On 28/06/2012 13:46, Wojciech Puchar wrote:
(unless we are talking temporary removal and re-insertion?)
nope, I'm talking about complete pair's crash when two disks die.
I do understand that's the possibility of such outcome (when two
disks in the same pair crash) is not high, but
when we have 12 or 24 disks in storage...
then may 6-12 filesystems. overall probability of double disk failure
is same, but you will loose 1/6-1/12 of data.
But the compromise is that you gain the complexity of maintaining more
filesystems and needing to figure out how to split your data across
multiple filesystems
The options today however seem to be only:
- RAID6 (suffers slow write speeds, especially for smaller files)
- RAID1 pairs with striping (raid0) over the top. (doesn't achieve max
speeds for small files. 2 disk failures a problem. No protection against
"silent corruption" of 1 disk)
- RAID1 pairs, plus some kind of intelligent overlay filesystem, eg
md-linear+XFS / BTRFS. With the filesystem aware of the underlying
arrangement it can theoretically optimise file placement and
dramatically increase write speeds for small files in the same manner
that RAID-0 theoretically achieves. (However, still no protection
against "silent" single drive corruption unless btrfs perhaps adds this
in the future?)
So given the statistics show us that 2 disk failures are much more
common than we expect, and that "silent corruption" is likely occurring
within (larger) real world file stores, there really aren't many battle
tested options that can protect against this - really only RAID6 right
now and that has significant limitations...
RAID1+XFS sounds very interesting. Curious to hear some failure testing
on this now. Also I'm watching btrfs with a 12 month+ view
Cheers
Ed W
