Re:

Hi Dragon,

On 06/18/2011 04:39 PM, Dragon wrote:

Monitor your background reshape with "cat /proc/mdstat".

When the reshape is complete, the extra disk will be marked "spare".

Then you can use "mdadm --remove".
-->after a view days the reshape was done and i take the disk out of the raid -> many thx for that

Good to hear.

quoted

at this point i think i take the disk out of the raid, because i need the space of

the disk.

Understood, but you are living on the edge.  You have no backup, and only one drive
of redundancy.  If one of your drives does fail, the odds of losing the whole array
while replacing it is significant.  Your Samsung drives claim a non-recoverable read
error rate of 1 per 1x10^15 bits.  Your eleven data disks contain 1.32x10^14 bits,
all of which must be read during rebuild.  That means a _13%_ chance of total
failure while replacing a failed drive.

I hope your 16T of data is not terribly important to you, or is otherwise replaceable.
--> nice calculation, where do you have the data from?
--> most of it is important, i will look for a better solution

The error rate is from Samsung, for your HD154UI drives:
http://www.samsung.com/latin_en/consumer/monitor-peripherals-printer/hard-disk-drives/internal/HD154UI/CKW/index.idx?pagetype=prd_detail&tab=specification

error rate = 1 / 1*10^15 = 1x10^-15

The rest comes from your setup:
11 disks * (1465138496 * 1024) bytes/disk * 8 bits/bytes = 1.32026560152e+14

% odds of failure = (data quantity * error rate) * 100%

[...]

--> and than, ext4 max size is actually 16TB, what should i do?

I've been playing with XFS.  The only significant maintenance drawback I've identified is that it cannot be shrunk.  Not even offline.  It's not really holding me back, though, as I tend to layer LVM on top of my raid arrays, then allocate to specific volumes.  I always hold back a substantial fraction of the space for future use of "lvextend".

--> for an end-user you have many knowledge about swraid ;)

Thank you.  I was a geek before I became an engineer :) .

Phil