I see a number of problems with DCE (apart from the fact that I won't have
cycles for quite a while to do such a huge project). Perhaps most of all
"who has really time to do it?".  

- Coda has and needs sophisticated bandwith adaptation.  
- Coda can use multicast MultiRPC to propagate files and other data to
replication servers
- Our bulk data transfer uses pretty nice side effects.  A huge amount of
coding would be needed to translate this into DCE stuff. 

How are we going to get this from DCE?  It may be there, but I don't know
about it. I did go to the bookstore and got some stuff (I had a pretty
neat book on my shelves too). 

Writing some packing routines is really not a problem (and we probably
have them anyway).  Similarly the cell issues are not problematic at the
network layer (and we should definitely learn from the DCE design, no
question about that) but mostly at the filesystem layer. What do we cache,
how do we efficiently propate extra data to the kernel (CellId's etc),
how do we translate user id's to Coda cell uids.  All of these would need
implementing anyway, and let's first do that.

Also, I worry that DCE is not "lightweight". We like "smallish" things
here and I guess we are bad offenders ourselves already. We don't want to
get worse.

The positive side of the coin is that we are working hard to separate our
subsystems more clearly.  This should make it a lot easier for DCE, or
pthreads or something else to be used later on.   We are thinking about
things like this from time to time, but dramatically short of man-power.
Also I find Coda pretty complex, and until recently I didn't perhaps
appreciate how difficult it would be to do such a project and end up with
a quality file system. 

I wish we had a full DCE installation here, I am sure I could learn a lot
about it that way. 

I am sure there is more to say about this...

- Peter -