headers
Kai Leibrandt | 1 Feb 2003 13:52
Picon
Favicon

Another mkfs.xfs RAID Question

Hi all,

Dragos' question about the su,sw options led me to have another look at
my Mylex RAID setup, and I now am confused too...
My RAID controller tells me that the primary RAID volume (a 3-disk RAID
5) has a Stripe Size of 64k and a Segment Size of 8k. So for this setup
should the sunit be set to 64k or to 8k (the segment size)??? The manual
fo the DAC says that the segment size is the preferred io size for
caching, but xfs has a preferred io size equal to the stripe size -
which to choose?

Many thanks,

Kai.
Permalink | Reply | 1 comment |
headers
Christian Guggenberger | 1 Feb 2003 14:04
Picon
Favicon

Yet Another mkfs.xfs RAID Question

Hi,

I'm going to set up a new 1.7 TB HW Raid 5 next week.
RAID 5 contains 11 disks, Stripe Unit should be (according to the manual)
128k.
so my mkfs.xfs options will be sunit=256, swidth=2560 for the data section, won't they?

I will definitly use internal log, so I'd like to ask, if I should use
logversion 2, and what sunit and swidth values I should use here?
I guess the same as for the data section???

Would I gain something from logversion 2?

thx
Christian
Permalink | Reply | 5 comments |
headers
Andries.Brouwer | 1 Feb 2003 17:45
Picon
Picon
Favicon

Re: system call documentation

    From kaos <at> ocs.com.au  Sat Feb  1 14:22:31 2003

    >Preparing the next man page release, I compared the list of
    >system calls for i386 in 2.4.20 with the list of documented
    >system calls. It looks like
    >
    >fgetxattr,
    > ...
    >are undocumented so far.

    *xattr* man pages are in the XFS tree and Andreas Gruenbacher's site,
    contents forwarded under separate copy.

    getxattr.2:    getxattr, lgetxattr, fgetxattr2
    listxattr.2:    listxattr, llistxattr, flistxattr
    removexattr.2:    removexattr, lremovexattr, fremovexattr
    setxattr.2:    setxattr, lsetxattr, fsetxattr

Good. Thanks!

However,

.\" (C) Andreas Gruenbacher, February 2001
.\" (C) Silicon Graphics Inc, September 2001

there is no indication that redistribution (of possibly modified
copies) is permitted.

Andries
(Continue reading)

Permalink | Reply | 7 comments |
headers
Walt H | 1 Feb 2003 18:07
Picon

Re: Yet Another mkfs.xfs RAID Question

Christian Guggenberger wrote:
> Hi,
> 
> I'm going to set up a new 1.7 TB HW Raid 5 next week.
> RAID 5 contains 11 disks, Stripe Unit should be (according to the manual)
> 128k.
> so my mkfs.xfs options will be sunit=256, swidth=2560 for the data section, won't they?
> 
> I will definitly use internal log, so I'd like to ask, if I should use
> logversion 2, and what sunit and swidth values I should use here?
> I guess the same as for the data section???
> 
> Would I gain something from logversion 2?
> 
> thx
> Christian
> 
> 

Are you using software md raid or hardware raid? If it's hardware raid 
5, the logversion argument shouldn't matter. Software raid is another 
story. I recently setup a file/database server with a six disk software 
raid 5 setup. I had time to try different raid chunksizes as well as 
experiment with version 1 vs. version 2 logs. What I found, for my case, 
version 2 logs for xfs really helped out in create/delete operations. 
Particularly deletes. Sequential read/writes were unaffected by the 
version differences. I ran many Bonnie++ runs as well as created a 
script that created 2000 directories with 10000 files in each directory 
and then proceeded to delete the whole lot. Each result was timed, 
although I don't have any numbers for you. I remember version 2 logs as
(Continue reading)

Permalink | Reply | 0 comments |
headers
Martin K. Petersen | 1 Feb 2003 22:03

Re: Another mkfs.xfs RAID Question

>>>>> "Kai" == Kai Leibrandt <k_leibrandt <at> hotmail.com> writes:

Kai> Dragos' question about the su,sw options led me to have another
Kai> look at my Mylex RAID setup, and I now am confused too...  My
Kai> RAID controller tells me that the primary RAID volume (a 3-disk
Kai> RAID 5) has a Stripe Size of 64k and a Segment Size of 8k. So for
Kai> this setup should the sunit be set to 64k or to 8k (the segment
Kai> size)??? The manual fo the DAC says that the segment size is the
Kai> preferred io size for caching, but xfs has a preferred io size
Kai> equal to the stripe size - which to choose?

Hmmm.  My guess would be that the Mylex uses 8k hardware sectors
internally and does read/modify/write on those.

Whether the 64K is stripe width or stripe unit, I don't know.

The real answer is the same as always: Try both settings with your
anticipated I/O load and see which one performs better.

--

-- 
Martin K. Petersen      http://mkp.net/
Permalink | Reply | 0 comments |
headers
Deborah CABESSA | 1 Feb 2003 23:19

Translation services - Services de traduction

Hello, 

As part of your operations, you perhaps sometimes require translation 
services.  We are a translation bureau offering services into a variety of languages. 
1) Do you ever need such services? 
2) Who in your organization is the person to contact? 

Best Regards, 

Deborah CABESSA 
Phone:  +33 (0)1 45 88 54 11 
Mobile:  +33 (0)6 11 77 57 08 

Bonjour, 

Dans le cadre de vos activités, il vous arrive peut-être parfois de recourir à des services de
traduction. Nous sommes un bureau de traduction et proposons des services dans diverses langues. 
1) Avez-vous besoin de services de ce type ? 
2) Dans votre entreprise, quelle est la personne à contacter ? 

