Distributed systems

Andy said earlier that he'd like to see distributed systems
implemented in user space. I agree, because I feel that it will take
quite a lot of trial and error to hit upon a scheme than works well in
all cases, and flexibility will be important (and VSTa allows servers
to be killed or replaced at runtime). Once things are working well,
performance could be analysed to see whether a kernel implementation
is needed or not.

I imagined a similar architecture to Andy outlined, and from what I
can see, the biggest problem will probably be in the naming scheme
used for message forwarding. If a file-based naming system is used
(arguably a good idea, even for distributed object systems like DOE),
then hacking namer, and then writing a multiplexing server to run
above the TCP/IP or rs232c server would seem to be enough to get
something running. The only problem with such a system would be that
every system would need to be running a local copy the multiplexing
server so they can understand where the messages should be sent...
Received on Sun Feb 27 17:24:07 1994