IPC and Sockets

[fostel at ncsu.UUCP]

    Someone suggested a way to extend pipes to serve as a better IPC system,
    by attaching filesystem names .. etc.  Someone else responded that Berkley
    had done just that and the result was sockets.  Well, the history is very
    long on this item and Berkley is far from the first (or likely the last)
    word on the subject.  The oldest reference I have is a paper by Carl
    Sunshine at Rand, circa early 70's.  The name chosen then was "port",
    a much nicer one than "socket" except for the popularity of sockets in
    networking jargon.  A port was just what you might expect.  To allow an
    interrupt like system (essential for the server systems such IPC is often
    used for) a new system call was added, the "await" call, which allowed
    you to go to sleep until there was activity on the port.  Also a "capac"
    call to allow polling on the availability of data as an alternative to
    await.

    Similar though I think not quite as nice was Berkleys MPX files, Bells
    Fifi files and a host of varioations on the original Rand model.
    Decendents of the Port model are still in use, e.g NBS networking modeling.
    (sorry Fif-o- files.) I have not yet seen a good description of BSD4.2
    so I don't know if berkley has advanced the state of the art or just
    re-written it.  <- last time someone confessed ignorance of 4.2, there was
    a rather snide remark about all the ways of learning about it and the
    lack of weight behind an opinion from such a person.  I have learned that
    you can never tell whether a UNIX anything is good until you try to use
    it, since the written descriptions tend to be _______ (supply your favorite
    word) ->
    ----GaryFostel----