Tux

...making Linux just a little more fun!

Linux Software RAID- comments/suggestions

Amit Kumar Saha [amitsaha.in at gmail.com]


Thu, 20 Mar 2008 14:17:01 +0530

Hello,

I started playing with Software RAID sometime back and have posted my notes here at http://amitksaha.blogspot.com/2008/03/software-raid-on-linux-part-1.html

Would be nice to have some feedback/comments.

Ben: Can i contribute to LG as an article?

Thanks, Amit

-- 
Amit Kumar Saha
*NetBeans Community
Docs Coordinator*
Writer, Programmer, Researcher
http://amitsaha.in.googlepages.com
http://amitksaha.blogspot.com


Top    Back


Ben Okopnik [ben at linuxgazette.net]


Thu, 20 Mar 2008 13:37:05 -0400

On Thu, Mar 20, 2008 at 02:17:01PM +0530, Amit Kumar Saha wrote:

> Hello,
> 
> I started playing with Software RAID sometime back and have posted my
> notes here at http://amitksaha.blogspot.com/2008/03/software-raid-on-linux-part-1.html
> 
> Would be nice to have some feedback/comments.
> 
> Ben: Can i contribute to LG as an article?

To make my last note a little clearer: go ahead and *format it* per LG standards and send it to me, and I'll be happy to check it out. Please see http://linuxgazette.net/faq/author.html for details.

-- 
* Ben Okopnik * Editor-in-Chief, Linux Gazette * http://LinuxGazette.NET *


Top    Back


Amit Kumar Saha [amitsaha.in at gmail.com]


Thu, 20 Mar 2008 23:16:39 +0530

On 3/20/08, Ben Okopnik <[email protected]> wrote:

> On Thu, Mar 20, 2008 at 02:17:01PM +0530, Amit Kumar Saha wrote:
>
> > Hello,
>  >
>  > I started playing with Software RAID sometime back and have posted my
>  > notes here at http://amitksaha.blogspot.com/2008/03/software-raid-on-linux-part-1.html
>  >
>  > Would be nice to have some feedback/comments.
>  >
>  > Ben: Can i contribute to LG as an article?
>
>
> To make my last note a little clearer: go ahead and *format it* per LG
>  standards and send it to me, and I'll be happy to check it out. Please
>  see http://linuxgazette.net/faq/author.html for details.

No rephrasing required? I am thinking of adding some introductory para to it.

-- 
Amit Kumar Saha
*NetBeans Community
Docs Coordinator*
Writer, Programmer, Researcher
http://amitsaha.in.googlepages.com
http://amitksaha.blogspot.com


Top    Back


Ben Okopnik [ben at linuxgazette.net]


Thu, 20 Mar 2008 14:17:02 -0400

On Thu, Mar 20, 2008 at 11:16:39PM +0530, Amit Kumar Saha wrote:

> On 3/20/08, Ben Okopnik <[email protected]> wrote:
> > On Thu, Mar 20, 2008 at 02:17:01PM +0530, Amit Kumar Saha wrote:
> >
> > > Hello,
> >  >
> >  > I started playing with Software RAID sometime back and have posted my
> >  > notes here at http://amitksaha.blogspot.com/2008/03/software-raid-on-linux-part-1.html
> >  >
> >  > Would be nice to have some feedback/comments.
> >  >
> >  > Ben: Can i contribute to LG as an article?
> >
> >
> > To make my last note a little clearer: go ahead and *format it* per LG
> >  standards and send it to me, and I'll be happy to check it out. Please
> >  see http://linuxgazette.net/faq/author.html for details.
> 
> No rephrasing required? I am thinking of adding some introductory para to it.

Hey, you're the author - you get to figure that part out. I'm just some guy who edits stuff. :)

-- 
* Ben Okopnik * Editor-in-Chief, Linux Gazette * http://LinuxGazette.NET *


Top    Back


Amit Kumar Saha [amitsaha.in at gmail.com]


Fri, 21 Mar 2008 10:43:43 +0530

> Hey, you're the author - you get to figure that part out. I'm just some
>  guy who edits stuff. :)

Please find attached the article.

