Brian,
I have worked hard on trying to get SI working under native x86_64.
While discovering many issues through trial and error the only progress
I have managed to make was increasing my daily aspirin intake.
If you want some first hand contact with an Opteron box for testing I
would be more than happy to provide you an environment via ssh.
The kernel is very key, in fact getting a x86_64 kernel to boot via PXE
and function has been unsuccessful. The many different compilations will
have problems varying from having no console device available, being
unable to pivot root and issues mounting the ramdisk. A x86_64 kernel
will not function with the 32-bit ramdisk and boel binaries tree. I
managed a very turbulent recompile of the ramdisk and boel binaries tree
as native x86_64 but I still could not get the PXE kernel to boot to the
point where the contents of the ramdisk were relevant.
One of the most frustrating sticking points was booting a 32-bit
kernel/ramdisk and with minor tweaks to the install script being able to
partition, mkfs and rsync the install image to the client. It all worked
rather well but when the 32-bit kernel tried to chroot in to the client
image tree it would fail because bash in the client image is a 64-bit
executable which the 32-bit kernel and chroot process would fail on.
x86_64 64-bit can execute a x86 32-bit binary but not vice-versa.
I tried embedding a 32-bit environment into the client image for the
sole purpose of getting the chroot to run and performing lilo and the
post install network script processing. I could not make it run.
I can provide you ssh to an SI server with a Opteron node attached to a
private network, serial console and a network controlled power switch
for reboot/recovery purposes. Just let me know.
Jeff
On Thu, 2004-02-19 at 08:40, Brian Elliott Finley wrote:
> Ole,
>
> I'm interested in this, and undoubtedly we'll be including official
> support for this architecture in the future, but I don't know that any
> of the development team have access to such hardware. Do we?
>
> Correct me if I'm wrong, but the key here is the kernel. Assuming we
> get an appropriate kernel running, then binaries compiled for an x86
> architecture will work fine, no?
>
> -Brian
>
>
> Thus spake Ole Holm Nielsen ([EMAIL PROTECTED]):
> > I'm grateful for Jeff's feedback, although it's not what I
> > wanted to hear :-( We're really happy with SystemImager
> > because it lets us clone a customized node's disk image.
> > The last time I looked at Kickstart it used RPM
> > packages, leaving a lot of customization to be done
> > afterwards. AutoYast is for SuSE Linux only, right ?
> >
> > If we decide for AMD64 machines (which perform very
> > great on our codes), we'll need really efficient network
> > installation methods for hundreds of nodes. Are there
> > any scaleable alternatives to SystemImager ? I suppose
> > that IBM delivers proprietary solutions for eServer AMD64
> > clusters...
> >
> > /Ole
--
Jeff Johnson <[EMAIL PROTECTED]>
Western Scientific, Inc
http://www.wsm.com
Powerful cluster computing, server and storage solutions.
"Rome did not create a great Empire by holding meetings. They did it by
killing all those who opposed them."
-------------------------------------------------------
SF.Net is sponsored by: Speed Start Your Linux Apps Now.
Build and deploy apps & Web services for Linux with
a free DVD software kit from IBM. Click Now!
http://ads.osdn.com/?ad_id=1356&alloc_id=3438&op=click
_______________________________________________
Sisuite-users mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/sisuite-users
