[rhelv6-beta-list] Can RHEL6 installer align the disk partition with 4k-sector or raid stripe size?

Nico Kadel-Garcia nkadel at gmail.com
Sat Jun 12 13:53:16 UTC 2010


On Fri, Jun 11, 2010 at 12:41 PM, Steve Bonneville <sbonnevi at redhat.com> wrote:
>
> Bryan J. Smith wrote:
>> > When I did the install, I *did* have the installer remove all existing
>> > partitions on the disk, so that could have fixed the old partition table.
>> > If I had not done that, it wouldn't surprise me to see /dev/sda1 still
>> > start at LBA 63 (from the previous RHEL 5 installation).
>>
>> Removing the old partition table does _not_ necessarily change the
>> _existing_ geometry.  I'm sure this is done for "safety" with how
>> the system might be presenting the geometry, the same geometry is
>> re-used.  I'd bet on it.  One can force Anaconda to use a geometry
>> though, which would solve the issue.
>>
>> Best yet is to use the disk vendor's tool.  ;)
>>
>> As far as documentation on the matter, these are issues with the PC and
>> other OSes.  These are not Linux details.  The myriad of issues that
>> have come about from endless geometry assumptions/presentations by both
>> PC firmware (let alone BIOS Int13h Disk Services itself), and Windows,
>> is a long history of agony and pain.
>
> Best as I understand it, the new alignment basically does what recent
> versions of Windows does, which is to ignore geometry and use LBAs to
> just align to the sector size, with a +1M offset to the start of the
> first partition.  We don't worry about breaking on cylinder boundaries
> anymore, we just worry about breaking on the physical sector boundary.

This does nothing for virtualization guest images whose "disk" is
presented by the virtualization software and may reside on a local or
network serviced disk image whose underlying hardware is 4096 byte
blockbased. It's the misalignment of the blocks between the guest OS
and the underlying local or network hardware that causes tremendous
overhead, and can apparently cause serious problems for some NetApp
NFS servers. The virtualization guests have considerable difficulty
knowing about this until the virtualization software can present
virtual disks with such configurations.

That would, in fact, be a better fix, but until such disks are more
fully supported, we need to deal with it otherwise. I've finally
gotten '%pre' scripting and some funky logvol settings working for
anaconda to build VMware guests that are 4096 byte aligned, but it's
nasty. I'd love to see it in a configuration option for anaconda for
precisely such systems.




More information about the rhelv6-beta-list mailing list