Hope its good!

Thanks, Amit


Top    Back


Rick Moen [rick at linuxmafia.com]


Thu, 20 Mar 2008 22:51:28 -0700

Quoting Amit Kumar Saha ([email protected]):

> Please find attached the article.

Amit, ideally, article submissions should go to "[email protected]", and not broadcast to this mailing list's entire membership. Thanks.


Top    Back


Kapil Hari Paranjape [kapil at imsc.res.in]


Fri, 21 Mar 2008 14:31:37 +0530

On Thu, 20 Mar 2008, Rick Moen wrote:

> Quoting Amit Kumar Saha ([email protected]):
> 
> > Please find attached the article.
> 
> Amit, ideally, article submissions should go to
> "[email protected]", and not broadcast to this mailing list's
> entire membership.  Thanks.

But since he did here are some (unfortunately mostly negative) criticisms:

	1. The article seems to say that installing "mdadm"
	is the same as installing kernel support for RAID. This is
	not true. Ideally, if you find "/proc/mdstat" does not exist
	you should first check whether the appropriate modules exist
	(under /lib/modules/$(uname -r)/kernel/drivers/md/).
 
	2. The article suggests using two consecutive partitions on
	a single disk under raid level0 which seems to be rather
	pointless unless there are some tests which one wants to run.
 
	3. The mdadm approach is one of the different ways to
	support raid under linux (some others being lvm and evms).
	The article fails to mention these others and the differences.
 
	4. Without a clear indication of the tests to be run, there
	is not much in the article which one cannot get from "man mdadm".
 
	5. Why use "Software RAID"? This question should at least be
	addressed in the article.

So this is a good start to an article but is well short of the real thing.

Regards,

Kapil. --


Top    Back


Ben Okopnik [ben at linuxgazette.net]


Fri, 21 Mar 2008 12:14:09 -0400

On Fri, Mar 21, 2008 at 02:31:37PM +0530, Kapil Hari Paranjape wrote:

> On Thu, 20 Mar 2008, Rick Moen wrote:
> > Quoting Amit Kumar Saha ([email protected]):
> > 
> > > Please find attached the article.
> > 
> > Amit, ideally, article submissions should go to
> > "[email protected]", and not broadcast to this mailing list's
> > entire membership.  Thanks.
> 
> But since he did here are some (unfortunately mostly negative)
> criticisms:

[snip]

Y'know, we're always looking for more tech editors to do exactly this kind of thing. If you ever feel like contributing in that respect, just let me know; as Shakespeare has Henry V say to Montjoy, "I know thy quality."

-- 
* Ben Okopnik * Editor-in-Chief, Linux Gazette * http://LinuxGazette.NET *


Top    Back


Amit Kumar Saha [amitsaha.in at gmail.com]


Fri, 21 Mar 2008 22:03:47 +0530

On 3/21/08, Kapil Hari Paranjape <[email protected]> wrote:

>
> But since he did here are some (unfortunately mostly negative)
>  criticisms:
>
>         1. The article seems to say that installing "mdadm"
>         is the same as installing kernel support for RAID. This is
>         not true. Ideally, if you find "/proc/mdstat" does not exist
>         you should first check whether the appropriate modules exist
>         (under /lib/modules/$(uname -r)/kernel/drivers/md/).

Thanks; I am teaching myself- so seeing it work that way, I thought, may be installing 'mdadm' inserts the kernel module as well.

This actually inserts the 'md' support kernel module:

$ sudo modprobe md-mod
Password:
>
>         2. The article suggests using two consecutive partitions on
>         a single disk under raid level0 which seems to be rather
>         pointless unless there are some tests which one wants to run.

Fair. But, I am just experimenting and learning and hopefully my experiments will help me conclude some best practices.

