Hello,
I don't understand why you wold want gcc running on your diskless
nodes ... is this to be a compiler farm?
Any way, I think that my setup will interest you. I have my diskless
compute nodes boot from a root+boot floppy disk. The root filesystem
is a ramdisk and BProc (http://bproc.sourceforge.net/) is used to
migrate processes from the master node (which has gcc lam/mpi and all
other software needed installed) to these compute nodes. And, to
provide a common filesystem between all nodes, I'm using PVFS
(http://www.parl.clemson.edu/pvfs/).
The diskless nodes are very minimalistic; it contains ony the kernel,
Busybox (http://www.busybox.net/) and a few userland programs needed
to bring PVFS and BProc up. The majority of the programs that I wish
to run on the cluster are staticly linked ... but for the few that
aren't, the libs are copied over to the ramdisk from the pvfs
directory at boot time. For this reason, I'm using an 8MB ramdisk
(the root filesystem takes about 2MB of this).
BProc is a very nice tool since the programs that you wish to run on
the cluster don't actually need to exist on the remote nodes. And
lam/mpi can integrate with it also ... therefore lam is only installed
on the master node.
Hope this gave you an idea of what you may want to do.
../kaveh
|
