mkfs.xfs -d su=XXX,sw=YYY and future volume resize

mkfs.xfs -d su=XXX,sw=YYY and future volume resize

[Marcin Sura]

[-- Attachment #1.1: Type: text/plain, Size: 425 bytes --]

Hi,

When I create filesystem I can specify su and sw paramteres which match
underlying storage device.
So for example, If I have raid 6 volume made of 6 + 2 disks with stripe
size of 256k I create xfs with:

mkfs.xfs -d su=256k,sw=6 <disk>

But what if I will extend original volume by 2 disks. sw will 8. Will be
there a performance panelty compared to filesystem created from scratch
with correct sw parameter?

BR
Marcin

[-- Attachment #1.2: Type: text/html, Size: 569 bytes --]

[-- Attachment #2: Type: text/plain, Size: 121 bytes --]

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: mkfs.xfs -d su=XXX,sw=YYY and future volume resize

[Dave Chinner]

On Thu, Jul 14, 2016 at 10:19:02PM +0200, Marcin Sura wrote:
> Hi,
> 
> When I create filesystem I can specify su and sw paramteres which match
> underlying storage device.
> So for example, If I have raid 6 volume made of 6 + 2 disks with stripe
> size of 256k I create xfs with:
> 
> mkfs.xfs -d su=256k,sw=6 <disk>
> 
> But what if I will extend original volume by 2 disks. sw will 8. Will be
> there a performance panelty compared to filesystem created from scratch
> with correct sw parameter?

That depends on your workload more than anything. If the workload
benefits from correct stripe alignment of your data and IO, then it
will definitely perform worse if you change the stripe width so that
allocation is no longer able to align correctly to the physical
geometry of the storage.

-Dave.
-- 
Dave Chinner
david@fromorbit.com

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: mkfs.xfs -d su=XXX,sw=YYY and future volume resize

[Carlos Maiolino]

On Thu, Jul 14, 2016 at 10:19:02PM +0200, Marcin Sura wrote:
>    Hi,
>    When I create filesystem I can specify su and sw paramteres which match
>    underlying storage device.
>    So for example, If I have raid 6 volume made of 6 + 2 disks with stripe
>    size of 256k I create xfs with:
>    mkfs.xfs -d su=256k,sw=6 <disk>
>    But what if I will extend original volume by 2 disks. sw will 8. Will
>    be there a performance panelty compared to filesystem created from
>    scratch with correct sw parameter?
>    BR
>    Marcin

Complementing Dave's answer, you can also mount the filesystem with new
alignment configuration.

You can use: sunit=value and swidth=value for mounting the FS with the new
settings. man xfs for more info.

Cheers

> _______________________________________________
> xfs mailing list
> xfs@oss.sgi.com
> http://oss.sgi.com/mailman/listinfo/xfs


-- 
Carlos

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: mkfs.xfs -d su=XXX,sw=YYY and future volume resize

[Dave Chinner]

On Fri, Jul 15, 2016 at 11:46:47AM +0200, Carlos Maiolino wrote:
> On Thu, Jul 14, 2016 at 10:19:02PM +0200, Marcin Sura wrote:
> >    Hi,
> >    When I create filesystem I can specify su and sw paramteres which match
> >    underlying storage device.
> >    So for example, If I have raid 6 volume made of 6 + 2 disks with stripe
> >    size of 256k I create xfs with:
> >    mkfs.xfs -d su=256k,sw=6 <disk>
> >    But what if I will extend original volume by 2 disks. sw will 8. Will
> >    be there a performance panelty compared to filesystem created from
> >    scratch with correct sw parameter?
> >    BR
> >    Marcin
> 
> Complementing Dave's answer, you can also mount the filesystem with new
> alignment configuration.
> 
> You can use: sunit=value and swidth=value for mounting the FS with the new
> settings. man xfs for more info.

That doesn't change the fact that everything that already existed in
the filesystem would now be misaligned and hence have performance
issues on read and/or overwrite. Only newly written data will be
aligned to the new stripe width.

Also, keep in mind that stripw width cahnges are an alignment change
you generally canot make through the mount options. Static metadata
such as allocation group headers are laid out according to the
physical geometry of the filesystem, and that cannot be changed.
hence the mount options can only change the stripe unit/width when
they don't change the alignment constraints. e.g. doubling of the
stripe width will work because the alignemtn of the stripe units is
still the same. However, changing from 6 disks to 8 won't work
because it is a different physical alignment...

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: mkfs.xfs -d su=XXX,sw=YYY and future volume resize

[Carlos Maiolino]

On Sat, Jul 16, 2016 at 09:07:25AM +1000, Dave Chinner wrote:
> On Fri, Jul 15, 2016 at 11:46:47AM +0200, Carlos Maiolino wrote:
> > On Thu, Jul 14, 2016 at 10:19:02PM +0200, Marcin Sura wrote:
> > >    Hi,
> > >    When I create filesystem I can specify su and sw paramteres which match
> > >    underlying storage device.
> > >    So for example, If I have raid 6 volume made of 6 + 2 disks with stripe
> > >    size of 256k I create xfs with:
> > >    mkfs.xfs -d su=256k,sw=6 <disk>
> > >    But what if I will extend original volume by 2 disks. sw will 8. Will
> > >    be there a performance panelty compared to filesystem created from
> > >    scratch with correct sw parameter?
> > >    BR
> > >    Marcin
> > 
> > Complementing Dave's answer, you can also mount the filesystem with new
> > alignment configuration.
> > 
> > You can use: sunit=value and swidth=value for mounting the FS with the new
> > settings. man xfs for more info.
> 
> That doesn't change the fact that everything that already existed in
> the filesystem would now be misaligned and hence have performance
> issues on read and/or overwrite. Only newly written data will be
> aligned to the new stripe width.
> 
> Also, keep in mind that stripw width cahnges are an alignment change
> you generally canot make through the mount options. Static metadata
> such as allocation group headers are laid out according to the
> physical geometry of the filesystem, and that cannot be changed.
> hence the mount options can only change the stripe unit/width when
> they don't change the alignment constraints. e.g. doubling of the
> stripe width will work because the alignemtn of the stripe units is
> still the same. However, changing from 6 disks to 8 won't work
> because it is a different physical alignment...
> 

Thanks for the detailed info, it was my bad to have not predicted such details


> Cheers,
> 
> Dave.
> -- 
> Dave Chinner
> david@fromorbit.com
> 
> _______________________________________________
> xfs mailing list
> xfs@oss.sgi.com
> http://oss.sgi.com/mailman/listinfo/xfs

-- 
Carlos

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs