1. UM (University of Michigan)

posted Oct 30, 2010, 4:56 AM by Jeff Ogden   [ updated Sep 6, 2014, 9:39 PM ]
MTS was created at the University of Michigan starting in 1966, so they didn't have to learn about it in the same sense as the other MTS sites. UM did, however, have to make a decision to develop and run it.

Robert Bartels was Director of UM's academic Computing Center. Bernie Galler, Bruce Arden, and Frank Westervelt were Associate Directors at the Computing Center. Frank was also Director of the ARPA funded CONCOMP project. According to Mike Alexander, Scott Gerstenberger, and Bruce Arden it was Bartels and Westervelt who  contacted Lincoln Labs. to obtain LLMPS and pushed the effort to develop MTS forward at Michigan.

The story of how MTS got started at Michigan has been told in several places:
From pages 23 to 25 of An Interview with Bernard A. Galler:

Now, getting back to the Computing Center activity, in the early 1960s, we needed to continue to expand our
computing facilities, and we talked to IBM about the 360 family. They were interested in time sharing, but not too
interested. They had talked to MIT; the MIT people were very active. They had had success with their CTSS
system, they had ideas about how to build computers better for time-sharing systems, and we knew about those
developments. We also knew that IBM and MIT had had big arguments and had had a falling out. They were not
really talking to each other. We began to talk to IBM about what we needed. And at first, we had some cooperation
with MIT and their GE system. They invited us, in fact, to join them in developing an elaborate time-sharing system,
which ultimately became MULTICS, with GE. We decided, with some internal discussions, that if we did that, we'd
always be second fiddle to MIT, that we thought we knew a lot ourselves, and it would be better if we didn't join that
particular consortium. Also, there was a great deal of secrecy of what was going on there, because it was considered
a proprietary product for General Electric, and that just wasn't our style. We decided to talk to IBM about how to
upgrade their computer to be better for time sharing. Later on, it turns out that the MIT people accused us of stealing
our ideas, and we didn't think so. We had some very soul-searching discussions. Did we, in fact, not treat what we
did know about them properly? We decided that we acted ethically, but for about ten years, the MIT people didn't
talk to us; in fact, were pretty negative in talking about us to other people. But it blew over finally, and now we're
friends, and it's okay.

The discussions we had with IBM took the form of "You've got this new 360 system. It needs some additional
hardware in order to make time sharing work better." We had some software ideas, we published them in the ACM
journal, and we did influence IBM to add some additional registers, and, in fact, it was virtual storage, the beginning
of virtual storage. MIT had developed one view of it, and one way of doing it; ours turned out to be a different
abstract model of what it should look like, and so it ended up different on the IBM machines. We asked IBM to build
a machine for us with that stuff on it, and we said we would develop the operating system for it, because we knew
how to do operating systems. Just give us the hardware that we need. And they began to build the model 360/66M.
The M, we think, was Michigan. It may have been for "modified"; we don't know. But in any case, they started to
build that for us. We mentioned what we were doing to some other people - Perlis at Carnegie, the General Motors
people, and so on - and they got interested and began to tell IBM that they wanted the same thing. And IBM finally
realized that they had a potential product. So they committed to building a new machine, called a 360/67, and they
said, "Because we've got these customers, we'd better build an operating system for it." So they came to us and said,
"We're going to build a number of them, and we'll provide an operating system." And what they did was they
brought together six of their most influential customers. We were one of them, General Motors, Lincoln Labs,
Princeton University, somebody else, and we were the inner six that were advising IBM on what the operating system
should look like. Unfortunately, each of us had different needs. We needed to run lots of small problems for
students, General Motors needed to have very large data sets for their automobile model designs, Lincoln Labs
needed ... they argued for a long time; we finally talked them out of it ... two special bits in each word so they could
recognize a specific kind of data coming in from a satellite to Hawaii, and that sort of thing, and everybody had
different goals. IBM tried to accommodate everybody by putting everything in, everything, they only missed the
kitchen sink. The system grew. It was called TSS, Time-Sharing System. It grew and grew and became so unwieldy
that on Black Monday, January 18, 1967, they sent representatives to everybody at that time who had ordered that
computer - there were at least a couple hundred universities by that time that had heard about it and were waiting for
it and had ordered machines - they sent people out that day to say to everybody at once, "Sorry, we are canceling
the operating system. The machine is now going to be called an experimental machine, and we're not so sure we are
going to deliver it." And so on. And they lost 120 orders that day. Michigan decided to stick with it. We had
confidence in it, and we said, "We think it will work. We think the ideas are good." That was in January. We began
to have some meetings ourselves as to what to do about this, and we realized that we'd better do it ourselves and
take advantage of the fact that Mike Alexander, who had joined us as a programmer, was playing around with those
little time-sharing systems that people had developed by that time, and was somewhat familiar with the
multi-programming system from Lincoln Labs, and we said, "Let's develop our own based on that, and make it
available to the campus." The model 67 had just arrived, and it was possible, using very elementary techniques, to
get a few people on in a time-sharing mode. We said to people on the campus, "We will deliver services in May free
of charge on this thing, and in June we will start charging you." It was a rather elementary system at that time, but
because we adopted the standard IBM linkage conventions, we were able to bring in all kinds of software from other
places and just dump them into our system, and it worked. So it grew very rapidly. All of that software was public
domain at the time. There was no question about being able to use it in terms of copyright and so forth. The
computer software industry hadn't started yet. That really started, in terms of selling software and so forth, in 1969,
when IBM decided to unbundle their software and charge for it separately. So we were still in a period when people
were exchanging software quite freely, and certainly all of what IBM delivered with their computer was considered
public domain. We didn't use the IBM operating system, but we were able to take whole sections from it like the
FORTRAN translator and so on.

