Re: very large data-offset?

[NeilBrown <neilb@suse.com>]

[-- Attachment #1: Type: text/plain, Size: 1800 bytes --]

On Mon, Dec 04 2017, Duane wrote:

> On 2017-12-03 05:51 PM, NeilBrown wrote:
>> On Thu, Nov 30 2017, Duane wrote:
>>
>>> Why is the data offset set so big? I created a 3x4TB RAID5 array and the
>>> data offset was 128MB. Chunk size was the default 512kB.
>> It is less than 0.1% of the device...
>>
>>> I cannot see why such a large offset is used. I would think the data
>>> offset need only be at most the chunk size plus the space (1 sector) for
>>> the superblock and bitmap.
>> It is insurance.  If you want to change the chunksize later, having a
>> lot of head-room will allow the reshape to go much faster.
>>
>>> When reshaping the array, I am prompted to use an external file, so I
>>> don't see that mdadm ever uses the space.
>> Citation needed.... what version of mdadm, what kernel?  What reshape
>> command?
> kernel:  9.64-1-lts

I don't know what that means?  Maybe 4.9.64-1-lts.
That's nice and recent.

> mdadm:  mdadm - v4.0 - 2017-01-09
> action:  reduce the number of raid devices

Ahh.  Reducing the number of devices doesn't use the head-space, it used
the end-space.  As you reduce the size of the array when doing this,
there is always lots of end-space.  So I'm surprised that it would want
a backup file.
However without specifics (mdadm -E of devices before the reshape, and
exact command given) I won't be looking into why it might.

Thanks,
NeilBrown


>>>
>>> I tried making some test arrays and got much smaller sizes. A 3x1GB
>>> RAID5 array with 64k chunks had a 1MB data offset.
>>>
>>>
>>> If I make a 7x4TB RAID5 array with 64kB chunks, is there a problem with
>>> setting the data offset to around 2MB?
>> Only that it might reduce your options in the future, though probably
>> not by much.
>>
>> NeilBrown

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 832 bytes --]