Notes from "Calculating a Natural World: Scientists, Engineers, and Computers During the Raise of U.S. Cold War Research"

From: Jeff Ogden <jco@umich.edu>

Subject: Re: Request for Information Regarding UM and IBM History

Date: June 1, 2017 at 1:04:59 PM EDT

Included below are some excerpts from a book I mentioned earlier, Calculating a Natural World: Scientists, Engineers, and Computers During the Raise of U.S. Cold War Research, by Atsushi Akera, 2007, MIT Press. I think these quotes give some hints about the relationship between IBM and U-M in the early years of computing. The book really only talks about the academic side of things at U-M and doesn’t go into the relationship with IBM on the administrative side (Data Systems Center and Hospital Data Systems Center).  The book has a lot of early history, but the U-M part of the story that is covered runs from the early 1950s through 1971 or so.

pages 298-299: Origins of Michigan's Computing Center

  The immediate antecedent of Michigan's first computing center was its Statistical

Research Laboratory (SRL). Cecil Craig, a well-known statistician on

the facu lty, convinced the administration to set up a tabulating facility in

the Horace Rackham School of Graduate Studies in 1947. Following the

same technical trajectory being followed elsewhere, this evolved into a

computing facility used by many members of the faculty. Citing the interest

the Midac had generated, Craig convinced the dean of the graduate

school that Michigan ought to get a computer for research and instructional

use. In the meantime, the negotiations between MIT and IBM had

taken place. As an outgrowth, IBM announced a broader Educational Contribution

Pmgram that offered, in effect, a 60 percent discount to academic

institutions seeking to lease any of its smaller machines. This included

IBM's magnetic drum computer, the IBM 650, which was installed in the

SRL in March 1956.[44] About a year and a half later, IBM's new Applied

Science Director, Charles DeCarlo, approached Michigan's Vice President,

William Stirton. DeCarlo offered a deal similar to its new arrangement with

MIT. This would have been IBM's second regional academic computing

facility. However, Stilton rejected DeCarlo's offer. In part, the offer was not

as sweet as the one given to MIT. But more important, Stlrton was unwilling

to appmve an installation open to other Midwestern universities. He

remained more guarded about committing Michigan's facilities and facu lty

to other universities. Stlrton may have also felt that such a service center

would produce unfair competition for regional businesses seeking to enter

the emerging computer services market. In any case, the offer was more

attractive when it originated with the institution in question, rather than

when tendered by a private firm.”[45]

  DeCarlo persisted in his effotts, recognizing that Stitton was an important

person to enlist. It helped that Stirton, in turn, could not remain deaf

to the groundswell of interest in computers. Immediately after the IBM 650

was installed, eighty students from Carr's Math 173 besieged the new facility.

Eighty-three percent of the machine's first 20,000 hours were spent on

instructional use or non-sponsored research, verifying that there was broad

interest in computers.[46]


page 309: 

  Still, it was significant that Michigan's administrators recognized the

intellectual merits of the work conducted within its computing center.

Bartels and the members of his executive committee publicized the work

on GAT. MAD, and other systems as academic achievements. The administration

responded in kind, authorizing Bartels to augment the Computing

Center's programming staff to expand on this work. IBM also

contributed to the pot, in establishing a new fund for systems programming

research at the center.


page 321: 

  A more serious crisis resulted from a subtle shift in IBM's educational discount

policy. IBM's "standard educational contribution" amounted to a 60

percent discount on standard rental rates, but the contract Michigan signed

with IBM was somewhat more complex. Michigan's contract stipulated

that the university would charge no fees for educational or non-sponsored

research use of the computer, while charging ail sponsored research projects

a "full commercial rate" set at 1/176th of the normal monthly charges.

Michigan was a lso required to limit sponsored research use to 70.4 hours

per month. IBM included this provision to prevent the Joss of business

from sponsored research projects that could afford to pay for computer

time. But as IBM came to dominate the market, it grew more concerned

about how a complex agreement, irrespective of its legal standing, would

play into accusations of market manipulation. In March 1963, IBM proposed

to amend the agreement to provide Michigan with a Oat 60 percent

discount for a ll academic use. Seeing no obvious implications for his

bottom line, Bartels agreed to the amendment."

