* Erik Dykema <ed2019@columbia.edu> [2004-03-03 19:35]:
| Hi-
| I think it would be very useful to support installing onto NFS
| volumes.
| At my site we operate these big netapp NFS file servers (filers).
| They have no intelligence except to serve NFS volumes, i.e. there is no
| real 'os' on them.
| If we could support installing TO an NFS root, we could use diskless
| workstations & servers that would install to their own volume on the
| filers. Then, the filer takes care of stuff like redundancy.
| This might also help for a situation where the NFS master server is
| one architecture (i386) and the machine you want to install 'to' is
| different, i.e. powerpc / mainframe.
| Also, I think that it would be very useful if DI supported
| installing FROM an NFS volume. Using an NFS volume is more flexible for
| testers than the initrd.gz files. If the kernel supported NFS root, could
| just mount it's root directory & init & etc, it would be possible to tinker
| with a running installer system, add drivers as needed, etc. without
| having to re-make the initrd.gz file, which is not easy.
ok. I know what you mean.
Bringing nfs support to d-i shouldn't be very hard. But you must create
your own installation medium, I guess.
First you must modify the existing d-i kernel to support nfs filesystems
(in client mode). Next you must create udebs for the portmap package and
the nfs client tools. Maybe you need also a small udeb, which ask the
user for the nfs mounting parameters (server, port, etc.). And then you
must mount the nfs volume into /target. (this udeb must provide
mounted-partitions). After this d-i should be able to install to an nfs
volume.
Create the udebs, and d-i will support it :-)