Bug#614632: Disk partitions not created along cylinder boundaries

[Jaap Winius]

Quoting Andreas Henriksson :


Thanks for your bug report (and sorry for the very late followup).

On Tue, Feb 22, 2011 at 07:51:28PM +0100, Jaap Winius wrote:

Package: netinstall
Version: Debian GNU/Linux 6.0.0 "Squeeze" - Official i386 NETINST Binary-1,
20110205-14:34


A late response indeed: that was almost six and a half years ago! I  
think I was using netinstall to install Debian squeeze, and then  
noticed afterwards that the partitions that had been created did not  
start on cylinder boundaries like I expected. Or, at least on sector  
2048. So, I figured netinstall was not operating fdisk properly. I  
think. It was, after all, a long time ago. :-)


Just close report.

Cheers,

Jaap

Bug#614632: Disk partitions not created along cylinder boundaries

[Andreas Henriksson]

Control: tags -1 + moreinfo

Hello Jaap Winius,

Thanks for your bug report (and sorry for the very late followup).

On Tue, Feb 22, 2011 at 07:51:28PM +0100, Jaap Winius wrote:
> Package: netinstall
> Version: Debian GNU/Linux 6.0.0 "Squeeze" - Official i386 NETINST Binary-1,
> 20110205-14:34
> 
> When a disk is partitioned, the partitions are later found to not end/start
> on cylinder boundaries. For example:
> 
> ~# fdisk -l /dev/sda
> 
> Disk /dev/sda: 1000.2 GB, 1000204886016 bytes
> 255 heads, 63 sectors/track, 121601 cylinders
> Units = cylinders of 16065 * 512 = 8225280 bytes
> Sector size (logical/physical): 512 bytes / 512 bytes
> I/O size (minimum/optimal): 512 bytes / 512 bytes
> Disk identifier: 0x0007d8f1
> 
>Device Boot  Start End  Blocks   Id  System
> /dev/sda1   1 973 7811072   82  Linux swap / Solaris
> Partition 1 does not end on cylinder boundary.
> /dev/sda2   * 973  121602   968949760   fd  Linux raid autodetect
> ~#
[...]

In other words fdisk is reporting about a potential compatibility problem.
I don't see why you think this warning is false. Of course CHS adressing
is deprecated and full compatibility with it might now be very important
in this day and age, but still if fdisk detects an issue why not
report it? If you think CHS-related compatibility issue reporting
should only happen in some kind of pedantic mode, then please feel
free to discuss such a change on the upstream mailing list.

Please note that debian-installer does not normally use fdisk to create
partitions, so it's definitely possible that the other tools has
issues which fdisk detects. (If I'm not mistaken parted is normally
the tool used by debian-installer.)

I'm marking this this bug as 'moreinfo' as I'm not really sure what's
supposed to happen to consider resolved. I also don't really see
any issue here to begin with, so unless someone fills me in I'll
likely end up closing this bug report.

Regards,
Andreas Henriksson

Bug#614632: Disk partitions not created along cylinder boundaries

[Colin Watson]

On Tue, Feb 22, 2011 at 07:51:28PM +0100, Jaap Winius wrote:
 When a disk is partitioned, the partitions are later found to not
 end/start on cylinder boundaries. For example:

This is entirely intentional.  No remotely modern disk requires
partitions to be aligned on cylinder boundaries, and cylinder alignment
is very bad indeed for performance on many modern disks (especially
SSDs, but also others).  It is a relic of the old days.  I would be
extremely surprised if anyone is still running Debian on a disk old
enough for this to matter.

The only reason you might still need cylinder alignment is if you have a
buggy BIOS that gets confused by non-cylinder-aligned partitions, or
sometimes if you're trying to dual-boot with an old version of Windows.
In this case, you can pass partman/alignment=cylinder as a boot
parameter to the installer.  You should not do this unless you have
problems that go beyond error messages from fdisk.

 The above example was actually the result of an earlier amd64
 version of the Debian squeeze netinstaller, downloaded on the 6th of
 November 2010. (The i386 version mentioned at the beginning of this
 report was found to have this same problem yesterday and also
 involved the partitioning a set of 1 TB disks that were configured
 in a RAID1 array. Until then I had not noticed that anything was
 amiss with my earlier installations.)
 
 However, an i386 version downloaded on the 19th of June 2010 does
 not have this problem and I was able to properly complete the
 installation yesterday by starting the installation and partitioning
 the disks with this older version, but then aborting that and
 completing the procedure with the most recent version.

You have not described why cylinder alignment was bad for you, aside
from the bogus error message from fdisk (use the -c option to turn this
off).  Did anything actually go wrong?  Perhaps you were just confused
by fdisk being stuck in the past?

 My work method also involved using all space available on the RAID1
 volume in question for LVM2. However, after doing this I saw that
 something 56.2 bytes of disk space were marked as unusable. This
 was the case with the disk systems that did not eventually turn out
 to end on cylinder boundaries. When it did work properly the other
 day, the left over unusable space was exactly 512 bytes.

