The "stateless" image I refer to is what is actually booted on the compute 
node containing the image to be captured.  It's called stateless because it is 
loaded completely in RAM and does not maintain any state when a reboot occurs.

The partimage binary is part of this stateless image and actually runs on the 
compute node.  It does not run on the management node.  The management node 
does not have block level access to the disk on the compute node to be able to 
capture the image from the disk.

I'll try to describe the process a little better.  The management node issues 
a reboot command to the compute node.  The compute node uses PXE to load and 
boot a kernel (vmlinuz), initial RAM disk (initrd.img), and a root filesystem 
(rootimg.gz) from the management node.  All three of these together make up 
the stateless image.  Once the compute node is booted with the stateless 
image, it uses NFS to mount some things from the management node, and then 
runs some xcat postscripts, one of which is the partimageng postscript.  This 
postscript determines what partitions are on the compute node and, depending 
on how the postscript is configured, uses partimage or partimageng to capture 
an image of the compute node disk that is then saved to the management node.  
When it is finished capturing the image, it notifies xcat on the management 
node and then reboots.  xcat reconfigures itself to tell the compute node to 
boot off of disk at next boot.  When the compute node comes up, it uses PXE to 
ask the management node how to boot.  The management node tells it to boot off 
of disk.

I hope that clarifies how the system works.  If any of it is unclear, please 
ask for further clarification.


On Wednesday June 01, 2011, Sunil Venkatesh wrote:
> Josh,
> I had one more clarification.
> partimage binaries run in the management node to capture an (stateless)
> image from the compute node right? In that case, is there a need for
> these binaries to go into the rootimg.gz??
> My assumption is, partimage runs on the management node (an intel blade
> in our case) to capture a stateless image from a compute node (a power 7
> blade) and stores these images under " /install " of the management
> node. Please correct me if I am wrong here.
> Regards,
> Sunil
> On 6/1/11 9:58 AM, Josh Thompson wrote:
> > Hash: SHA1
> > 
> > On Tuesday May 31, 2011, Sunil Venkatesh wrote:
> >> Hi,
> >> 
> >> I used the steps that were mentioned under
> >> 
> >>
> >> mag e+and+partimage- ng+to+xCAT+2.x+%28unofficial%29
> >> 
> >> to enable partimage support for xcat. I wasn't sure if I need to change
> >> references to x86&  x86_64 (as directories) to reflect the ppc
> >> architecture, as the web page says "The architecture for the node must
> >> always be set to x86 for this..". I have with me the vmlinuz (kernel
> >> image) and initrd for the capture process. The 2 nodeset commands
> > 
> > By this, do you mean you have vmlinuz and initrd for your power blades,
> > not the ones linked to off of the page you listed above?  If you do,
> > that's a good start.  However, you'll also need rootimg.gz.  rootimg.gz
> > is the root filesystem for the stateless image.  It also contains the
> > partimage and partimageng binaries.  Assuming partimage or partimageng
> > can actually capture partitions from power systems, you'll need to
> > compile at least one of them to run on power.  For the rootimg.gz image
> > I provided, I compiled them statically so that I didn't have to worry
> > about including any library dependencies in rootimg.gz.
> > 
> > It would be a good idea to research how to use xcat's genimage command to
> > generate stateless images to learn how to do this.
> > 
> > If there's any part of the above that you don't fully understand, please
> > ask me to clarify it.  Until you have a stateless image that you can
> > deploy to your power blades, there's no point in trying to debug any VCL
> > specific items.
> > 
> > Josh
> > - --
> > - -------------------------------
> > Josh Thompson
> > VCL Developer
> > North Carolina State University
> > 
> > my GPG/PGP key can be found at
> > Version: GnuPG v2.0.17 (GNU/Linux)
> > 
> > PZYAniP2Eam7nxgiDWUnp5sKPYPO4OMa
> > =exBV
> > -----END PGP SIGNATURE-----
Josh Thompson
Systems Programmer
Advanced Computing | VCL Developer
North Carolina State University

my GPG/PGP key can be found at

All electronic mail messages in connection with State business which
are sent to or received by this account are subject to the NC Public
Records Law and may be disclosed to third parties.

Attachment: signature.asc
Description: This is a digitally signed message part.

Reply via email to