Excerpts from transarc.external.info-afs: 17-Aug-98 Re: Cross-cell
replication James [EMAIL PROTECTED] (1223*)
> > if you have the proper tokens for each cell, you can do a 'vos dump'
> > of the volume from one cell piped to a 'vos restore' of the volume in the
> > other cell. basically, you use 'vos dump' and 'vos restore' to make copies
> > on a per-volume basis. this is one area in which afs wins over dfs,
> since such
> > a simple operation can't be done in dfs.
> > anne
>
> But would that be an incremental update, like a vos release? It looks simple
> to me, as long as your WAN can tolerate the bandwidth surge.
>
> Cross-cell replication has appeared on more than a few wish lists to Transarc,
by the way.
(Caveat: not product-planning-level official.)
Generally, DFS can do this about as well as can AFS. The limitation of
having appropriate tokens for each cell is annoying, but you can (or
should be able to) put cross-cell users on admin lists in DFS.
The big problem is that the access-control information isn't guaranteed
to be meaningful between different cells. Of the owner/group/ACL
information of such files, what should stay the same, and what should
change? Numeric values of the st_uid/st_gid/ACL entries? Print-names?
In DCE there's a complication that any ACL includes a reference to the
security UUID of the cell to which st_uid and st_gid refer, as well as
the security UUID of any inter-cell elements on the ACL; so when there's
an ACL, the ACL (with UID and GID) completely specifies the
access-control information irrespective of what cell the file is in.
When there's no ACL and there's only UID and GID for a DFS file, the
cell qualification is implicit. You can copy the bits, but the cell
qualifier for ACL-less cells changes to the new local cell, and for
ACL'ed files, all files now look like they were created via cross-cell
authentication.
Craig
