In any nontrivial network of Suns where there is a lot of NFS
cross-mounting, when one Sun goes down, the others often
freeze up. Some machine tries to access the down one, and
(getting no response) repeats indefinitely. This causes it to
appear dead to some messages (what is actually happening is
that it is locked up in what should have been a brief
excursion to a higher
spl level). Then another machine
tries to reach either the down machine or the pseudo-down
machine, and itself becomes pseudo-down. The first machine to
discover the down one is now trying both to access the down
one and to respond to the pseudo-down one, so it is even
harder to reach. This situation snowballs very quickly, and
soon the entire network of machines is frozen - worst of
all, the user can't even abort the file access that started
the problem!
Many of NFS's problems are excused by partisans as being an
inevitable result of its
statelessness, which is held to be
a great feature (critics, of course, call it a great
of
Unix's alleged bogosity; ITS had a working NFS-like
shared file system with none of these problems in the early