XFS sunit/swidth for raid10

XFS sunit/swidth for raid10

[Peter Rabbitson]

Hi,
How does one determine the XFS sunit and swidth sizes for a software 
raid10 with 3 copies?

Thanks

Peter

Re: XFS sunit/swidth for raid10

[dean gaudet]

On Thu, 22 Mar 2007, Peter Rabbitson wrote:

> Hi,
> How does one determine the XFS sunit and swidth sizes for a software raid10
> with 3 copies?

mkfs.xfs uses the GET_ARRAY_INFO ioctl to get the data it needs from 
software raid and select an appropriate sunit/swidth...

although i'm not sure i agree entirely with its choice for raid10:

                *sunit = md.chunk_size >> 9;
                *swidth = *sunit * md.raid_disks;

i'd think it would depend on the layout of the raid10 (near, far, 
offset)... for near2 on 4 disks i'd expect swidth to be only 2*sunit... 
but for far2 on 4 disks i'd expect 4*sunit... but i'm not sure.  it 
probably doesn't matter.

-dean

Re: XFS sunit/swidth for raid10

[Peter Rabbitson]

dean gaudet wrote:
> On Thu, 22 Mar 2007, Peter Rabbitson wrote:
> 
>> Hi,
>> How does one determine the XFS sunit and swidth sizes for a software raid10
>> with 3 copies?
> 
> mkfs.xfs uses the GET_ARRAY_INFO ioctl to get the data it needs from 
> software raid and select an appropriate sunit/swidth...
>
> although i'm not sure i agree entirely with its choice for raid10:

So do I, especially as it makes no checks for the amount of copies (3 in 
my case, not 2).

>it probably doesn't matter.

This was essentially my question. For an array -pf3 -c1024 I get swidth 
= 4 * sunit = 4MiB. Is it about right and does it matter at all?

Re: XFS sunit/swidth for raid10

[dean gaudet]

On Thu, 22 Mar 2007, Peter Rabbitson wrote:

> dean gaudet wrote:
> > On Thu, 22 Mar 2007, Peter Rabbitson wrote:
> > 
> > > Hi,
> > > How does one determine the XFS sunit and swidth sizes for a software
> > > raid10
> > > with 3 copies?
> > 
> > mkfs.xfs uses the GET_ARRAY_INFO ioctl to get the data it needs from
> > software raid and select an appropriate sunit/swidth...
> > 
> > although i'm not sure i agree entirely with its choice for raid10:
> 
> So do I, especially as it makes no checks for the amount of copies (3 in my
> case, not 2).
> 
> > it probably doesn't matter.
> 
> This was essentially my question. For an array -pf3 -c1024 I get swidth = 4 *
> sunit = 4MiB. Is it about right and does it matter at all?

how many drives?

-dean

Re: XFS sunit/swidth for raid10

[Peter Rabbitson]

dean gaudet wrote:
> On Thu, 22 Mar 2007, Peter Rabbitson wrote:
> 
>> dean gaudet wrote:
>>> On Thu, 22 Mar 2007, Peter Rabbitson wrote:
>>>
>>>> Hi,
>>>> How does one determine the XFS sunit and swidth sizes for a software
>>>> raid10
>>>> with 3 copies?
>>> mkfs.xfs uses the GET_ARRAY_INFO ioctl to get the data it needs from
>>> software raid and select an appropriate sunit/swidth...
>>>
>>> although i'm not sure i agree entirely with its choice for raid10:
>> So do I, especially as it makes no checks for the amount of copies (3 in my
>> case, not 2).
>>
>>> it probably doesn't matter.
>> This was essentially my question. For an array -pf3 -c1024 I get swidth = 4 *
>> sunit = 4MiB. Is it about right and does it matter at all?
> 
> how many drives?
> 

Sorry. 4 drives, 3 far copies (so any 2 drives can fail), 1M chunk.

Re: XFS sunit/swidth for raid10

[dean gaudet]

On Fri, 23 Mar 2007, Peter Rabbitson wrote:

> dean gaudet wrote:
> > On Thu, 22 Mar 2007, Peter Rabbitson wrote:
> > 
> > > dean gaudet wrote:
> > > > On Thu, 22 Mar 2007, Peter Rabbitson wrote:
> > > > 
> > > > > Hi,
> > > > > How does one determine the XFS sunit and swidth sizes for a software
> > > > > raid10
> > > > > with 3 copies?
> > > > mkfs.xfs uses the GET_ARRAY_INFO ioctl to get the data it needs from
> > > > software raid and select an appropriate sunit/swidth...
> > > > 
> > > > although i'm not sure i agree entirely with its choice for raid10:
> > > So do I, especially as it makes no checks for the amount of copies (3 in
> > > my
> > > case, not 2).
> > > 
> > > > it probably doesn't matter.
> > > This was essentially my question. For an array -pf3 -c1024 I get swidth =
> > > 4 *
> > > sunit = 4MiB. Is it about right and does it matter at all?
> > 
> > how many drives?
> > 
> 
> Sorry. 4 drives, 3 far copies (so any 2 drives can fail), 1M chunk.

my mind continues to be blown by linux raid10.

so that's like raid1 on 4 disks except the copies are offset by 1/4th of 
the disk?

i think swidth = 4*sunit is the right config then -- 'cause a read of 4MiB 
will stride all 4 disks...

-dean