesystem tommorow after I get my caffine IV