I'm fresh back from LPAR class, and am setting up my first partitioned
system. I have my vio server up and am starting to set up virtual
disks for my partitions to use. I plan on using logical volumes, so I
can carve my 73GB drives into smaller disks for my partitions to boot
from. In doing this, I ran across a bit in my class manual that says:

"If a logical volume is exported as a virtual device, all physical
volumes that make up the volume group in which the logical volume is
contained must be attached to the same adapter".

Now, I can do this (just barely - these adapters only have 3 disk
slots each!), but I wonder why. I had planned on using LVM mirroring
for these logical volumes, but putting them on a single adapter
introduces a single point of failure.

Can anyone explain why this limitation exists, and if it has perhaps
been lifted in some more recent release of the vioserver or the
hypervisor? I'm running vioserver version 1.4.1, and my host system
is a 561 at firmware level 322 (which just came out a few weeks ago).
I plan on running AIX 5.3 06 01 in my partitions.

rodak <robert.macnutt@nghs.com> wrote:
> I'm fresh back from LPAR class, and am setting up my first partitioned
> system. I have my vio server up and am starting to set up virtual
> disks for my partitions to use. I plan on using logical volumes, so I
> can carve my 73GB drives into smaller disks for my partitions to boot
> from. In doing this, I ran across a bit in my class manual that says:

> "If a logical volume is exported as a virtual device, all physical
> volumes that make up the volume group in which the logical volume is
> contained must be attached to the same adapter".

> Now, I can do this (just barely - these adapters only have 3 disk
> slots each!), but I wonder why. I had planned on using LVM mirroring
> for these logical volumes, but putting them on a single adapter
> introduces a single point of failure.

> Can anyone explain why this limitation exists, and if it has perhaps
> been lifted in some more recent release of the vioserver or the
> hypervisor? I'm running vioserver version 1.4.1, and my host system
> is a 561 at firmware level 322 (which just came out a few weeks ago).
> I plan on running AIX 5.3 06 01 in my partitions.

The way it was explained to me is that VIO client writes do not go
through the LVM layer on the VIO server. Furthermore, as I understand it,
the entire client LV needs to be contained on a single PV on the VIO server.
If you are looking to avoid a single point of failure, create a second VIO
server, allocate another LV from it to the client and use LVM mirroring at
the client level.

Regards,
Gregory Young

On Oct 16, 7:56 pm, Gregory Young <use...@outland.cc> wrote:
> rodak <robert.macn...@nghs.com> wrote:
> > I'm fresh back from LPAR class, and am setting up my first partitioned
> > system. I have my vio server up and am starting to set up virtual
> > disks for my partitions to use. I plan on using logical volumes, so I
> > can carve my 73GB drives into smaller disks for my partitions to boot
> > from. In doing this, I ran across a bit in my class manual that says:
> > "If a logical volume is exported as a virtual device, all physical
> > volumes that make up the volume group in which the logical volume is
> > contained must be attached to the same adapter".
> > Now, I can do this (just barely - these adapters only have 3 disk
> > slots each!), but I wonder why. I had planned on using LVM mirroring
> > for these logical volumes, but putting them on a single adapter
> > introduces a single point of failure.
> > Can anyone explain why this limitation exists, and if it has perhaps
> > been lifted in some more recent release of the vioserver or the
> > hypervisor? I'm running vioserver version 1.4.1, and my host system
> > is a 561 at firmware level 322 (which just came out a few weeks ago).
> > I plan on running AIX 5.3 06 01 in my partitions.
>
> The way it was explained to me is that VIO client writes do not go
> through the LVM layer on the VIO server. Furthermore, as I understand it,
> the entire client LV needs to be contained on a single PV on the VIO server.
> If you are looking to avoid a single point of failure, create a second VIO
> server, allocate another LV from it to the client and use LVM mirroring at
> the client level.
>
> Regards,
> Gregory Young

Thanks, Gregory. Yes, you're right, two VIO servers does make a lot
more sense. I was considering that anyway. About the LV being
contained within a single PV - do you have any reference where I can
verify that? If it's a limitation that has perhaps been removed with
more recent software, I'd like to know, since I may have the need to
create a virtual LV that's larger than a single PV, and it would be
simpler to create it as a single LV.

On Oct 17, 9:33 am, rodak <robert.macn...@nghs.com> wrote:
> On Oct 16, 7:56 pm, Gregory Young <use...@outland.cc> wrote:
>
>
>
> > rodak <robert.macn...@nghs.com> wrote:
> > > I'm fresh back from LPAR class, and am setting up my first partitioned
> > > system. I have my vio server up and am starting to set up virtual
> > > disks for my partitions to use. I plan on using logical volumes, so I
> > > can carve my 73GB drives into smaller disks for my partitions to boot
> > > from. In doing this, I ran across a bit in my class manual that says:
> > > "If a logical volume is exported as a virtual device, all physical
> > > volumes that make up the volume group in which the logical volume is
> > > contained must be attached to the same adapter".
> > > Now, I can do this (just barely - these adapters only have 3 disk
> > > slots each!), but I wonder why. I had planned on using LVM mirroring
> > > for these logical volumes, but putting them on a single adapter
> > > introduces a single point of failure.
> > > Can anyone explain why this limitation exists, and if it has perhaps
> > > been lifted in some more recent release of the vioserver or the
> > > hypervisor? I'm running vioserver version 1.4.1, and my host system
> > > is a 561 at firmware level 322 (which just came out a few weeks ago).
> > > I plan on running AIX 5.3 06 01 in my partitions.
>
> > The way it was explained to me is that VIO client writes do not go
> > through the LVM layer on the VIO server. Furthermore, as I understand it,
> > the entire client LV needs to be contained on a single PV on the VIO server.
> > If you are looking to avoid a single point of failure, create a second VIO
> > server, allocate another LV from it to the client and use LVM mirroring at
> > the client level.
>
> > Regards,
> > Gregory Young
>
> Thanks, Gregory. Yes, you're right, two VIO servers does make a lot
> more sense. I was considering that anyway. About the LV being
> contained within a single PV - do you have any reference where I can
> verify that? If it's a limitation that has perhaps been removed with
> more recent software, I'd like to know, since I may have the need to
> create a virtual LV that's larger than a single PV, and it would be
> simpler to create it as a single LV.

No it's per design, you are stuck with that limitation. The reason is
this:
- on a VIO you create a VG with disks from different adapters
- you move an adapter from one LPAR to the other
- you just broke your VG

So to avoid this, they limit us! Solution: 2 VIO, one adapter each,
mirror on client.

Nic