Satya, 

During our time together Michael and I had two further ideas.  I am not
attaching priority tickets to any of these, but I think they are good
ideas.

ps. note that I am cc'ing this to the newly created discussion lists.

Put your WWW pointer at www.coda.cs.cmu.edu.

Peter

1) Coda on other Unix platforms (all of them). HOW?

While implementing a minicache is clearly platform dependent, there is a
catch all solution (with a small performance penalty):

use Bob Baron's ideas: adapt an (existing) free user level nfs server to
accept requests from the kernel and pass them to Venus. This nfs server
can be totally kernel independent, so we could compile a (modified to
accept packets from this server) Venus on all platforms without kernel
work.

Let's call this NFS<->Venus server "rigoletto".

Of course we will probably not get the performance out of this that we get
out of our minicaches, but the advantage of being able to deal with all
platforms in one blow is compelling -- particularly since we can avoid
commercial Unix kernel stuff in this way. 

----------------------------

2) Vice on major Unix platforms (all of them). HOW?

Michael and I are pretty certain the the userlevel iopen, icreate etc.
library will have poor performance (think about a server with 50000 files,
now realise that lookup always does a linear directory search).  

However, we can easily adapt the user level library to be smarter and find
the inode quickly (e.g. build a directory tree).

This allows us to port Vice (independently of Venus) to the major Unix
platforms.  This is very important for large scale installation who might
want to use Coda.


--------------------------

Note that in each case, lwp will still have to be ported, and layering
this on top of an existing thread library could help us deal with
portability here (Posix threads exist more or less on all platforms).