Tim Bates <[EMAIL PROTECTED]> writes: > The reason generally is that /usr can be put on another system entirely > and remotely mounted, because as you said it can be read only. That way > a whole cluster of machines can use precisely the same /usr directory, > which makes systems administration somewhat easier. The top-level /bin > and similar directories are for those files that are necessary even if > the network-mounted /usr is unavailable. If all your binaries were in > /bin and that was mounted from a remote machine, and that remote machine > was down, the local machine would be entirely unusable. So having both > is a tradeoff between having the essential files locally, whilst having > the bulk of them able to be provided remotely.
You have a good point. It is much easier to bring a system back up if it has a local root filesystem when /usr is mounted over network. Thanks for clarifying the issue to me :) The thing is, there is no /usr in GNU (Hurd that is). But Hurd is special because it has a much larger potential for special filesystems that can bring in files from different sources and present them in a common namespace. For example, a special filesystem could be used instead of stow to have software packages in their own directories while still being able to access them from common bin, lib, sbin etc. directories. But I think there's also a unionfs on *BSD that can be used to union multiple filesystems to one. /Pelle
