Can't I share the files from my head node to the slave nodes without making
the compressed root file system image?
let's be clear: the initrd is loaded by the bootloader (disk bios
or pxe) into memory at the same time as the kernel. it's normally
meant to be a transient set of config scripts/tools that loads
modules necessary for building the _real_ mount tree. once that's
done, the root is pivoted and the initrd is umounted (freeing the
memory). this is mainly to permit a distro to include a kernel
which is generic wrt devices, filesystems, etc. this normally
preceded the usual unix init script (/etc/rc.sysinit if sysv-ish).
but the mechanism could also include a dhcp client that builds
a mount tree that includes an nfs mount as its root. in another thread
I was pointing out that this is slightly distasteful because the node
has already dhcp'd once, and it would be nice to avoid throwing that
information away. (/proc/cmdline is the only way I can see to pass
such information through to the initrd or init scripts.)
What I am asking exactly is, If I just mention the directories which I want
to share in the /etc/exports, is it enough?
listing your desired exports there and making sure nfsd et al are running,
sure. if nfs is already running, you need to call exportfs.
(but this assumes that you're either using kernel net autoconf and
nfsroot, or have hacked an initrd to do such.)
you should see (in the nfs server's logs) the client attempting to
mount that export.
_______________________________________________
Beowulf mailing list, Beowulf@beowulf.org
To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf
