Dinner Talk by Rey Johnson at the DataStorage '89 Conference
on September 19, 1989 held in San Jose, California



    Good evening.  Thank you for the introduction.  I’m delighted to be able to spend the evening with you.
    I’ll be talking this evening about some of the highlights in the development of the first random access disk product – the IBM RAMAC 350 file – The why, where, when and how of the first disks.
    This is the product that spawned the magnetic disk storage industry – an industry that has since come to generate an annual revenue of 23 billion dollars. I think it fair to say that the RAMAC 350 has carved a place in history for itself.
    I am sure that many in my audience were not born in 1951 when my story begins.  Let me therefore set the stage for the events, most of which took place in San Jose, a few blocks from here.
    In 1951, San Jose was a city of 100,000 people.  Its economy was based mainly on agriculture.  At the same time, IBM was a rapidly growing company, with data processing as it main business.  Its revenue in 1951 was about $250,000,000.
    After IBM had developed two large computers that were intended strictly for scientific applications, IBM management saw little evidence that computers would become profitable business products.  In fact, in 1951, an extensive survey of all potential computer customers yielded statistics that indicated that 17 or 18 computers would saturate the market.
    At that time, Thomas J. Watson, Jr., was succeeding his father, Thomas J. Watson, Sr., as president of IBM.  He decided to build 19 scientific computers.  The first of them was completed in 1952.
    IBM’s business in punched card equipment was growing as rapidly as we would build products.  Expansion was in the air at IBM.
    In 1951, IBM corporate headquarters in New York decided to establish a research laboratory on the West Coast.  Research was a term that at that time covered all types of engineering activity.
    The West Coast was chosen because IBM’s customers in the aircraft industry were creating innovations and modifications that were considered to be potential products.  For example, the first card programmed calculator originated with engineers at the Northrop Corporation.
    The bay area was chosen as the site of the new IBM research laboratory because it was between Los Angeles and Seattle, where those innovative customers were based.
    San Jose was chosen as the specific site because IBM already had a punched card plant here, at 16th and St. John.  This plant housed the district manager, an accounting department and a cafeteria.  Under the direction of Roger Williams, IBM’s community relations were very good.
    During the first week of January 1952, I was told of my appointment as West Coast laboratory manager.  I was told that I would have free rein in hiring a staff of 30 to 50, and that I would be free to chooses projects to work on.  One-half of my projects were to be new IBM products and one-half were to be devices in support of customer’s special engineering needs.  No projects were to be duplicates of work in progress in other IBM laboratories.  The laboratory was to be dedicated to innovation.
    My first act as manager of the new laboratory was to rent a building, and the second act was to place an ad in all West Coast daily papers, announcing that IBM was opening a laboratory in San Jose.  The ad noted that positions were available for scientists, engineers and technicians, and it brought in 400 applications.
    The IBM Research and Engineering Laboratory opened its doors at 99 Notre Dame, a few blocks from here, on February 1, 1952.
    I was told that my flair for innovative engineering was a major consideration in my selection to manage the new laboratory.  During 18 years with the IBM Endicott laboratory, I had had responsibility for numerous IBM products -- test scoring, mark sensing, time clock products, key punches, matrix and non-impact printers and random card file devices.  By 1952, I held over 50 patents, some of them fairly good.
    To be given freedom to choose our projects and our staff made the San Jose laboratory an exciting opportunity, especially since funding was guaranteed -- at least for a few years.
    The first few months of 1952 were consumed largely in interviewing and hiring a balanced staff of talented and experienced engineers, technicians and administrative personnel.
    Except for one person from each of our two New York laboratories, and one engineer from my department in Endicott, New York, we were under orders not to recruit people from the eastern sites of IBM.  As a result, our first crew all came from the West.
    Among the first projects, undertaken during the start-up were a non-impact printer, a test scoring machine, source recording equipment and a random access replacement for tub files.
    It was the search for an automatic random access system to replace tub files that led us to explore magnetic systems.
