lilo vs grub

[Alexandre Oliva]

On Oct 21, 2003, Alan Cox <alan at redhat.com> wrote:

>> >> http://www.redhat.com/archives/fedora-test-list/2003-October/msg02399.html
>> 
>> > Lilo boot works with suitable bioses if either drive fails.
>> 
>> Even if the disk geometries are different, and the raid members are in
>> different partition numbers and/or different locations in the entire
>> disk?  If so, how many raid1 replicas can it support?

> It really depends on the BIOS. A lot of more industrial bioses will try
> hda, see it has failed and try hdc. Providing each lilo is mapping the
> raid copy on that disk, not the other copies it works well.

Then I don't see how it could possibly work in the case of installing
the boot loader in a partition, instead of in the MBR, if the replicas
happen to be in different locations within the disk.  Unless it stores
the mappings for all different replicas, and it actually knows which
replica it's booting from.  This is one of the tricky cases I pointed
out in my e-mail above.  If lilo doesn't support this, claiming it
supports raid1 is misleading at best.  I can believe it supports one
particular case of raid1, that happens to be one of the common cases,
but I very much doubt any boot loader can support the other cases I
pointed out, simply because there's only so much code+information you
can store in the boot sector, and to get more info you have to be able
to know where to get it from.

Now let's go back to the industrial bioses you mentioned.  Ok, hda
failed, so the bios tries hdc.  lilo has to know it has to load from
hdc as well, right?  But how does it know which number it should use?
If it reads from disk 0x80, it may get failures for reading from hda,
or it may be getting sectors from the remapped hdc.  Now assume it got
it right, and worked.  But then, I want to replace the broken hda.  So
I insert a new disk there, and voila, can't reboot.

Oh, it's different from that?  The bios won't remap the disks?  Then
how is lilo going to know to boot off hdc (say 0x81) in the first
place?  And if it does, when I remove the broken disk, hdc will become
0x80, and the system will no longer boot.

And this is all assuming the partitions are actually at the same
locations in the disks.

So, what's really the point of being smart about raid1 /boot?  You
really can't tell how the BIOS is going to react, or how you plan on
recovering from the failure of the primary boot disk.  It's always
guessing, and where's guessing, there's room for mistakes.

I'd much rather know how perfectly well what my disaster-recovery
scenario is and be prepared for that.  This means I have to set it up
myself with the assumptions I know I can make.  Anything else is
misleading at best.

IMNSHO :-)

-- 
Alexandre Oliva   Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer                 aoliva@{redhat.com, gcc.gnu.org}
CS PhD student at IC-Unicamp        oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist                Professional serial bug killer