>>>>> Manuel Bouyer writes:

Manuel> (and the 'proto=udp' amd option is not obvious).

Actually, that option is extremely non-obvious, since you have to put
it in at least two places to be absolutely sure to get UDP for all
mounts. You need it in each amd mount map (to get UDP for remote NFS
mounts), but you also need it in the overall amd.conf file. Without
the latter amd's top-level mount point will be using TCP (amd's
"simulated" NFS server, that triggers the "real" mount). If that uses
TCP a machine that we have will crash every day. After adding:

map_options = opts:=proto=udp

to /etc/amd.conf the machine has been up for over a week (we already
had proto=udp in all the mount maps).

Note that this can not be the "server goes down" problem, there must
be some other problem with TCP mounts too. So I strongly support
Manuel's proposal to switch back amd to UDP.

        -- Urban