Cordialement, 

Deborah CABESSA 
Tél : +33 (0)1 45 88 54 11 
Mobile : +33 (0)6 11 77 57 08
Permalink | Reply | 0 comments |
headers
Martin K. Petersen | 1 Feb 2003 22:20

Re: Yet Another mkfs.xfs RAID Question

>>>>> "Christian" == Christian Guggenberger <Christian.Guggenberger <at> physik.uni-regensburg.de> writes:

Christian> I'm going to set up a new 1.7 TB HW Raid 5 next week.  RAID
Christian> 5 contains 11 disks, Stripe Unit should be (according to
Christian> the manual) 128k.  so my mkfs.xfs options will be
Christian> sunit=256, swidth=2560 for the data section, won't they?

Yup.

Christian> I will definitly use internal log, so I'd like to ask, if I
Christian> should use logversion 2, and what sunit and swidth values I
Christian> should use here?  I guess the same as for the data
Christian> section???

128KB log alignment seems a bit of an overkill.

Does your controller state which chunk size it uses internally?  Most
controllers use 4-16KB blocks for RAID5.  So try aligning your log to
values in that neighbourhood.

There's no swidth for the log, btw.

--

-- 
Martin K. Petersen      http://mkp.net/
Permalink | Reply | 3 comments |
headers
Christian Guggenberger | 2 Feb 2003 15:26
Picon
Favicon

Re: Yet Another mkfs.xfs RAID Question]

On Sat, Feb 01, 2003 at 04:20:46PM -0500, Martin K. Petersen wrote:
> >>>>> "Christian" == Christian Guggenberger <Christian.Guggenberger <at> physik.uni-regensburg.de> writes:
> 
> Christian> I'm going to set up a new 1.7 TB HW Raid 5 next week.  RAID
> Christian> 5 contains 11 disks, Stripe Unit should be (according to
> Christian> the manual) 128k.  so my mkfs.xfs options will be
> Christian> sunit=256, swidth=2560 for the data section, won't they?
> 
> Yup.
> 
> 
> Christian> I will definitly use internal log, so I'd like to ask, if I
> Christian> should use logversion 2, and what sunit and swidth values I
> Christian> should use here?  I guess the same as for the data
> Christian> section???
> 
> 128KB log alignment seems a bit of an overkill.
> 
> Does your controller state which chunk size it uses internally?  Most
> controllers use 4-16KB blocks for RAID5.  So try aligning your log to
> values in that neighbourhood.
> 
thanks for your quick answer!
The only Documentation about stripe or chunk size I got from the vendor is,
to use 32k chunk size for random read/write optimaziation or 128k chunk for
sequentiell read/write optimaziation... No mention about what the contoller
does internally!

so I will stay with logversion 1 and sunit, switdh arguments for data
section as mentioned above.
(Continue reading)

Permalink | Reply | 1 comment |
headers
Walt H | 2 Feb 2003 16:06
Picon

Re: Yet Another mkfs.xfs RAID Question]

Christian Guggenberger wrote:
> On Sat, Feb 01, 2003 at 04:20:46PM -0500, Martin K. Petersen wrote:
> 
>>>>>>>"Christian" == Christian Guggenberger <Christian.Guggenberger <at> physik.uni-regensburg.de> writes:
>>
>>Christian> I'm going to set up a new 1.7 TB HW Raid 5 next week.  RAID
>>Christian> 5 contains 11 disks, Stripe Unit should be (according to
>>Christian> the manual) 128k.  so my mkfs.xfs options will be
>>Christian> sunit=256, swidth=2560 for the data section, won't they?
>>
>>Yup.
>>
>>
>>Christian> I will definitly use internal log, so I'd like to ask, if I
>>Christian> should use logversion 2, and what sunit and swidth values I
>>Christian> should use here?  I guess the same as for the data
>>Christian> section???
>>
>>128KB log alignment seems a bit of an overkill.
>>
>>Does your controller state which chunk size it uses internally?  Most
>>controllers use 4-16KB blocks for RAID5.  So try aligning your log to
>>values in that neighbourhood.
>>
> 
> thanks for your quick answer!
> The only Documentation about stripe or chunk size I got from the vendor is,
> to use 32k chunk size for random read/write optimaziation or 128k chunk for
> sequentiell read/write optimaziation... No mention about what the contoller
> does internally!
(Continue reading)

Permalink | Reply | 0 comments |
headers
Stephen Lord | 2 Feb 2003 16:11
Picon
Favicon

Re: Yet Another mkfs.xfs RAID Question

On Sat, 2003-02-01 at 15:20, Martin K. Petersen wrote:
> >>>>> "Christian" == Christian Guggenberger <Christian.Guggenberger <at> physik.uni-regensburg.de> writes:
> 
> Christian> I'm going to set up a new 1.7 TB HW Raid 5 next week.  RAID
> Christian> 5 contains 11 disks, Stripe Unit should be (according to
> Christian> the manual) 128k.  so my mkfs.xfs options will be
> Christian> sunit=256, swidth=2560 for the data section, won't they?
> 
> Yup.
> 
> 
> Christian> I will definitly use internal log, so I'd like to ask, if I
> Christian> should use logversion 2, and what sunit and swidth values I
> Christian> should use here?  I guess the same as for the data
> Christian> section???
> 
> 128KB log alignment seems a bit of an overkill.
> 
> Does your controller state which chunk size it uses internally?  Most
> controllers use 4-16KB blocks for RAID5.  So try aligning your log to
> values in that neighbourhood.

4K is usually enough to fix performance issues, but it may be dependent
on the underlying raid too.

Steve
Permalink | Reply | 2 comments |

Gmane