> If you do this, then sharing files between cells becomes a
> nightmare.
not necessarily. if the convention were to give all files
canonical names like /afs/foreign/..., and if the local cell
also appeared there, then the scheme would work.
unfortunately, this wouldn't do any good, because folks would
either embed names like /afs/local/... in their utilities and
discourse, and lose the global namespace, or they would employ
/afs/foreign/localcell/... in which case the hack wouldn't do
anyone much good.
hacking at the namespace is not the answer, though. the real
problem is that afs does not cache attributes on the disk, so
/afs/* has to be stat-ed after every reboot. ("has to" is not
the right term, but if you're on a next or a mac or some other
broken system, you have no choice sometimes.) furthermore, the
default value for the in-core attribute cache is way, way too
small, so that even after /afs/* is stat-ed, the results are
likely to be flushed before they're needed again.
incidentally, in the course of hacking afs for disconnected
operation, we "fixed" the former problem. the latter is simply
a matter of changing your /etc/rc (and buying more memeory,
maybe). (although my afsd hung when i requested 10,000 stat
entries ...)
peter