  This time it was the university's legal counsel who advised Bartels that

the Computing Center would have to reduce the rate it charged sponsored

research proJects from $460 per hour to $175 per hour because of the

amendment. Since there was no longer any explicit provision for free …


page 330-332: Michigan's Success

  Intractable difficulties also plagued the effort at IBM. Time sharing still was

not one of IBM's top priorities, and designing a system that remained as

compatible as possible with the rest of the System/360 series was an especially

difficult challenge. In August 1966, IBM advised Michigan that it was

unlikely that they could deliver their machine before December, whereas

it had been scheduled to arrive early in the fa ll. Worse yet, IBM told

Michigan that the accompanying software, Time Sharing System (fSS),

would not be released until the following April, and with no guarantee as

to its performance. In January 1967, this was followed with a more formal

announcement that IBM had blocked further sales of the IBM 360/67, and

was going to release TSS only for experimental and educational use.”[42]

  This brought Michigan squarely back into the midst of computer time

sharing. Bartels had announced the inauguration of a new time-sharing

service before he leamed of IBM's difficulties. Feeling compelled to deliver

on their commitment, Arden, Westervelt, and others began scrutinizing

the situation. They came to the conclusion that IBM had lost confidence

In their own system after a software simulation revealed that the IBM

360/67, when In time-sharing mode, would have 1/10 the computational

power of the older IBM 7094. From his knowledge of the hardware, Westervelt

did not believe this could be true. He proceeded to carry out an independent

eva luation of the IBM 360/67's performance. This led him to the

conclusion that the source of the difficulty was IBM's time-sharing software,

not the hardware. In the meantime, two of Arden's staff members

had obtained the source code for the Lincoln Terminal System (LTS), an

experimental time-sharing system developed separately at MIT's Lincoln

Laboratories. Arden authorized his staff to port LTS first to their IBM 7090,

and then their new IBM 360/67. At this point, Michigan also received a

grant from NSF."

  Michigan was able to get a time-sharing system into operation, and to

do so in short order. A rudimentary version of the Michigan Terminal

System (MTS) was up and running by May 1967. The system attained an

overall capacity for simultaneous users comparable to Corbato's crss by

November. By the following August, Arden and his staff were able to release

a more extensive and robust version of MTS that made full use of the IBM

360/67's hardware modifications. At the end of 1968, MTS was the only

large-scale time-sharing system that was operating reliably.[44]

  There was a technical basis for Michigan's success. One of the main technical

challenges of time sharing had to do with the limited size of a computer's

core memory. However, both Arden and Galler had already

encountered a similar problem while working on high-level languages.

They had developed a utility that automatically loaded sections of the

MAD compile•·'s translation tables into core memory. Conceptually, by

extending this idea from a single application to the operating system as a

whole, it was possible to create an illusion that a machine had a very large

core memory. This illusion was available to programmers as well as any

other user, so that the abstraction simplified Michigan's systems programming

work. The efficient implementation of this scheme required special

hardware modifications- precisely the ones At'den and his colleagues

requested of IBM. Technically, this involved having special circuits that

made it possible to rapidly reallocate different sections of core memory.

Through such work, Michigan became one of the sites that contributed to

the development of "virtual memory.""

  The ideas central to virtual memory had originated with MIT, not

Michigan. It was Jack Dennis, one of MIT's electrical engineering facu lty

members, who first presented the important idea of "segmentation."

Dennis was designing yet another time-sharing system for the Digital

Equipment Corporation's (DEC) PDP-I, a tiny machine with very severe

resource constraints. By Arden's own admission, Dennis's paper led to

obvious questions of implementation. Drawing on the same work by

Dennis, Fano's major disagreement with IBM had to do with IBM's initial

unwillingness to incorporate similar dynamic reallocation hardware into

its time-sharing system.[46]

  Yet it was significant that only Michigan's Computing Center possessed

both the requisite hardware and the pressing need to implement a largescale

computer time-sharing system. In comparison to the competing

demands of MIT's Computation Center and Project MAC, Michigan was

able to follow the more productive middle road.

A few notes to give a little more context:

page 299: William E. Stilton is mentioned as a U-M vice president. In 1957 Stilton became the U-M’s new vice president for liaison with industry, business, and professional groups. He went on to be the director of the U-M’s then new Dearborn Center while continuing his role as vice president.

page 331: The details associated with the development of LTS / MTS on an IBM 7090 are simply wrong. Mike and Bruce can tell you more about this. There are also a couple of items on the MTS Archive Web site that talk about this misunderstanding a bit:

There are some more items that may be most interesting to you, although there are other items in the Myths and misconceptions section of the MTS Archive site that might be of interest too:

   -Jeff

Comments