>
>         3. The mdadm approach is one of the different ways to
>         support raid under linux (some others being lvm and evms).
>         The article fails to mention these others and the differences.

Agree. Like the traditional 'raidtools'- which is very nicely described in the "Software RAID How-to" which is referred.

>
>         4. Without a clear indication of the tests to be run, there
>         is not much in the article which one cannot get from "man mdadm".

Its Ben's decision- I am open for rejection as well :-)

>
>         5. Why use "Software RAID"? This question should at least be
>         addressed in the article.

It does- gives a single reason in the first few lines

I appreciate the feedback and welcome more :-)

Thanks, Amit

-- 
Amit Kumar Saha
*NetBeans Community
Docs Coordinator*
Writer, Programmer, Researcher
http://amitsaha.in.googlepages.com
http://amitksaha.blogspot.com


Top    Back


René Pfeiffer [lynx at luchs.at]


Fri, 21 Mar 2008 18:13:46 +0100

On Mar 21, 2008 at 2203 +0530, Amit Kumar Saha appeared and said:

> On 3/21/08, Kapil Hari Paranjape <[email protected]> wrote:
> > But since he did here are some (unfortunately mostly negative)
> >  criticisms:
> >
> >         1. The article seems to say that installing "mdadm"
> >         is the same as installing kernel support for RAID. This is
> >         not true. Ideally, if you find "/proc/mdstat" does not exist
> >         you should first check whether the appropriate modules exist
> >         (under /lib/modules/$(uname -r)/kernel/drivers/md/).
>
> Thanks; I am teaching myself- so seeing it work that way, I thought,
> may be installing 'mdadm' inserts the kernel module as well.
>
> This actually inserts the 'md' support kernel module:
>
> $ sudo modprobe md-mod
> Password:

Yes, but you might want to check the full MD subsystem. It has its own modules for RAID0/1/4/5/6 that also need to be present.=20

lynx@miranda:~$ grep MD /boot/config-2.6.24.3
[...]
CONFIG_MD=3Dy
CONFIG_BLK_DEV_MD=3Dm
CONFIG_MD_LINEAR=3Dm
CONFIG_MD_RAID0=3Dm
CONFIG_MD_RAID1=3Dm
CONFIG_MD_RAID10=3Dm
CONFIG_MD_RAID456=3Dm
CONFIG_MD_RAID5_RESHAPE=3Dy
CONFIG_MD_MULTIPATH=3Dm
CONFIG_MD_FAULTY=3Dm
[...]
lynx@miranda:~$=20

Grab a kernel tree, run "make menuconfig" and browse the MD section. The descriptions carry some interesting information.

> >         2. The article suggests using two consecutive partitions on
> >         a single disk under raid level0 which seems to be rather
> >         pointless unless there are some tests which one wants to run.
>
> Fair. But, I am just experimenting and learning and hopefully my
> experiments will help me conclude some best practices.

That's good news, and articles can have more that one part. :)

> >         3. The mdadm approach is one of the different ways to
> >         support raid under linux (some others being lvm and evms).
> >         The article fails to mention these others and the differences.
>
> Agree. Like the traditional 'raidtools'- which is very nicely
> described in the "Software RAID How-to" which is referred.

Make sure that you meantion how traditional raidtools is. The mdadm tools work a lot better. Also make sure to mention how to mark the RAID partitions as autodetect (by writing 0xfd in the partition type).

> >         4. Without a clear indication of the tests to be run, there
> >         is not much in the article which one cannot get from "man mdadm".
>
> Its Ben's decision- I am open for rejection as well :-)

But you can influence Ben's decision by adding content. ;)

> >         5. Why use "Software RAID"? This question should at least be
> >         addressed in the article.
>
> It does- gives a single reason in the first few lines

Well, there are some more reasons for and against. From the top of my head I can tell you about the following things:

 - Software RAID is independent of proprietary management software,
   maintaining a software RAID work the same way on all machines that
   run Linux. This can be an important advantage when your staff is
   poorly trained and the management console to your proprietary RAID
   system is written by equally poorly trained staff.
 
