[SCC_Active_Members] CICS and Multics

[Olin Sibert]

There's a lot of information about Multics at Tom Van Vleck's
excellent Multicians web site:
    www.multicians.org

In addition, I'm coordinating the grandly-titled Multics
History Project, an effort to collect and scan (and catalog and
deliver to the Museum) all the Multics-related documents we can
find, starting with MIT's collection (120K pages, now about 70%
done).

We're looking for all sorts of Multics documents to round out
the collection; if you have any, please get in touch with me!

Grady's Multics kernel is likely to be primarily of historical
value--and as "several boxes" of cards, I can't imagine it's
anything but a binary image of some early part of the 645 boot
process--a box of 2000 cards can encode at most 54K Multics words,
whereas even the first part of a modern Multics boot image was
over 100K words.

As for source code, the Museum already has a full set of Multics
release tapes for the final Multics releases (MR12.x, late 1980s).
These have been read and converted to modern bits by Peter Flass.
In addition, I've collected a set of tapes for the MR10.2 release
(1984 or thereabouts), and there is fiche for earlier releases
from MR3 to MR9 (which we have no current plans to scan).  These
tapes include both binary and source code.

A direct port of any of this code to the Pentium (or any other
modern architecture) is probably not practical.  The primary
reasons are that the code is all in PL/I (with many Multics PL/I
extensions) or Multics assembler language and that it is heavily
dependent on the 36-bit word size (and on 9-bit characters and
18-bit halfword offsets) of the Multics hardware.

A more feasible (and historically interesting) approach is
emulation, which is eminently practical and would be much faster
than any real Multics hardware ever was (it topped out at 1.9
MIPS per CPU).  The CPU architecture is complex (particularly the
character and bit-string instructions), but it is reasonably
well-documented.  The real problem is the I/O devices, which are
not as well-documented or well-defined, and are themselves quite
complex.  I've had a CPU emulator at various stages of operation
over the years (but NOT for the string instructions), but it
never got past the first few dozen instructions on the boot tape
because it needed tape and console I/O, too.

Emulation for the 645 is more problematic; although the CPU
is similar (and much simpler), the I/O is even more mysterious,
and we have no source code from that era as a guide.

Cheers -- Olin Sibert

At 03:47 PM 7/15/2005, Van Snyder wrote:
>On Fri, 2005-07-15 at 11:36 -0600, Grady Booch wrote:
>
> > From the earliest days of working on software preservation with this
> > group, I'd collected several boxes of punched cards constituting the
> > Multics kernel, and I also have (on microfiche) the source code for an
> > early early version of IBM's CICS.
>
>A Multics kernel! Fantastic!  From my position of meager knowledge, it
>seems that the Pentium is a better Multics platform than the Honeywell
>6080 was (not least that one need not have a plumber to keep it going).
>Any hope of getting the rest of it?
>
>--
>Van Snyder                    |  What fraction of Americans believe
>Van.Snyder at jpl.nasa.gov       |  Wrestling is real and NASA is fake?
>Any alleged opinions are my own and have not been approved or disapproved
>by JPL, CalTech, NASA, Frederick Gregory, George Bush, or anybody else.
>_______________________________________________
>SCC_active mailing list
>SCC_active at computerhistory.org
>http://mail.computerhistory.org/mailman/listinfo/scc_active