Newcastle Connection/Distributed UNIX Systems

[rbw at rti-sel.UUCP]

We at the Research Triangle Institute have ported  the  New-
castle  Connection  to  a Gould Concept 32/8750 running 4.1c
BSD, a VAX 11/750 running 4.2 BSD, and a SUN 100  also  run-
ning 4.2 BSD connected via Ethernet.

Using the Newcastle Connection software, it is possible, for
example,  for  one  to use files and peripheral devices on a
remote machine, or to request remote execution of a program,
all  transparently.   It  is  very handy to 'cd' amongst the
various machines as  one  is  used  to  doing  on  a  single
machine.   And,  'rsh',  'rcp',  'rtar', 'rdump', ... become
unnecessary.

The hardware and software differences  between  these  three
machines  caused  several  expected and unexpected problems.
The byte ordering differences were addressed by the  Newcas-
tle  Connection  (but  had to be modified to work properly).
The networking code (provided by Ray  Essick  for  4.1a  but
modified  by  the  Newcastle distributors) had to be updated
for 4.1c and 4.2.  System calls not present in V7  UNIX  had
to  be  added  and tested.  The new 4.2 signal mechanism and
symbolic links caused some problems.  More  difficult  prob-
lems  were due to the necessity of interpreting binary files
(i.e.   directories,  utmp,  etc.)   between   heterogeneous
machines.   For example, the file(I) command does not under-
stand the magic numbers of  files  on  an  arbitrary  remote
machine.  We were able to make directory access work, but in
general this is a difficult aspect of distributed  computing
in a non-homogeneous environment.

We would like to get in touch with people  who  use  distri-
buted  UNIX  file  systems  (and/or  computation) to discuss
these and other problems, and their experiences.   What  are
the  strengths/weaknesses  of  other  systems (NETIX, Purdue
IBIS, Bell's V8, etc.).  Are there features  that  you  have
found you cannot live without? What misfeatures have you run
into?   What   are   your   plans   for   distributed   file
systems/computing in the future?


    Bob Warren  (mcnc!rti-sel!rbw)

    Research Triangle Institute
    PO Box 12194
    Herbert Building
    Research Triangle Park, N. C.   27709