change existing RAID10 to "far layout"

[Reindl Harald]

Am 31.12.2011 16:11, schrieb Roberto Ragusa:
> On 12/31/2011 02:11 AM, Reindl Harald wrote:
>>>> for better performance you could use RAID10 in the far layout instead (raid10,f2),
>>>> this is a direct replacement with the enhanced raid10 driver, which gives double
>>>> the sequential read speed compared to raid1. Use "--level=10 -p f2" as additional
>>>> parameters when creating arrays with mdadm.
>>
>> hmm is there a way to change this in an existing RAID10?
>> i think i would like this for / where read-performance is much
>> more important and /dev/md2 is perfectly with near-layout
>> for write-performance (hughe data, virtual machines)
> 
> It is apparently impossible to change the layout from near to far.
> If you think about how different they are, it is difficult to
> imagine how a conversion could be done.
> 
> The best option is to boot from a rescue image, copy the root partition
> somewhere else, destroy and redo the RAID and copy the root partition back.
> 
> In that case, you may want to consider the offset layout too.
> After some experiments when making a filesystem for VM files, the best
> option proved to be offset=2 with chunksize=4MiB.
> Very good linear read speed, and good write performance even on random writes.
> 
> # cat /proc/mdstat
> Personalities : [raid10]
> md10 : active raid10 sdd5[3] sdc5[2] sdb3[1] sda3[0]
>       285196288 blocks super 1.2 4096K chunks 2 offset-copies [4/4] [UUUU]
> 
> unused devices: <none>
> # hdparm -t /dev/md10
> 
> /dev/md10:
>  Timing buffered disk reads:  922 MB in  3.02 seconds = 305.11 MB/sec

OK, so better leave it as it is :-)

> 200 MB/Secnod are not too bad and i considered it only if
it is not too much risk, but not risk setup the machine again
while there are two physical cloesn running perfect at all

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 262 bytes
Desc: OpenPGP digital signature
URL: <http://lists.fedoraproject.org/pipermail/users/attachments/20111231/49f2c0be/attachment-0001.sig>