 - When considering software RAID think of the performance. All RAID
   algorithms are done by the system CPU and every block has to be
   copied over the system's data bus (i.e. sda1 <-> IO controller <-> RAM
   possibly CPU <-> IO controller <-> sdb1).=20

Happy RAIDing, :) René.


Top    Back


Rick Moen [rick at linuxmafia.com]


Fri, 21 Mar 2008 11:54:12 -0700

Quoting Amit Kumar Saha ([email protected]):

> On 3/21/08, Kapil Hari Paranjape <[email protected]> wrote:
> >         2. The article suggests using two consecutive partitions on
> >         a single disk under raid level0 which seems to be rather
> >         pointless unless there are some tests which one wants to run.
> 
> Fair. But, I am just experimenting and learning and hopefully my
> experiments will help me conclude some best practices.

Doing spanning (RAID0) or mirroring (RAID1) within a single drive spindle is an excellent way to thrash that physical drive to death while enduring miserable performance -- though most people would say that's something to be avoided, not deliberately sought.

> >         3. The mdadm approach is one of the different ways to
> >         support raid under linux (some others being lvm and evms).
> >         The article fails to mention these others and the differences.
> 
> Agree. Like the traditional 'raidtools'- which is very nicely
> described in the "Software RAID How-to" which is referred.

My understanding is that the traditional raidtools are now considered obsolete, in favour of mdadm -- and that, at least to that degree, the Software RAID HOWTO is likewise out of date. Outdated write-ups are one of the hazards one encounters in this area.

> It's Ben's decision- I am open for rejection as well :-)

But, you know, you could just send in a revision.


Top    Back


Ben Okopnik [ben at linuxgazette.net]


Fri, 21 Mar 2008 15:40:55 -0400

On Fri, Mar 21, 2008 at 06:13:46PM +0100, Ren? Pfeiffer wrote:

> On Mar 21, 2008 at 2203 +0530, Amit Kumar Saha appeared and said:
> > 
> > Its Ben's decision- I am open for rejection as well :-)
> 
> But you can influence Ben's decision by adding content. ;)

That's really, really good advice that Ren? is giving you, Amit. Rather than going through submission/rejection/rewriting/resubmission times N, why not consider improving it as much as you can, now that you've got a couple of good new perspectives? This would make your article better, as well as increasing your knowledge on the topic.

Frankly, I'd say that your "mistake" in submitting the article here in TAG turned out to be an excellent thing - live tech editing from TAG is a huge benefit. Take advantage of it while it's here; this special, one-time deal is subject to availability and may end soon to the incredible demand! [1] If you order today, you get not only the thing itself, but a set of these highly polished, indestructible, guaranteed- for-life - say, have you seem my demonstration with the tomato? Oh, never mind.

[1] Where the hell's Ron Popeil when you need him? I can't do this bit worth a damn.

-- 
* Ben Okopnik * Editor-in-Chief, Linux Gazette * http://LinuxGazette.NET *


Top    Back


Amit Kumar Saha [amitsaha.in at gmail.com]


Sat, 22 Mar 2008 09:31:37 +0530

On 3/21/08, Kapil Hari Paranjape <[email protected]> wrote:

>  [snip]..
>
>         1. The article seems to say that installing "mdadm"
>         is the same as installing kernel support for RAID. This is
>         not true. Ideally, if you find "/proc/mdstat" does not exist
>         you should first check whether the appropriate modules exist
>         (under /lib/modules/$(uname -r)/kernel/drivers/md/).

What if the modules are not present? Will installing 'mdadm' build the Kernel module as well and then insert it?

[snip]..

