Re: Annotations, sets, & servers, and redirects

	From: "R. Martin Roscheisen" <rmr@cs.stanford.edu>

	So, if an annotation set spans more than one server, how does access
	control work then.  Consider the case of adding members to access
	control groups.  

Assuming you want the same access control across the whole annotation set
(which need not be the case, incidently) then the group membership list
could be replicated across all the servers that need it.  Depending on
how big the group is and how many servers there are, this might
be reasonable.  Another alternative is that the authentication could
be done by way of the base server, or yet another server that specializes
in authentication.  The access credentials would be forwarded to the real
annotation server.

	OK, let's design an authorization update protocol.
	All security issues considered ?  Really secure ?  
	Is it worth the overhead in protocol complexity ?  

This distributed group idea is interesting, and no doubt challenging,
but quite independent of annotations, so let's proceed on without it.

	Are there other solutions to the hotspot problem to attain the same
	effect ?  (e.g. replicating sets and splitting up members over sites
	once they subscribe--as a pragmatic one, etc.)

Replication of annotations is essential to scalability for large
numbers of readers.  Replication is almost independent of annotations
too, so that's another protocol we can off-load on some other group. 
But replication is even more important for annotations than general
documents, as I point out in my paper
because annotations are like news postings, and people often want to
find out as soon as one has been posted to a forum they are interested
in.  So notification needs to be tied to replication.