.so macs
.TL "System Dependencies"

  Throughout the design of this project I have tried to make it as independent
as possible upon the transport mechanism and the version of \*(UX used.
The primitives in \fItrans.c\fP depend heavily upon using virtual circuit
transport service ports, and the fake system calls correspond to those of
4.2BSD \*(UX. The fake system calls for a different version of \*(UX, and an
implementation of \fItrans.c\fP using, say, the ``sockets'' of 4.2 BSD \*(UX
should be interchangeable.

  A program written for 4.2 could remotely be executed on a machine running
4.1 using the 4.2 fake system calls. (A library of compatibility functions
would also have to be used to map unreplaced 4.2 functions into their 4.1
equivalents.)

  It is also quite feasible to implement a RFS server in an operating system
other that \*(UX. It would have to present a \*(UX system call interface to the
alien system's filestore.

.TL Conclusion

  As a topic, Remote Procedure Calls and transparent networking are not
thoroughly explored. The subject of this project is a good one. The goals are
simply and clearly defined but there are many problems to be confronted in
producing a satisfactory solution. The surprisingly small amount of software I
produced to implement a fully working, if simple remote execution facility is
the tip of a mountain of thought.

  The solutions to difficult issues such as signal handling and performing
forks and execs transparently have manifested themselves in the basic design of
the system. For this reason it has been difficult in this report to justify
some decisions as they were mentioned.

  But those decisions, I think, were good. The remotely executing versions of
standard, unmodified system programs such as \fBecho\fP, \fBprintenv\fP,
\fBcat\fP and \fBls\fP behave identically to their local counterparts. I
implemented only the system calls necessary for these test programs as I found
them to be necessary; there are several dozen calls to be replaced for a fuller
system for which implementation will be a simple job of imitation. Only a few
such as \fBdup2\fP will require a little thought.

  I am confident that I have produced a system whose generality and flexibility
will make it easy to extend, and which will indeed enable large programs such
as \fBtroff\fP to be executed remotely. The generality is such that I see no
reason why a program should not undergo loading with both RFS and REX to
produce a program which both accesses files on many machines and executes
remotely!

.so concpic.t

.ce
rcopy file1@host3 file2@host4