On Mar 9, 2011, at 2:14 PM, Chris Murphy wrote:
> 
> OK so are you going to take the definition of "silent data corruption" to 
> mean any sort of error where the user isn't explicitly notified? What if they 
> aren't even passively notified?
> 
> I have had two completely fakaked Time Machine backups that continued to 
> backup, with no complaints, and no complaints on startup or when mounting the 
> disk. Upon restore, errors. Could not complete the restore. Disk Utility? 
> With one of the drives, there were file system errors they were fixed, the 
> restore still would not restore. The other drive, completely clean with Disk 
> Utility, would not restore.
> 
> I would not consider these silent data corruption events, despite not being 
> notified of a problem in advance, if they resulted from file system or data 
> corruption due to a kernel panic or power failure. That's the premise of the 
> article and I don't agree with it.
> 
> In any event, that one is expected to take precautions with RAID 5/6 or even 
> RAID-Z with respect to power management does not mean the incidence of silent 
> data corruption is higher. It simply means *IF* it happens to a RAID 5 array 
> the problems can rapidly become magnified requiring significant 
> contingencies. I might still be able to suck off a bunch of data from a 
> non-array (or RAID 1 disk) despite file system corruption, or the corruption 
> of even 10% of the disk - I will have *better* recovery from RAID 5 if that 
> same event happens to 1 disk, but much worse recovery if that problem is 
> propagated through the entire array. I don't think this is a secret. Again, 
> RAID 5 is not a new thing.
> 
> But I totally disagree with the wording used that implies there is a high 
> incidence of silent data corruption inherent to a RAID 5 system. Usually 
> those systems have better drives, better interfaces, better cables, and more 
> resilient OS, with UPS systems. The net of that is silent data corruption 
> would be less likely by far than a non-arrayed setup of the same capacity.
> 
> Let me use perhaps an imperfect analogy. Cars and airplanes. Cars = non-array 
> and airplanes = RAID 5. The likelihood you're going to have an accident with 
> a car is astronomically higher than an airplane. But if you have an accident 
> in an airplane the incidence of death is higher. Death is contingent on the 
> other happening first, which is very unlikely with airplane travel.
> 


A power failure or sudden crash is not the only place corruption can enter.


But on the Time Machine side, if you want a more reliable time machine backup 
solution, build or buy a box running Nexenta that uses ZFS and then create AFS 
services on it that you can target your Time Machine to.  I am in the process 
of implementing this in my office.

My hosting business actually also uses Nexenta and has two HW RAID Cards (Areca 
with battery back NVRAM).  One HW Raid is a Raid 6 and one a Raid 5 using 
bigger base disks.  Then I use these two volumes in a ZFS mirror, along with a 
mirrored SSD ZIL.  The server sits in a UPS and generator backed data center as 
well.  Provides good performance and I should be able to tolerate a lot of 
disks going out at once.

_______________________________________________
MacOSX-admin mailing list
[email protected]
http://www.omnigroup.com/mailman/listinfo/macosx-admin

Reply via email to