>- It's easier to do at time 0 than with a "mature" cell. Retro-fitting
>stinks.
>- I like the MR AFS approach
>- I dislike the DMIG approach and what is currently offered with the
>network backup products.
>- It's best to organize the data and meta-data the same way, with the
>same granularity whether it's on or off line. (Ask yourself why AFS
>scales so well compared to some other distributed systems!)
>
>Last but not least, since it seems that the vendors are not expedient in
>addressing this concern, perhaps an "Open Source" project is warranted?
Well, it seems that right now, MR-AFS is the best canidate; it's
not open source, but what with AFS is? (I know about Arla, but a
server is much harder :-/). I agree that retro-fitting a live cell
stinks, but it is possible to migrate your cell to MR-AFS fairly
easily (we cheated; our MR-AFS data is in another cell, and we
cross-realm authenticate into it).
I think MR-AFS is 3/4ths the way there; if some more people pitch in,
it could be really nitfy.
--Ken
