Hamish Moffatt wrote: > I've been playing with this recently; I have a whole lot of motherboards > and net cards which I want to use to get some parallel processing happening. > > I suggest the NFS-Root and NFS-Root-Client mini-howtos. What I have > ended up doing is creating a tree with just enough of /bin, /sbin/, > /etc, /var and /tmp to boot up in it, then the client will mount > the server's /bin, /sbin, /home and /usr over the top. I have about 3.5mb > per client and I am hoping to thin this out some more. (Unfortunately, > the startup scripts require bash, and bash + libreadline + curses etc > are huge.) This way I have effectively got a Debian system remotely, > although it can't be used for package installation/deinstallation etc > of course because /usr is readonly, and /var and /etc are machine-specific.
An alternative is to use the nametrans patch for the kernel (I think this has actually been integrated into the 2.1.x kernel). This lets you nfs mount server:/ on the clients at their root directory. The few files (/etc/hostname, /etc/init.d/network, etc) that need to be different for each computer are given special names, like /etc/init.d/network#HOST=foo#, and the client named foo will see that file as /etc/init.d/network. This simplifies administration a lot - you only have to install everything once and it's visible to all your computers. It also uses only a few extra K per client. -- see shy jo -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
