[linux-lvm] Configuring boot partition with LVM

[linux-lvm] Configuring boot partition with LVM

[Jim Cain]

All,

Has anyone been successful in creating a completely LVM-managed system?
Specifically, I'm wondering about the problem with needing LVM-enabled
kernel on a boot floppy for installation purposes.

Or is this not a good idea? Is it better to have a non-LVM root partition
on a separate drive, and then use LVM for everything else?

Thanks for the input,
Jim

Re: [linux-lvm] Configuring boot partition with LVM

[Fabian.Herschel]

On Fri, Feb 04, 2000 at 12:29:34PM -0500, Jim Cain wrote:
> All,
> 
> Has anyone been successful in creating a completely LVM-managed system?
> Specifically, I'm wondering about the problem with needing LVM-enabled
> kernel on a boot floppy for installation purposes.

At this point of time lilo is not ready to boot a kernel fom a lv. Lilo
should also learn how to boot a kernel fom a striped volume...

Greetings
Fabian



> 
> Or is this not a good idea? Is it better to have a non-LVM root partition
> on a separate drive, and then use LVM for everything else?
> 
> Thanks for the input,
> Jim

-- 
-----------------------------------------------------------------------
Fabian Herschel                                 SuSE Rhein/Main AG     
Tel:    +49-6196-50951-23                       Mergenthalerallee 45-47
Fax:    +49-6196-409-607                        D-65760 Eschborn
mailto: Fabian.Herschel@suse.de                 http://www.suse.de      

Re: [linux-lvm] Configuring boot partition with LVM

[Luca Berra]

On Mon, Feb 07, 2000 at 10:37:19AM +0100, Fabian.Herschel@suse.de wrote:
> At this point of time lilo is not ready to boot a kernel fom a lv. Lilo
actually you can, if you play tricks with your partition table :)
but being real, you will need to move vgscan into kernel space.

> should also learn how to boot a kernel fom a striped volume...
this seems impossible to me, unless you add a boot area
to the VG structure.
lilo is a multi stage boot process
a small 512-byte loader is placed in the mbr
this loader then loads lilo itself from the disk
now if you want a striped vg for /boot
you must also find somewere to store the second stage boot loader.
and it must be possible to load it from a 512-byte program.

L.

-- 
Luca Berra -- bluca@comedia.it
    Communication Media & Services S.r.l.

Re: [linux-lvm] Configuring boot partition with LVM

[Jos Visser]

>  Lilo
> should also learn how to boot a kernel fom a striped volume...

Hi,

I don't think you fully appreciate the request you're making here.
Commercial Unices such as HP-UX and Solaris can not boot from striped
volumes for about the same reasons why it would be difficult to have
LILO read from a striped volume: the implications are just too
horrendous to comtemplate.

Typically I don't care that my boot partition/lv can not be striped. The
I/O rates into that volume are just not high enough to need striping.
Why not put our efforts to something else which is more important right
now.?

++Jos

-- 
All the rivers run into the sea; yet the sea is not full; unto
the place from whence the rivers come, thither they return again.

Re: [linux-lvm] Configuring boot partition with LVM

[Fabian.Herschel]

Hi,

On Mon, Feb 07, 2000 at 05:37:50PM +0100, Jos Visser wrote:
> >  Lilo
> > should also learn how to boot a kernel fom a striped volume...
> 
> Hi,
> 
> I don't think you fully appreciate the request you're making here.

The question (not my answer!!) was why is it complicate to boot fom a
lv. Ans i anserwed that lilo cant do that at this point of time!
I also pointed to an advanced problem (the striping) so it should
be clear thats not only a thing of a "simple" patch in lilo. 

My itention was to show WHY theere are a lot of problems for getting
linux to boot from "any" lv.

> Commercial Unices such as HP-UX and Solaris can not boot from striped
> volumes for about the same reasons why it would be difficult to have
> LILO read from a striped volume: the implications are just too
> horrendous to comtemplate.
> 
> Typically I don't care that my boot partition/lv can not be striped. The

Me to. A small /boot does not need to be striped. In the near time it
also need to be lv controlled...

> I/O rates into that volume are just not high enough to need striping.
> Why not put our efforts to something else which is more important right
> now.?

Yes, but we should also explain (or show) restrictions to our collegs on 
this list! 

