What are the correct mkfs.xfs parameters for a lying WD-EARS HDD?

What are the correct mkfs.xfs parameters for a lying WD-EARS HDD?

[Matthias Schniedermeyer]

Hi


More than a month ago i bought 4 Western Digital WD15EARS (1.5 TB) which 
are (AFAIK) the first general/commercial available 4k sector SATA-HDDs.

Unfortunatly the HDDs lie about the 4k physical sector size and the most 
prominent drawback is a worse than abysmal delete performance.
("Normal" Read & Write-performance is OK)

So if i wanted to (re-)mkfs the filesystems what would the correct 
parameters be?

Kernel/Userspace is pretty recent (Debian-SID):
mkfs.xfs version 3.1.1, kernel v2.6.33, util-linux 2.16.0
Not that that should matter when the HDDs lies.





Bis denn

-- 
Real Programmers consider "what you see is what you get" to be just as 
bad a concept in Text Editors as it is in women. No, the Real Programmer
wants a "you asked for it, you got it" text editor -- complicated, 
cryptic, powerful, unforgiving, dangerous.

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

Re: What are the correct mkfs.xfs parameters for a lying WD-EARS HDD?

[Eric Sandeen]

Matthias Schniedermeyer wrote:
> Hi
> 
> 
> More than a month ago i bought 4 Western Digital WD15EARS (1.5 TB) which 
> are (AFAIK) the first general/commercial available 4k sector SATA-HDDs.
> 
> Unfortunatly the HDDs lie about the 4k physical sector size and the most 
> prominent drawback is a worse than abysmal delete performance.
> ("Normal" Read & Write-performance is OK)
> 
> So if i wanted to (re-)mkfs the filesystems what would the correct 
> parameters be?
> 
> Kernel/Userspace is pretty recent (Debian-SID):
> mkfs.xfs version 3.1.1, kernel v2.6.33, util-linux 2.16.0
> Not that that should matter when the HDDs lies.

Recent kernel+util-linux-ng++fdisk+parted+xfsprogs -should- do the right thing for you....
Oh, but this was maybe the drive that didn't output the right stuff when queried.

Make sure your partitions, if any, are on 4k boundaries.(*)  older fdisk at least
won't do this by default, not sure about parted.

once that is done, tell mkfs.xfs "-s size=4096" to set the 4k sector size
(again, all-upstream should do this magically for sane drives)

-Eric

(*)unless the drive has an offset to make 512-sector 63 line up on a nice
boundary... in which case I guess you could experiment with perf both
ways to be sure...

> 
> Bis denn
> 

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

Re: What are the correct mkfs.xfs parameters for a lying WD-EARS HDD?

[Matthias Schniedermeyer]

On 08.03.2010 16:45, Eric Sandeen wrote:
> Matthias Schniedermeyer wrote:
> > Hi
> > 
> > 
> > More than a month ago i bought 4 Western Digital WD15EARS (1.5 TB) which 
> > are (AFAIK) the first general/commercial available 4k sector SATA-HDDs.
> > 
> > Unfortunatly the HDDs lie about the 4k physical sector size and the most 
> > prominent drawback is a worse than abysmal delete performance.
> > ("Normal" Read & Write-performance is OK)
> > 
> > So if i wanted to (re-)mkfs the filesystems what would the correct 
> > parameters be?
> > 
> > Kernel/Userspace is pretty recent (Debian-SID):
> > mkfs.xfs version 3.1.1, kernel v2.6.33, util-linux 2.16.0
> > Not that that should matter when the HDDs lies.
> 
> Recent kernel+util-linux-ng++fdisk+parted+xfsprogs -should- do the right thing for you....
> Oh, but this was maybe the drive that didn't output the right stuff when queried.

hdparm -I /dev/sdg | grep "Sector size"
        Logical/Physical Sector size:           512 bytes

hdparm -V
hdparm v9.27

I vagely remember a posting on LKML that said it's a "known problem" 
that the WD??EARS lie about physical sector size.

> Make sure your partitions, if any, are on 4k boundaries.(*)  older fdisk at least
> won't do this by default, not sure about parted.

It's an encrypted loop at am 4k offset to leave space for a fake MBR 
(and more importantly a signature for a custom udev-script).

> once that is done, tell mkfs.xfs "-s size=4096" to set the 4k sector size

So that should be enough?
Time for backup/mkfs/restore then.





Bis denn

-- 
Real Programmers consider "what you see is what you get" to be just as 
bad a concept in Text Editors as it is in women. No, the Real Programmer
wants a "you asked for it, you got it" text editor -- complicated, 
cryptic, powerful, unforgiving, dangerous.

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

Re: What are the correct mkfs.xfs parameters for a lying WD-EARS HDD?

[Matthias Schniedermeyer]

On 09.03.2010 00:47, Matthias Schniedermeyer wrote:
> On 08.03.2010 16:45, Eric Sandeen wrote:
> > Matthias Schniedermeyer wrote:
> > > Hi
> > > 
> > > 
> > > More than a month ago i bought 4 Western Digital WD15EARS (1.5 TB) which 
> > > are (AFAIK) the first general/commercial available 4k sector SATA-HDDs.
> > > 
> > > Unfortunatly the HDDs lie about the 4k physical sector size and the most 
> > > prominent drawback is a worse than abysmal delete performance.
> > > ("Normal" Read & Write-performance is OK)
> > > 
> > > So if i wanted to (re-)mkfs the filesystems what would the correct 
> > > parameters be?
> > 
> > once that is done, tell mkfs.xfs "-s size=4096" to set the 4k sector size
> 
> So that should be enough?
> Time for backup/mkfs/restore then.

backup/mkfs/restore is finished.

