Path: utzoo!attcan!uunet!husc6!bloom-beacon!apple!bionet!agate!saturn!perry@apollo.COM
From: perry@apollo.COM (Jim Perry)
Newsgroups: comp.os.research
Subject: Re: Earliest use of Threads (was Re: THREADS, Light weight processes)
Message-ID: <5368@saturn.ucsc.edu>
Date: 4 Nov 88 16:13:00 GMT
Sender: usenet@saturn.ucsc.edu
Organization: Apollo Computer, Chelmsford, MA
Lines: 39
Approved: comp-os-research@jupiter.ucsc.edu


In article <5338@saturn.ucsc.edu> fouts@lemming. (Marty Fouts) writes:
>
>I offer, only half humorously, TSO running under OS/360, as the first
>example of "multiple THREADS of control executing within a single
>address space"

Again, "threads" or multitasking is a very old technology, a natural
outgrowth of the fact that I/O is too slow to be done synchronously.
In more recent operating systems (well, in Unix, which seems to be
what everyone means by OS) I/O scheduling is done among processes,
i.e. when a process initiates an I/O operation it is suspended until
that operation completes.  Other systems allow various mechanisms to
allow the process to continue past the I/O call, with some means of
announcing when pending operations complete.  A program using such a
facility is multi-threaded in the simplest sense, and a more elegant
model is easily constructed from such a system.

The DTSS system (Dartmouth TimeSharing System, though they call it
something else these days) relies heavily on this model.  Far from 
the more common approach these days, the command processor (shell
equivalent) is a single program, which holds open all user terminals
with a task (thread, LWP) to handle each.  The original system was
based on an assembly language model, and so the notion of a task was
loose and up to each application; over the years subroutine 
packages provided cleaner interfaces in Dartmouth's dialect of XPL 
and then PL/I.  A few years ago the OS, PL/I compiler and runtime 
system were modified to provide an excellent multitasking model
that's cleaner and more usable than anything I've seen since (to
be fair, if you get to control the OS, the language, and the runtime
system, and are unconcerned about portability, you have a lot more
leeway than e.g. trying to make something that does all this in
portable C).  I know that IBM PL/I and Ada also provide language
support for multitasking, and have been somewhat surprised not to
see these mentioned.  Has the CS world really forgotten everything 
that came before C and UNIX, or is that just a feature of the UNIX
bias of usenet?
-- 
Jim Perry   perry@apollo.com    Apollo Computer, Chelmsford MA