Greetings
Fabian

-- 
-----------------------------------------------------------------------
Fabian Herschel                                 SuSE Rhein/Main AG     
Tel:    +49-6196-50951-23                       Mergenthalerallee 45-47
Fax:    +49-6196-409-607                        D-65760 Eschborn
mailto: Fabian.Herschel@suse.de                 http://www.suse.de      

Re: [linux-lvm] Configuring boot partition with LVM

[Jos Visser]

In HP-UX, your boot lvol (/stand) must be an HFS, and it must *not* be
striped. It may be mirrored, but to boot from the mirrored lvol you must
create a complete secondary boot disks (using pvcreate -B and then after
having created the mirrors running lvlnboot).

In my opinion, mirroring should be made a part of the LVM because then
the LVM will be in complete control of the mirrors, which allows other
nice things such as lvsplit/lvmerge and others.

++Jos

Denis JARRIL wrote:
> 
> I don't agree.
> With HP-UX, you can boot from anywhere. You just have to create a contiguous
> and HFS LV to boot on. I don't know where the problem comes from on Linux
> but I tried to make a root miror and it didn't work. The test of the md
> volumes was good but the system stopped just after, when it passes from the
> ramdisk to the disk. Bad thing. If this problem is resolved, LVM on Linux
> will be able to work.
> 
> ----- Message d'origine -----
> De : Jos Visser <josv@osp.nl>
> � : <Fabian.Herschel@suse.de>
> Cc : <linux-lvm@msede.com>
> Envoy� : lundi 7 f�vrier 2000 17:37
> Objet : Re: [linux-lvm] Configuring boot partition with LVM
> 
> >  Lilo
> > should also learn how to boot a kernel fom a striped volume...
> 
> Hi,
> 
> I don't think you fully appreciate the request you're making here.
> Commercial Unices such as HP-UX and Solaris can not boot from striped
> volumes for about the same reasons why it would be difficult to have
> LILO read from a striped volume: the implications are just too
> horrendous to comtemplate.
> 
> Typically I don't care that my boot partition/lv can not be striped. The
> I/O rates into that volume are just not high enough to need striping.
> Why not put our efforts to something else which is more important right
> now.?
> 
> ++Jos
> 
> --
> All the rivers run into the sea; yet the sea is not full; unto
> the place from whence the rivers come, thither they return again.

-- 
All the rivers run into the sea; yet the sea is not full; unto
the place from whence the rivers come, thither they return again.

Re: [linux-lvm] Configuring boot partition with LVM

[Andreas Dilger]

Christopher Shaulis writes:
> I can't remember exactly how they explained it, but IBM has a good
> reason for not allowing mirrored dump devices. I think the idea is that
> in order to catch the system dumping in the LVM code itself, you have to
> write the dump without the aid of the LVM. I think the way that works is
> that it precomputes a list of offset:extent lists for the device when it
> is brought online.

I think that the issue isn't when the dump is written to a mirrored LV
(as the dump will only go to one or the other copy of the LV), but when
it is read back there is no way to determine which copy you will read
from (I/O load balancing causes reads to come from either PV), so you
may get intermittent garbage inside the dump...

Cheers, Andreas
-- 
Andreas Dilger  \ "If a man ate a pound of pasta and a pound of antipasto,
                 \  would they cancel out, leaving him still hungry?"
http://www-mddsp.enel.ucalgary.ca/People/adilger/               -- Dogbert

Re: [linux-lvm] Configuring boot partition with LVM

[Luca Berra]

On Thu, Feb 10, 2000 at 12:14:04AM -0700, Andreas Dilger wrote:
> Christopher Shaulis writes:
> > I can't remember exactly how they explained it, but IBM has a good
> > reason for not allowing mirrored dump devices. I think the idea is that
> > in order to catch the system dumping in the LVM code itself, you have to
> > write the dump without the aid of the LVM. I think the way that works is
> > that it precomputes a list of offset:extent lists for the device when it
> > is brought online.

Dunno 'bout IBM, with HP you can setup dump to a mirrored LV, but it
will dump only on the first PV, this is due to dumping being handled
by the firmware, which knows nothing about LVM

dump LVs must be contiguous on disk, but you can have many of them

> I think that the issue isn't when the dump is written to a mirrored LV
> (as the dump will only go to one or the other copy of the LV), but when
> it is read back there is no way to determine which copy you will read
> from (I/O load balancing causes reads to come from either PV), so you
> may get intermittent garbage inside the dump...
This is one issue, another nice pitfall is that dump is usually read
from disk after mirror reconstruction has begun (remember, if it crashed
mirror copies aren't synchronized), so you may not exactly found what you expect
in the dump :)

With HP you must specify a couple of options in lv creation to disable
mirror reconstruction.

L.

-- 
Luca Berra -- bluca@comedia.it
    Communication Media & Services S.r.l.