I'm currently testing delete-performance.
And it improved to abysmal performance. (As expected)

Read-latency, while deleting, has also improved drastically. Before the 
FS was near unusable while deleting. But that is also to be expected 
from such a HDD when it is doing RMW like there is no tomorrow. ;-)






Bis denn

-- 
Real Programmers consider "what you see is what you get" to be just as 
bad a concept in Text Editors as it is in women. No, the Real Programmer
wants a "you asked for it, you got it" text editor -- complicated, 
cryptic, powerful, unforgiving, dangerous.

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

Re: What are the correct mkfs.xfs parameters for a lying WD-EARS HDD?

[Gim Leong Chin]

Hi,


You can use GPT fdisk (gdisk) to do the partitioning.  The author had the exact same problem with the performance of WD 4 kB sector drives.  The tool automatically sets the partition start and end at 4 kB boundaries.  You can in fact set the partition start and end at any sector you want.

It is the firmware in the WD drives that translates the physical 4 kB sector to 512 byte sectors.


GL

--- On Wed, 10/3/10, Matthias Schniedermeyer <ms@citd.de> wrote:

> From: Matthias Schniedermeyer <ms@citd.de>
> Subject: Re: What are the correct mkfs.xfs parameters for a lying WD-EARS HDD?
> To: "Eric Sandeen" <sandeen@sandeen.net>
> Cc: xfs@oss.sgi.com
> Date: Wednesday, 10 March, 2010, 5:41 AM
> On 09.03.2010 00:47, Matthias
> Schniedermeyer wrote:
> > On 08.03.2010 16:45, Eric Sandeen wrote:
> > > Matthias Schniedermeyer wrote:
> > > > Hi
> > > > 
> > > > 
> > > > More than a month ago i bought 4 Western
> Digital WD15EARS (1.5 TB) which 
> > > > are (AFAIK) the first general/commercial
> available 4k sector SATA-HDDs.
> > > > 
> > > > Unfortunatly the HDDs lie about the 4k
> physical sector size and the most 
> > > > prominent drawback is a worse than abysmal
> delete performance.
> > > > ("Normal" Read & Write-performance is
> OK)
> > > > 
> > > > So if i wanted to (re-)mkfs the filesystems
> what would the correct 
> > > > parameters be?
> > > 
> > > once that is done, tell mkfs.xfs "-s size=4096"
> to set the 4k sector size
> > 
> > So that should be enough?
> > Time for backup/mkfs/restore then.
> 
> backup/mkfs/restore is finished.
> 
> I'm currently testing delete-performance.
> And it improved to abysmal performance. (As expected)
> 
> Read-latency, while deleting, has also improved
> drastically. Before the 
> FS was near unusable while deleting. But that is also to be
> expected 
> from such a HDD when it is doing RMW like there is no
> tomorrow. ;-)
> 
> 
> 
> 
> 
> 
> Bis denn
> 
> -- 
> Real Programmers consider "what you see is what you get" to
> be just as 
> bad a concept in Text Editors as it is in women. No, the
> Real Programmer
> wants a "you asked for it, you got it" text editor --
> complicated, 
> cryptic, powerful, unforgiving, dangerous.
> 
> _______________________________________________
> xfs mailing list
> xfs@oss.sgi.com
> http://oss.sgi.com/mailman/listinfo/xfs
> 


      New Email names for you! 
Get the Email name you&#39;ve always wanted on the new @ymail and @rocketmail. 
Hurry before someone else does!
http://mail.promotions.yahoo.com/newdomains/sg/

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

Re: What are the correct mkfs.xfs parameters for a lying WD-EARS HDD?

[Martin K. Petersen]

>>>>> "Eric" == Eric Sandeen <sandeen@sandeen.net> writes:

Eric> Make sure your partitions, if any, are on 4k boundaries.(*) older
Eric> fdisk at least won't do this by default, not sure about parted.

You can use the expert mode in fdisk to ensure that the partition begins
on a 4KB boundary.


Eric> (*)unless the drive has an offset to make 512-sector 63 line up on
Eric> a nice boundary... in which case I guess you could experiment with
Eric> perf both ways to be sure...

EARS is the one that's lying about the physical block size.  It's
zero-aligned unless somebody messed with the jumper.

-- 
Martin K. Petersen	Oracle Linux Engineering

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

Re: What are the correct mkfs.xfs parameters for a lying WD-EARS HDD?

[Eric Sandeen]

Martin K. Petersen wrote:
>>>>>> "Eric" == Eric Sandeen <sandeen@sandeen.net> writes:
> 
> Eric> Make sure your partitions, if any, are on 4k boundaries.(*) older
> Eric> fdisk at least won't do this by default, not sure about parted.
> 
> You can use the expert mode in fdisk to ensure that the partition begins
> on a 4KB boundary.

and use -u to give it sector units ...

> 
> Eric> (*)unless the drive has an offset to make 512-sector 63 line up on
> Eric> a nice boundary... in which case I guess you could experiment with
> Eric> perf both ways to be sure...
> 
> EARS is the one that's lying about the physical block size.  It's
> zero-aligned unless somebody messed with the jumper.

Ok, that's good.

Thanks Martin!

-Eric

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