So we were in business quite rapidly, and MTS, as the new system was called, the Michigan Terminal System, grew.
It was quite good, and people - Alexander and Boettner were the two who really developed it. Other people jumped
in and helped developed stuff. I remember Westervelt and O'Brien and other people were there. It was a very
exciting time, lots of new developments, and especially the demonstration that the model 67 worked.

This is from an e-mail message that Mike Alexander sent me in June 2010:

We got the 360/67 in January and Don and I spent the next 10 months implementing paging and virtual memory.  We used the 360/67 in model 65 mode until November when we switched it to model 67 mode and started using virtual memory and paging. 

One other anecdote: a week or two after we turned in virtual memory support I ran into Bernie in the corridor and he asked me when we would be switching it on.  The change was transparent enough that he hadn't noticed it.

In retrospect it's amazing that we went from getting our first 360 in the spring of 1966 to running a virtual memory system by the end of 1967.  I guess we didn't know it was impossible so we just did it.  Of course it was pretty limited VM with little sharing of memory and only 24 bit addressing.



On Dec 2, 2010 at 1:58 PM PST, Jim Bodwin wrote:

Also, don't forget the install at UM Human Genetics.  One interesting story about the reliability of MTS from the HG install:  MTS required a reboot to switch to/from daylight savings time.  I went over the day after the time change to reboot it and discovered that the last time it had been rebooted was 6 months prior when I had last switched it to/from DST.



On September 6, 2014, at 11:29 AM, Jeff Ogden wrote:

It turns out that Lincoln Manual 78 is available on the web (once I knew what to look for). If you are looking for something to read, you might try:

     -Jeff

On Sepember 6, 2014, at 12:02 PM, Mike Alexander wrote:

Now that takes me back.  It’s been a long time since I’ve seen that document.

There must have been an earlier version of it since we got LLMPS at least a year earlier than January 67.  By then MTS was well underway.  January 67 was a milestone month for MTS since that’s when the first version of the DSR interface was implemented.  We had LLMPS doing spooling for the 7090 by the spring of 1966.  I think it was sometime in the summer of 1965 that I went to Lincoln Labs to take a look at it.  I remember I got a lift in a private plane from someone visiting UM from Boston and I think he was here for the Engineering Summer Conferences that year.  I’m pretty sure there was a version of this document available then, perhaps it was a draft that hadn’t been published yet.

             Mike
Comments