Something 56.2 bytes?  That doesn't make sense ...

Yes, there may be a little more unusable space now, but it should be no
more than a megabyte - in other words, a negligibly tiny fraction of the
size of the disk.

Regards,

-- 
Colin Watson   [cjwat...@debian.org]



-- 
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org

Bug#614632: Disk partitions not created along cylinder boundaries

[Jaap Winius]

Quoting Colin Watson cjwat...@debian.org:


This is entirely intentional.  No remotely modern disk requires
partitions to be aligned on cylinder boundaries, and cylinder alignment
is very bad indeed for performance on many modern disks ...


That's good to hear! I was concerned, because fdisk does not show this  
non-problem on any of the older systems that I have running  
currently and I did not want to run into any performance problems  
later on.



You have not described why cylinder alignment was bad for you, aside
from the bogus error message from fdisk (use the -c option to turn this
off).  Did anything actually go wrong?


No, nothing else seemed to be wrong. I was suspected there might be a  
(small) performance penalty, but never actually noticed anything.



Perhaps you were just confused by fdisk being stuck in the past?


That looks to be the case.


Yes, there may be a little more unusable space now, but it should be no
more than a megabyte - in other words, a negligibly tiny fraction of the
size of the disk.


Strange, though, that the version of the partitioner from last June  
should give a different result (on the same disk) than the later ones  
-- a result that fdisk does not complain about.


Oh well. It sounds like it might be helpful if this bug report was  
moved to the util-linux package (fdisk). Can you arrange this?


Thanks very much for your explanation!

Cheers,

Jaap



--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org

Bug#614632: Disk partitions not created along cylinder boundaries

[Colin Watson]

reassign 614632 util-linux
retitle 614632 fdisk: error message about cylinder alignment confuses people 
into thinking the partitioner is wrong
thanks

On Wed, Feb 23, 2011 at 02:54:52PM +0100, Jaap Winius wrote:
 Quoting Colin Watson cjwat...@debian.org:
 This is entirely intentional.  No remotely modern disk requires
 partitions to be aligned on cylinder boundaries, and cylinder alignment
 is very bad indeed for performance on many modern disks ...
 
 That's good to hear! I was concerned, because fdisk does not show
 this non-problem on any of the older systems that I have running
 currently and I did not want to run into any performance problems
 later on.

OK, good.

 Yes, there may be a little more unusable space now, but it should be no
 more than a megabyte - in other words, a negligibly tiny fraction of the
 size of the disk.
 
 Strange, though, that the version of the partitioner from last June
 should give a different result (on the same disk) than the later
 ones -- a result that fdisk does not complain about.

We made this change to the partitioner last March, but depending on what
branch of the Debian installer you were using, you might well have been
using a version last June that defaulted to cylinder alignment.

 Oh well. It sounds like it might be helpful if this bug report was
 moved to the util-linux package (fdisk). Can you arrange this?

Done with this message.  See also the similar bug #602628.

Cheers,

-- 
Colin Watson   [cjwat...@debian.org]



-- 
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org

Bug#614632: Disk partitions not created along cylinder boundaries

[Jaap Winius]

Package: netinstall
Version: Debian GNU/Linux 6.0.0 Squeeze - Official i386 NETINST  
Binary-1, 20110205-14:34


When a disk is partitioned, the partitions are later found to not  
end/start on cylinder boundaries. For example:


~# fdisk -l /dev/sda

Disk /dev/sda: 1000.2 GB, 1000204886016 bytes
255 heads, 63 sectors/track, 121601 cylinders
Units = cylinders of 16065 * 512 = 8225280 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disk identifier: 0x0007d8f1

   Device Boot  Start End  Blocks   Id  System
/dev/sda1   1 973 7811072   82  Linux swap / Solaris
Partition 1 does not end on cylinder boundary.
/dev/sda2   * 973  121602   968949760   fd  Linux raid autodetect
~#

The above example was actually the result of an earlier amd64 version  
of the Debian squeeze netinstaller, downloaded on the 6th of November  
2010. (The i386 version mentioned at the beginning of this report was  
found to have this same problem yesterday and also involved the  
partitioning a set of 1 TB disks that were configured in a RAID1  
array. Until then I had not noticed that anything was amiss with my  
earlier installations.)


However, an i386 version downloaded on the 19th of June 2010 does not  
have this problem and I was able to properly complete the installation  
yesterday by starting the installation and partitioning the disks with  
this older version, but then aborting that and completing the  
procedure with the most recent version.


My work method also involved using all space available on the RAID1  
volume in question for LVM2. However, after doing this I saw that  
something 56.2 bytes of disk space were marked as unusable. This was  
the case with the disk systems that did not eventually turn out to end  
on cylinder boundaries. When it did work properly the other day, the  
left over unusable space was exactly 512 bytes.




--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org