Although we are still in the just-getting-started stage with regard to AFS, we
are looking a little ahead to when:
- most of our users will still be NFS-only
- some will be experimenting with or productively using AFS
- DFS will be available on some of our machines.
Our question of the day is: Before we try "selling" AFS to a lot of people who
are already familiar and happy with NFS, how reasonable is it for us to assert
that such a conversion is worthwhile despite the looming presence of DFS? In
particular, what can we tell system managers and users about the extent of
expected AFS/DFS happy co-existence at an individual site? (At my site, I
think that AFS/DFS co-existence is an interesting issue, but AFS as "training"
for DFS is not.)
I have already seen a partial (unclear) answer to our question (and I am
hoping for something more definitive). What I have seen so far is a "Transarc
AFS Statement of Direction", which appears as section 7.3 of an AFS HEPnet
Evaluation (a collaborative effort by National HEPnet Management and Fermi
National Accelerator Laboratory, dated June 1992).
In the cited Statement of Direction there is some discussion of Transarc's
AFS 3 to DCE DFS migration strategy. This strategy (it is said) "will
assist sites in making the transition to DCE, while still maintaining AFS 3
machines. It will offer options for communication between the two file systems,
both within a cell and between cells."
The above quote seems to suggest that an individual cell will be able to
house both AFS and DFS machines, and that users of one file system will be
able to access files from the other file system either within their own cell
or in remote cells.
However, the next paragraph of the Statement of Direction discusses a tool that
Transarc expects to offer for conversion of AFS 3 formatted files to DFS files.
With file conversion needed, I would have to guess that "communication between
the two file systems, both within a cell and between cells" would be something
less than I have inferred in the preceding paragraph.
So, how extensive will the happy co-existence be? Will individual cells
house both types of files? Will file servers be strictly one or the other?
Will an individual client machine be able to access both AFS and DFS
servers? Will it be possible for some users on a client machine to use
AFS commands while others use DFS? (Note: I'm not suggesting that AFS/DFS
communication "must" include all of these conveniences. I'm just trying to
get information that will allow us to guage how much we ought to push our
users toward AFS at this time.)
