best/easiest tool for diskless COW.
Robert G. Brown
rgb at phy.duke.edu
Wed Jun 20 09:40:09 EDT 2001
On Wed, 20 Jun 2001, Nelson Castillo wrote:
> I would like to know which tool is the bes for setting up a C.O.W.,
> and also the easiest to use. I use RedHat 7.1 and PVM.
> I know this is a C.O.W. and not a beowulf, this is a first step.
RH 7.1 plus the PVM and MPI distros, maybe MOSIX, maybe Condor,
installed via kickstart -- sounds good to me.
It's hard to beat kickstart for ease of use: Set up a dhcp record for
the host to be installed, point it at a boilerplate kickstart file that
you develop for your nodes/workstations (which can just about be e.g.
the Gnome Workstation default plus a few things like PVM), create a
kickstart boot floppy (one that goes straight into kickstart instead of
giving you an interactive menu) and pop the flop, boot, pop the flop,
boot, etc... Depending on the server bandwidth, you can easily
parallelize enough installs to complete a host every 2-5 minutes.
The only hassle I've encountered with this is that to give the booting
hosts fixed IP numbers you have to boot them once to get their ethernet
number and then boot them again once you've set up the dhcpd record that
maps the MAC address to the IP number. One could probably write a tool
or add a "cluster" feature to dhcpd to BOTH assign names and numbers in
the order they are received from a block AND rewrite the dhcpd.conf file
with the assignments, but AFAIK this still doesn't exist. With perl it
(or something automagical) wouldn't be too hard -- boot all the hosts
once with the ks floppy, run the script on /var/log/messages to extract
all the request MAC addresses, unique them against /etc/dhcpd.conf,
generate a boilerplate /etc/dhcpd.conf entry for each new MAC address
(with IP numbers and names assigned out of a block e.g. b01, b02 with IP
# 192.168.1.1, 192.168.1.2...), restart dhpcd, and reboot all the hosts
to be installed. Right now I tend to do the latter few steps by hand
because we don't install enough new hosts at any one time to payback the
time required to write and debug the script. Lazy me.
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
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
More information about the Beowulf