Why is sb->size set to 0 with raid0?

Why is sb->size set to 0 with raid0?

[Mario 'BitKoenig' Holbe]

Hello,

while examining a raid0 v1.2 superblock I found "Used Dev Size" seems
not to be set there, mdadm -E on a component device shows:
	          Magic : a92b4efc
	        Version : 1.2
	    Feature Map : 0x0
	...
	 Avail Dev Size : 2930274824 (1397.26 GiB 1500.30 GB)
	  Used Dev Size : 0
	    Data Offset : 16 sectors
	   Super Offset : 8 sectors
	          State : clean

Is this intentional?
This is just to let you know and for my curiosity. I don't know whether
this has negative side-effects or not. I found none until now.

Don't get me wrong, I like the (IMHO positive) side-effect of this: it
makes totally sense to me, not to reserve bitmap-space for raid0 - I
even downgraded to mdadm 3.1.1 for raid creation to get rid of the 1M
alignment, and I was short thinking about patching super1.c for this
array to get rid of the bitmap-space reservation as well.
Thus, I was just positively surprised to see that in fact there *is* no
bitmap-space reserved for raid0, wondered how and where this was
managed, and tracked it down to sb->size being set to 0.


regards
   Mario
-- 
You know, people think mathematics is complicated. Mathematics is the simple
bit. Its the stuff we can understand. Its cats that are complicated. I mean,
what is it in those little molecules and stuff that make one cat behave
differently than another, or that make a cat? And how do you define a cat?
I have no idea.                                 -- John Conway

Re: Why is sb->size set to 0 with raid0?

[Neil Brown]

On Wed, 21 Jul 2010 18:16:46 +0200
Mario 'BitKoenig' Holbe <Mario.Holbe@TU-Ilmenau.DE> wrote:

> Hello,
> 
> while examining a raid0 v1.2 superblock I found "Used Dev Size" seems
> not to be set there, mdadm -E on a component device shows:
> 	          Magic : a92b4efc
> 	        Version : 1.2
> 	    Feature Map : 0x0
> 	...
> 	 Avail Dev Size : 2930274824 (1397.26 GiB 1500.30 GB)
> 	  Used Dev Size : 0
> 	    Data Offset : 16 sectors
> 	   Super Offset : 8 sectors
> 	          State : clean
> 
> Is this intentional?
> This is just to let you know and for my curiosity. I don't know whether
> this has negative side-effects or not. I found none until now.
> 
> Don't get me wrong, I like the (IMHO positive) side-effect of this: it
> makes totally sense to me, not to reserve bitmap-space for raid0 - I
> even downgraded to mdadm 3.1.1 for raid creation to get rid of the 1M
> alignment, and I was short thinking about patching super1.c for this
> array to get rid of the bitmap-space reservation as well.
> Thus, I was just positively surprised to see that in fact there *is* no
> bitmap-space reserved for raid0, wondered how and where this was
> managed, and tracked it down to sb->size being set to 0.
> 

The 'Used Dev Size' is the amount of each device that is used in the array.
For RAID0 (and linear), there is no such number.  The devices can be of
different sizes and in each case "all" of the available space is used.
So in this case '0' means 'all'.  I should probably just remove the number
as it doesn't mean anything.

This has nothing to do with the bitmap.  Space reserved for the bitmap is
out side of the 'Avail Dev Size'.

NeilBrown

Re: Why is sb->size set to 0 with raid0?

[Mario 'BitKoenig' Holbe]

Neil Brown <neilb@suse.de> wrote:
> The 'Used Dev Size' is the amount of each device that is used in the array.
> For RAID0 (and linear), there is no such number.  The devices can be of
> different sizes and in each case "all" of the available space is used.

Hmm, RAID0 has a chunk size - but I guess this is what you meant with
the quotes around "all" :)

> This has nothing to do with the bitmap.  Space reserved for the bitmap is
> out side of the 'Avail Dev Size'.

Sure, it has: having a zero 'Used Dev Size' (i.e. sb->size = 0) results
in not reserving space for a bitmap - at least for v1 superblocks:

		array_size = __le64_to_cpu(sb->size);
		/* work out how much space we left for a bitmap */
		bm_space = choose_bm_space(array_size);

... which I like in case of RAID0 (as I said), but of which I don't know
if you do like it as well ;) - for example, because it also prevents a
bitmap when the RAID0 is later converted to RAID5.


regards
   Mario
-- 
Jene, die grundlegende Freiheit aufgeben wuerden, um eine geringe
voruebergehende Sicherheit zu erwerben, verdienen weder Freiheit noch
Sicherheit.
                                     -- Benjamin Franklin (1706-1790)

Re: Why is sb->size set to 0 with raid0?

[Roman Mamedov]

[-- Attachment #1: Type: text/plain, Size: 465 bytes --]

On Thu, 22 Jul 2010 15:34:56 +1000
Neil Brown <neilb@suse.de> wrote:

> For RAID0 (and linear), there is no such number.  The devices can be of
> different sizes and in each case "all" of the available space is used.

That sounds a bit surprising, excuse my ignorance, but how a RAID0 of
differently-sized devices is laid out? Is the "tail" of the larger device
simply used in linear mode, adding space to the striped "heads"?

-- 
With respect,
Roman

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 198 bytes --]

Re: Why is sb->size set to 0 with raid0?

[Neil Brown]

On Thu, 22 Jul 2010 18:50:54 +0600
Roman Mamedov <roman@rm.pp.ru> wrote:

> On Thu, 22 Jul 2010 15:34:56 +1000
> Neil Brown <neilb@suse.de> wrote:
> 
> > For RAID0 (and linear), there is no such number.  The devices can be of
> > different sizes and in each case "all" of the available space is used.
> 
> That sounds a bit surprising, excuse my ignorance, but how a RAID0 of
> differently-sized devices is laid out? Is the "tail" of the larger device
> simply used in linear mode, adding space to the striped "heads"?
> 

% man md

search for "RAID0"

read and enjoy.

NeilBrown

Re: Why is sb->size set to 0 with raid0?

[Mario 'BitKoenig' Holbe]

Roman Mamedov <roman@rm.pp.ru> wrote:
> That sounds a bit surprising, excuse my ignorance, but how a RAID0 of
> differently-sized devices is laid out? Is the "tail" of the larger device

rtfm :)

md(4):
   RAID0
...
       If  devices  in  the array are not all the same size, then once
       the smallest device has been exhausted, the RAID0 driver starts
       collecting  chunks  into  smaller  stripes  that  only span the
       drives which still have remaining space.


regards
   Mario
-- 
Und wie jede Sprache, so hat auch PHP ein Anwendungsgebiet, fuer das es
besonders gut geeignet ist. Soweit ich sehen kann, handelt es sich dabei
um das sogenannte "ins Knie schiessen".
                              -- Volker Birk in de.alt.sysadmin.recovery