Calculating swidth On A RAID6 Volume

Calculating swidth On A RAID6 Volume

[Andrew Debenham]

[-- Attachment #1.1: Type: text/plain, Size: 2165 bytes --]

Hello -

I'm am new to XFS and am in the process of trying to tune an XFS file system.  I spent some time searching through this mailing list's archives but wasn't able to find anything that matched the question that I have (I apologize in advance if this issue has already been addressed).

I have a 3ware 9650SE-12M installed and have a RAID6 unit with 11 x 1.82 TB SATA drives and stripe size of 256KB.  I found an earlier thread that discussed calculating the values for sunit and swidth for RAID5, which stated that the swidth would be "sunit x (# disks in stripe - 1)".  So, finally the question - would the swidth for a RAID6 volume be "sunit x (# disks in stripe -2)?

If that is the case, then here is what I would expect my values to be:
sunit = 512
swidth = 4608

Would that be correct?  Thanks in advance for your help in this matter!

- Andrew

________________________________
This email and any attachments may be confidential and are intended solely for the use of the individual to whom it is addressed.

If you are not the intended recipient of this email, the following stipulations govern the use of this information: You may not take any action based upon its contents. You may not copy or show this message or attachments to anyone. You should contact the sender and subsequently delete this message and all attachments.

Any views or opinions expressed are solely those of the author and do not necessarily represent those of Special Operations Technology, Incorporated.

While antivirus software has been applied, you should perform due diligence to check this email and attachments for the presence of viruses. No warranties or assurances are made in relation to the safety and content of this email and attachments. Special Operations Technology, Incorporated accepts no liability for any damages caused by any virus transmitted by or contained in this email and attachments.

No liability is accepted for any consequences arising from this email transmission whatsoever.

Special Operations Technology, Incorporated is a premier IT professional services firm focused in the government and law enforcement space.

[-- Attachment #1.2: Type: text/html, Size: 3586 bytes --]

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

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

Re: Calculating swidth On A RAID6 Volume

[Michael Weissenbacher]

> finally the question - would the swidth for a RAID6 volume be "sunit x
> (# disks in stripe -2)?
Yep, correct

> If that is the case, then here is what I would expect my values to be:
> sunit = 512
> swidth = 4608
>  
> Would that be correct?  Thanks in advance for your help in this matter!
Correct too, AFAICT

hth,
Michael

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

Re: Calculating swidth On A RAID6 Volume

[Michael Monnerie]

[-- Attachment #1.1: Type: Text/Plain, Size: 1898 bytes --]

On Dienstag, 20. Juli 2010 Andrew Debenham wrote:
> I'm am new to XFS and am in the process of trying to tune an XFS file
>  system.  I spent some time searching through this mailing list's
>  archives but wasn't able to find anything that matched the question
>  that I have (I apologize in advance if this issue has already been
>  addressed).

Thanks for reminding me, wanted to write a FAQ entry, done now:
http://xfs.org/index.php/XFS_FAQ#Q:_How_to_calculate_the_correct_sunit.2Cswidth_values_for_optimal_performance

> I have a 3ware 9650SE-12M installed and have a RAID6 unit with 11 x
>  1.82 TB SATA drives and stripe size of 256KB.  I found an earlier
>  thread that discussed calculating the values for sunit and swidth
>  for RAID5, which stated that the swidth would be "sunit x (# disks
>  in stripe - 1)".  So, finally the question - would the swidth for a
>  RAID6 volume be "sunit x (# disks in stripe -2)?
> 
> If that is the case, then here is what I would expect my values to
>  be: sunit = 512
> swidth = 4608
 
I think sunit should be as big as the RAID controllers stripe size, as 
it describes the smallest I/O XFS should do. So sunit=512 would make it 
possible that XFS writes a single sector, while the RAID controller 
needs to read/write 256KB with your stripe size, resulting in 
performance degradation. But I'm no XFS dev, so maybe wait for a 
clearing from someone who has deeper XFS knowledge than me.

sunit=262144
swidth=9

Would be my suggestion for your setup.

-- 
mit freundlichen Grüssen,
Michael Monnerie, Ing. BSc

it-management Internet Services
http://proteger.at [gesprochen: Prot-e-schee]
Tel: 0660 / 415 65 31

****** Aktuelles Radiointerview! ******
http://www.it-podcast.at/aktuelle-sendung.html

// Wir haben im Moment zwei Häuser zu verkaufen:
// http://zmi.at/langegg/
// http://zmi.at/haus2009/

[-- Attachment #1.2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 198 bytes --]

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

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

Re: Calculating swidth On A RAID6 Volume

[Michael Weissenbacher]

Michael,
>> If that is the case, then here is what I would expect my values to
>>  be: sunit = 512
>> swidth = 4608
>  
> I think sunit should be as big as the RAID controllers stripe size, as 
> it describes the smallest I/O XFS should do. So sunit=512 would make it 
> possible that XFS writes a single sector, while the RAID controller 
> needs to read/write 256KB with your stripe size, resulting in 
> performance degradation. But I'm no XFS dev, so maybe wait for a 
> clearing from someone who has deeper XFS knowledge than me.
> 
> sunit=262144
> swidth=9
> 
> Would be my suggestion for your setup.
> 
IIRC sunit and swidth are both specified in units of 512 bytes (don't
ask me why) so a sunit value of 512 would mean 256KB stripe size which
is correct :-)

If i am right, the FAQ should be corrected (hopefully some dev can
confirm/deny it)

cheers,
Michael

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

Re: Calculating swidth On A RAID6 Volume

[Emmanuel Florac]

Le Tue, 20 Jul 2010 16:13:29 +0200
Michael Weissenbacher <mw@dermichi.com> écrivait:

>  
> IIRC sunit and swidth are both specified in units of 512 bytes (don't
> ask me why) 

Because this is the disk block size for all disks drives ever made
except the very latest ones (4096 bytes blocks) and the special
RAID-formatted 520 bytes sometimes in use in the past.

