swapping over NFS

David Monro (davidm@cs.su.oz.au)
Sun, 7 Apr 1996 16:42:37 +1000 (EST)

Having played a bit with a diskless setup (using an eprom and the NFSroot
option) I have discovered that the main obstacle to a really useful diskless
CPU server is the inability to swap over nfs. Is anybody working on this? How
hard would it be?

The only other obstacles I can see to creating a setup where multiple sytems
share the same root filesystem is a modified version of mount that doesn't use
/etc/mtab but uses /proc/mounts, and could understand some fancy substitutions
in the fstab, eg

server:/usr/export/var.@HOSTNAME@ /var nfs defaults 0 0
server:/usr/export/swapfiles/swap.@HOSTNAME@ none swap defaults 0 0

would mount the appropriate things off the server without having a different
fstab for different clients. This however I can do myself. (I don't believe it
is possible to share /var, though I may be wrong. /var/log/wtmp at least must
persist over reboots (so it can't be on a memfs, unlike stuff in /var/run and
/var/lock) and must be unique to each machine).

Also, a memfs for /tmp on diskless machines would be nice - there is one
around, I believe. Does anybody know where it hides out and how stable it is?