Thanks, Amit


Top    Back


Kapil Hari Paranjape [kapil at imsc.res.in]


Sat, 22 Mar 2008 09:54:55 +0530

Hello,

On Sat, 22 Mar 2008, Amit Kumar Saha wrote:

> On 3/21/08, Kapil Hari Paranjape <[email protected]> wrote:
> >  [snip]..
> >
> >         1. The article seems to say that installing "mdadm"
> >         is the same as installing kernel support for RAID. This is
> >         not true. Ideally, if you find "/proc/mdstat" does not exist
> >         you should first check whether the appropriate modules exist
> >         (under /lib/modules/$(uname -r)/kernel/drivers/md/).
> 
> What if the modules are not present? Will installing 'mdadm' build the
> Kernel module as well and then insert it?

No. Your article assumes that you have a "stock" Debian kernel installed. For such a kernel, most features are built as modules and can be loaded dynamically as required.

If the user has compiled their own kernel where they did not compile in support for "md" either statically or as modules, then the installation of "mdadm" will work but the commands will not work.

Regards,

Kapil. --


Top    Back


Amit Kumar Saha [amitsaha.in at gmail.com]


Sat, 22 Mar 2008 09:57:15 +0530

On 3/22/08, Kapil Hari Paranjape <[email protected]> wrote:

> Hello,
>
>
>  On Sat, 22 Mar 2008, Amit Kumar Saha wrote:
>  > On 3/21/08, Kapil Hari Paranjape <[email protected]> wrote:
>  > >  [snip]..
>  > >
>  > >         1. The article seems to say that installing "mdadm"
>  > >         is the same as installing kernel support for RAID. This is
>  > >         not true. Ideally, if you find "/proc/mdstat" does not exist
>  > >         you should first check whether the appropriate modules exist
>  > >         (under /lib/modules/$(uname -r)/kernel/drivers/md/).
>  >
>  > What if the modules are not present? Will installing 'mdadm' build the
>  > Kernel module as well and then insert it?
>
>
> No. Your article assumes that you have a "stock" Debian kernel
>  installed. For such a kernel, most features are built as modules and
>  can be loaded dynamically as required.
>
>  If the user has compiled their own kernel where they did not compile
>  in support for "md" either statically or as modules, then the
>  installation of "mdadm" will work but the commands will not work.

Thanks for the re-confirmation. I was wondering that will be the answer to my query :-)

-- 
Amit Kumar Saha
*NetBeans Community
Docs Coordinator*
Writer, Programmer, Researcher
http://amitsaha.in.googlepages.com
http://amitksaha.blogspot.com


Top    Back


Amit Kumar Saha [amitsaha.in at gmail.com]


Sat, 22 Mar 2008 10:49:38 +0530

On 3/22/08, Ben Okopnik <[email protected]> wrote:

> On Fri, Mar 21, 2008 at 06:13:46PM +0100, Ren? Pfeiffer wrote:
>  > On Mar 21, 2008 at 2203 +0530, Amit Kumar Saha appeared and said:
>  > >
>
> > > Its Ben's decision- I am open for rejection as well :-)
>  >
>  > But you can influence Ben's decision by adding content. ;)
>
>
> That's really, really good advice that Ren? is giving you, Amit.
>  Rather than going through submission/rejection/rewriting/resubmission
>  times N, why not consider improving it as much as you can, now that
>  you've got a couple of good new perspectives? This would make your
>  article better, as well as increasing your knowledge on the topic.
>
>  Frankly, I'd say that your "mistake" in submitting the article here in
>  TAG turned out to be an excellent thing - live tech editing from TAG is
>  a huge benefit. Take advantage of it while it's here; this special,
>  one-time deal is subject to availability and may end soon to the
>  incredible demand! [1] If you order today, you get not only the thing
>  itself, but a set of these highly polished, indestructible, guaranteed-
>  for-life - say, have you seem my demonstration with the tomato? Oh,
>  never mind.
>
Not all mistakes necessarily lead to all negatives- its proved once again here!

Thanks guys!

-- 
Amit Kumar Saha
*NetBeans Community
Docs Coordinator*
Writer, Programmer, Researcher
http://amitsaha.in.googlepages.com
http://amitksaha.blogspot.com


Top    Back