[GE users] Building SGE 6.0 on a Warewulf cluster
john.hearns at clustervision.com
Wed Sep 29 23:51:41 BST 2004
[ The following text is in the "iso-8859-1" character set. ]
[ Your display is set for the "ISO-8859-10" character set. ]
[ Some special characters may be displayed incorrectly. ]
On Wed, 2004-09-29 at 23:43, Jeffrey B. Layton wrote:
> I'm going to build SGE 6.0 on a Warewulf (warewulf-cluster.org)
> cluster. Warewulf uses PXE or Etherboot to boot the nodes and
> send an image which is installed on a Ram drive on the nodes. The
> ram drive is pretty small (about 40 Megs at max), so I'd like to
> keep it that way. I'm still working through the SGE build instructions,
> but does anyone want to offer some friendly advice for building
> Oh, I forgot to mention that I can build SGE for Warewulf in a couple
> of ways. I can build the complete setup and install it on the head node
> and then build the compute node image from the head node but I need
> to know exactly what I need to include on the compute nodes. The
> second way is to build SGE on the head node and then build it and
> include it in the image for the compute nodes. I'd like to stick to the
> first method if possible. I'll just need to figure out what the minimum
> components are for the compute nodes and include those in the image.
All you need to have on the nodes is the /etc/init.d/sgeexecd script
All the binaries etc. etc. etc. live under $SGE_ROOT.
Once you have done an inst_sge on the master you will find the sgeexecd
script in $SGE_ROOT/default/common
Put the sgeexecd in your slave image, and fix up the runlevels
(there is a buglet in the Redhat chkconfig which doesn't recognise
the LSB init script headers, so be aware of an link in rc.X which looks
Edit /etc/services also to get the port set
$SGE_ROOT is then NFS-mounted from teh master node.
Say /usr/local/sge (your choice!)
I'm moving to using local spooling on slave nodes,
but for your diskless nodes just go with spooling to the $SGE_ROOT
Oh, and following on from an earlier thread make sure the sgeadmin user
is in the password file/NIS map/LDAP... on the slave nodes.
And you probably don't need to rebuild (unless you want to!)
Won't the courtesy binaries work for you?
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net
More information about the gridengine-users