[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Diskless nodes: the bootdisk



> But the slaves nodes was set up with harddisks; and I was needing to do
> its run through of floppies, when every node would have a floppy with
> its kernel image inside of the floppy

This is a somewhat awkward mode of operation... Floppies are flaky.

> (I cannot use dhcp/tftp to get the kernel images from the server,
> because the departament that this cluster have been build it's already
> belong of a internal network, with it's particular settings(...), ),

This is a misunderstanding. You can have a private network inside another
private network, no problem. You can have your cluster server serving as
the gateway between your private network and the Department's one. It can
run dhcpd for your nodes, broadcasting only in you internal network, and
serve the kernel via tftp, there is no problema with any of that:


          your private net   +-----------+   dept's private net
                             |your server|
 +------+                    |           |                  +------+
 | node |                    |           |                  | dept |
 |      |    192.168.10.*    |card   card|   192.168.0.*    |      | Internet
 |      |--------------------| 1       2 |------------------| gate |--->
 |      |    network mask    |           |   network mask   |      |
 +------+    255.255.255.0   | (gateway) |   255.255.255.0  +------+
                             |           |
                             +-----------+

You can use the other address ranges as well (which I don't remember now).

> so the file system would needed to be mounted on the server, through nfs
> (the ready-only directories), and in ram (ramdisks) the read-write
> directories..

You must have NFS filesystems served by the server in any case. In general
ram disks are used only for booting. You probably want to use your RAM for
your applications. You can have your r/w fs's in the server via NFS too.

> Does anybody have some suggestions about how to do this? The /etc/fstab
> file? I have seen some shell scripts,

The NFS root requires special kernel configuration to mount the root. You
will need to fix the fstab too. Note that people are tending to migrate to
initrd's and that all this may become obsolete in the future.

> but one thing that has made me confuse it's if the directory /etc need
> to be the same of the server..

No it does not. In fact, if should'n. To have a look at some somewhat
outdated documentation in Portuguese on how to do all this, look at:

latt.if.usp.br:/pmc
							Cheers,

----------------------------------------------------------------
        Jorge L. deLyra,  Associate Professor of Physics
            The University of Sao Paulo,  IFUSP-DFMA
       For more information: finger delyra@latt.if.usp.br
----------------------------------------------------------------



Reply to: