>> Sure, afsd could search for all the AFSDB records in the net and load
>> them into the kernel. (I think that some folks at UMich actually had an
>> afsd that worked this way.) But it still requires CellServDB to know
>> (statically) the names of all the AFS cells that might potentially be
>> examined.
A utility program which iterated over root.afs, pulled out all the
cells, and constructed a CellServDB based on this would be very, very
useful, and avoids most of the problems people have mentioned. Of
course, it does require that afs be running, but you can use fs
newcell to stuff info about new cells into the kernel after the cache
manager starts. If you have DNS problems, you use the version from
the prior (reboot|cron job|whatever), and wait for the next time to
try again.
Anyway, that's mostly brainstorming, but I think you could come up
with something useful based on this.
Marc
