[LAU] RAID and AVLinux

[Jörn Nettingsmeier]

Arnold Krille wrote:
> Unless I miss something on AVLinux:
> 
> On Saturday 03 October 2009 21:36:54 Jörn Nettingsmeier wrote:
>> Jonathan E. Brickman wrote:
>>>>  You should at least use RAID6.
>>> 1.  How does RAID6 differ from RAID5?
>> raid 6 uses a more elaborate algorithm than the simple parity of raid5,
>> to calculate two redundant bits. so you can recover from the failure of
>> two disks out of a raid6 array.
> 
> Loosing two disks means you got more then two coupled. If so, you have 
> definitely several of the same charge in that array. (Don't deny it, the human 
> laziness speaks against you.)
> Recovering will still do heavy load on the disks. Now calculate the 
> probability of two disks of the same charge failing without a third disk of 
> that charge failing during recovery. Pretty low, huh?

that remark is bogus. each disk design has a mean-time between failures
of X hours. that means of N disks, N/2 will have failed after time X for
a sufficiently high number of samples N.

the failure of each individual disk is a statistically independent
event. there is absolutely no synchronized self-destruct of disks
belonging to the same batch.
there is however the tendency of new disks to either die pretty fast or
last quite a while, and a very instructive long-term study by our
friends at google about aging disks. so if you're really careful with
your data, you don't wait for an old disk to fail but retire it after a
defined time. plus you should watch your smart data for early warning
signs. doing this makes large arrays quite manageable.

the problem these days is that disk sizes are getting ever larger.
that means more data is at risk, and the time to resync a degraded array
is also increasing. hence, raid5 might not be safe anymore. raid6 is
designed to not expose your data to a window of vulnerability at all -
one disk can always be swapped while the array remains fully redundant.