> so a sunit value of 512 would mean 256KB stripe size which
> is correct :-)

exactly. Or use su and sw to specify the size in k, m, etc.

-- 
------------------------------------------------------------------------
Emmanuel Florac     |   Direction technique
                    |   Intellique
                    |	<eflorac@intellique.com>
                    |   +33 1 78 94 84 02
------------------------------------------------------------------------

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

Re: Calculating swidth On A RAID6 Volume

[Michael Monnerie]

[-- Attachment #1.1: Type: Text/Plain, Size: 1123 bytes --]

On Dienstag, 20. Juli 2010 Emmanuel Florac wrote:
> > so a sunit value of 512 would mean 256KB stripe size which
> > is correct :-)
> 
> exactly. Or use su and sw to specify the size in k, m, etc.
 
Ah, I remember the difference now, that's why I use su= and sw=. This 
mix of different units is... irritating.

I corrected the FAQ now to use su+sw:
http://xfs.org/index.php/XFS_FAQ#Q:_How_to_calculate_the_correct_sunit.2Cswidth_values_for_optimal_performance
And I mentioned "swidth" and "sunit" in a sentence in the end.

I'd say using su+sw is more future proof than swidth+sunit, as 4K sector 
drives will become standard, and then using 512B units will be outdated 
anyway, right? And using su+sw seems simpler to me to explain. YMMV, 

-- 
mit freundlichen Grüssen,
Michael Monnerie, Ing. BSc

it-management Internet Services
http://proteger.at [gesprochen: Prot-e-schee]
Tel: 0660 / 415 65 31

****** Aktuelles Radiointerview! ******
http://www.it-podcast.at/aktuelle-sendung.html

// Wir haben im Moment zwei Häuser zu verkaufen:
// http://zmi.at/langegg/
// http://zmi.at/haus2009/

[-- Attachment #1.2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 198 bytes --]

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

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

Re: Calculating swidth On A RAID6 Volume

[Michael Monnerie]

[-- Attachment #1.1: Type: Text/Plain, Size: 615 bytes --]

On Dienstag, 20. Juli 2010 Emmanuel Florac wrote:
> exactly. Or use su and sw to specify the size in k, m, etc.
 
BTW, "man mount" only mentions "swidth/sunit", while "man mkfs.xfs" 
mentions "su/sw". Maybe someone can update the docs of mount?

-- 
mit freundlichen Grüssen,
Michael Monnerie, Ing. BSc

it-management Internet Services
http://proteger.at [gesprochen: Prot-e-schee]
Tel: 0660 / 415 65 31

****** Aktuelles Radiointerview! ******
http://www.it-podcast.at/aktuelle-sendung.html

// Wir haben im Moment zwei Häuser zu verkaufen:
// http://zmi.at/langegg/
// http://zmi.at/haus2009/

[-- Attachment #1.2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 198 bytes --]

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

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

Re: Calculating swidth On A RAID6 Volume

[Emmanuel Florac]

[-- Attachment #1.1: Type: text/plain, Size: 926 bytes --]

Le Tue, 20 Jul 2010 17:00:51 +0200
Michael Monnerie <michael.monnerie@is.it-management.at> écrivait:

> I'd say using su+sw is more future proof than swidth+sunit, as 4K
> sector drives will become standard, and then using 512B units will be
> outdated anyway, right?

Absolutely, and furthermore I'm wondering what's happening in the
case where the drives have 4096 bytes blocks; I suppose then sunit
should be expressed as a number of 4096 bytes blocks, and what about
swidth ? the hell if I know :) This is also probably a nice little nest
of coming filesystem bugs to be hatched :=)

-- 
------------------------------------------------------------------------
Emmanuel Florac     |   Direction technique
                    |   Intellique
                    |	<eflorac@intellique.com>
                    |   +33 1 78 94 84 02
------------------------------------------------------------------------

[-- Attachment #1.2: signature.asc --]
[-- Type: application/pgp-signature, Size: 197 bytes --]

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

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

Re: Calculating swidth On A RAID6 Volume

[Dave Chinner]

On Tue, Jul 20, 2010 at 05:27:04PM +0200, Emmanuel Florac wrote:
> Le Tue, 20 Jul 2010 17:00:51 +0200
> Michael Monnerie <michael.monnerie@is.it-management.at> écrivait:
> 
> > I'd say using su+sw is more future proof than swidth+sunit, as 4K
> > sector drives will become standard, and then using 512B units will be
> > outdated anyway, right?
> 
> Absolutely, and furthermore I'm wondering what's happening in the
> case where the drives have 4096 bytes blocks; I suppose then sunit
> should be expressed as a number of 4096 bytes blocks, and what about
> swidth ? the hell if I know :) This is also probably a nice little nest
> of coming filesystem bugs to be hatched :=)

No, the sky is not going to fall. ;)

The stripe unit is not related to sector size except for the fact
that sector size defines the minimum filesystem block size and the
stripe unit is an integer mutliple of the filesystem block size.
i.e.

mkfs.xfs -s size=512 -b size=512 -d sunit=1,swidth=1

is a valid configuration that gives single sector sunit/swidth
alignment. Hence sunit needs to be able to express sizes in
multiples of 512 bytes.

For a 4k sector size drive, the equivalent is:

mkfs.xfs -s size=4096 -b size=4096 -d sunit=8,swidth=8

and mkfs.xfs will ensure that any value of sunit that is not
a multiple of 8 (4k) will be rejected.....

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

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