kickstart install using NFS
Robert G. Brown
rgb at phy.duke.edu
Fri Oct 11 08:09:51 PDT 2002
On Fri, 11 Oct 2002, Henderson, TL Todd wrote:
> I just wanted to say thanks to everyone. I've got a kickstart floppy
> designed that sets up a node in about 5 minutes from reboot to reboot. I
> used dhcp with mac addresses to assign host names and ips. The whole
> install is about 350 meg.
> I guess I'm easily impressed, but this is pretty slick.
> One of these days I'll play around with the PXE boot options and eliminate
> the floppy.
Yeah, and also try the following (customized for your site):
cp /var/phy/xtmp/ks/initrd-everything.img /boot/initrd-everything.img
cp /var/phy/xtmp/ks/vmlinuz-install /boot/vmlinuz-install
grubby --add-kernel=/boot/vmlinuz-install \
--args="ks=http://install.phy.duke.edu/kscgi/kscgi lang= devfs=nomount ramdisk_size=8192" \
--make-default --title="reinstall" --initrd=/boot/initrd-everything.img
where grubby appears to be a part of mkinitrd.
You run this script as root on your system (after putting the initrd's
and install kernels in the right places) and reboot with NO floppy, from
whereever you might be. In five minutes, your system should reboot
itself at the end of a full KS install.
Lovely way to upgrade nodes consistently, recover from a possible client
compromise on a LAN (instead of screwing around looking for trojans,
just reinstall and five minutes later it is up, good as new), or just
impress the heck out of visitors.
So you only really need to do the floppy/ks install ONCE in the lifetime
of the node or its boot disk, whichever is shorter. That is to say,
maybe 70-80% of the time just once until you throw the node out the
window or use it for a doorstop.
PXE just makes it better, sort of, maybe -- skip the floppy even the
first time, but at the expense of having to screw around with PXE, which
is almost not worth it given that floppy drives cost $8-15 in bulk (in a
node purchase) and PXE is likely to add 20 minutes of bios-mucking per
node to the setup, setting up the boot order and reflashing NIC biosen
(which never seem to PXE-work out of the box and which needs the floppy
Hmm, $8 of OPM (other people's money) or 20 minutes or even more(!) of
my time, per node, decisions, decisions. Either way it is pretty easy.
Stuff like this makes WinXX, even the very latest and greatest variants
that will, after a fashion, install from a network image, an absolute
joke in terms of comparative scalability. One of these days Corporate
America is going to twig to the fact that instead of one MCSE per WinXX
server with twenty or thirty WinXX clients, one RHCE or equivalent can
manage a bunch of servers and as many clients as humanly possible,
literally limited by hardware downtime and human user support.
As in easily 100 per sysadmin (we're pushing 125 here, but our sysadmins
are starting to show signs of wear and tear because hardware failures
ARE happening almost daily with over 250 systems 0-4 years old on desks
and in racks throughout the building, a couple dozen printers, and well
over 100 users).
BTW, you do have to watch out for bios settings that require you and/or
a keyboard/monitor to be present at boot time. YMMV.
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
Robert G. Brown http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567 Fax: 919-660-2525 email:rgb at phy.duke.edu
More information about the Beowulf