Setting up nodes with built in hard disk

[Bill Broadley]

On Sun, Jun 16, 2002 at 03:52:59PM +0100, ds10025 at cam.ac.uk wrote:
> Hi
> 
> I'm building a basic Beowulf, with NFS support. Each nodel will have a 
> small 500MB hard disk.
> 
> What basic packages I need to install on each node?

My preference is for something like:

netboot -> initrd
Initrd:
	Is the disk partitioned correctly?
	  If yes continue
	  else partition
	Is the disk partitioned correctly?
      if not skip everything.
	Is /scratch formatted ?
	  if yes continue
	  if not mkfs
    If /scratch was successfully formatted mount
    Is swap setup?
	  if yes continue
      if not mkswap
    swapon <swap partition>
    mount -o ro central_fileserver:/dist/ro /
	mount -r rw central_fileserver:/dist/rw/<nodename> /var
    amd or autofs for homedirs.

That way adding nodes is as simple as adding dhcp entry, and creating
a /var for that node.

If a disk dies, the node still boots (no swap and no scratch of course).

That way I get the benefit of a local disk (high bandwidth for swap and 
/scratch), without having to maintain patches, applications, backups, 
and reinstal when a disk dies.

Works for me.  I did a redhat-6 version, a redhat-7 version, and am
working on a redhat-7.3 version.  I.e. all standard redhat rpm's +
changes to allow for a readonly /.  I currently have 60 diskless nodes
and am working on adding 30 (swap + scratch) more.

-- 
Bill Broadley
Mathematics/Institute of Theoretical Dynamics
UC Davis