    In 1952 IBM was producing sixteen billion Hollerith cards per year.  Each of these cards had to have information entered into it in the form of punched holes before it could be usefully processed by accounting machines.  Manual key punching was one of the most costly items in customer data processing operations.  In many applications, most of the information in a card was unchanged from week to week.  In a payroll application, for example, only hours worked may be new.  An automatic tub file would automatically enter status information and the key puncher operator would be relieved of punching anything but new data.
    After deciding that our random access component was to be based on a magnetic recording system, we proceeded to explore the most probable magnetic media.  We explored magnetic drums, magnetic tape loops, magnetic plates, magnetic tape strip bins, and even magnetic wires and rods.
    Rotating magnetic disks came out on top in our analysis, chiefly because of its rotational dynamics, the potential of multiple accesses and the efficient surface-to-size ratio.
    As time went on, our engineers became inspired by the possibility of developing a product that gave essentially instant access to file data, not only when connected to key punches but also when connected to accounting machines and maybe even to computers.
    Two events in 1953 turned out to be fortuitous for out disk project.  We ended an automatic data reduction project being done under contract with the McDonald Douglas Aircraft Company.  This released a half dozen talented electrical and system engineers, who were then available for reassignment to the disk project.
    The second fortuitous event was the receipt of a request to bid from the U.S. Air Force Supply Depot in Ohio.  They called for a material information flow device.  They wanted instant access to each of their 50,000 item inventory records.
    We were simultaneously studying file applications in wholesale grocery and wholesale paper supply companies in the bay area.
    We pooled these insights and information and prepared a set of specifications for a general-purpose random access memory.  These specifications recorded in February 1953, in his notebook by Art Critchlow, turned out to be almost identical to the RAMAC 350 disk file specifications announced two years later.
    Going from the wish list provided by the specifications to the reliable operating model required solving many technical problems.  With added staff available we proceeded to attack all key problems simultaneously.  What kind of disks could we use?  How could we get them to run true?  How could we best paint them with the iron oxide paint?  What kind of magnetic transducer should we use?  How could we keep the read/write head close to the disk without having it wear out?  And finally, how were we to move the head to any one of 50,000 tracks in less than one second or in one accounting machine cycle?
    We proceeded to test out ideas.
    We tested the dynamics of rotating disks by mounting 120 aluminum disks two feet in diameter on a shaft with about ¼ inch spacers and rotating this array at 3600 rpm.  One test run of this model allayed out fears about problems of excessive wind vibration, power requirements and even excessive disk wobble.
    However, one problem that turned out to be quite difficult was coating the disks with iron oxide paint to a uniform thickness and smooth finish.  The oxide paint we were using was essentially the same as was used to paint the Golden Gate Bridge.  One of the engineers suggested pouring the paint near the center of a rotating disk and allowing centrifugal force to spread a smooth uniform coat over the disk surface.  Another engineer found that filtering the paint through a silk stocking showed that by filling a tray of paper cups with just the right amount of paint, the coating thickness would be the same from disk to disk.  This system was used for many years.  It was later incorporated into the equipment that automated the process.
    Another group of engineers was assigned to develop a small thin head for the record and readback functions.  None of our staff knew much about magnetic recording.  So we hired a consultant, Al Hoagland, who at the time was a graduate student at UC Berkeley and an expert in magnetic recording.  About the only magnetic transducers in use in 1953 were those used with magnetic drum and magnetic tape equipment.  Both of these had entirely different space and positioning constraints than we had.
    Early in the development of the read/write head we decided to protect the head against wear by using air pressure with nozzles in the face of the flat head.  The airflow spaced the head a uniform distance from the sometimes wobbly disks.  Air pressure was also used to force the head toward the disk after it reached its destination.
    Stored bit density at the center tracks was made the same as the state of the art density in magnetic drums, 100 bits to the inch and 20 tracks to the inch.  This was better than a 4,000% improvement over punched cards in information density and the data was alterable and erasable.
    In our first file model two-foot diameter oxide-coated disks were mounted on a horizontal shaft at ½-inch intervals.  Fifty-one disks gave 100 inside surfaces.  Two opposite facing heads were mounted on one access arm..  The access arm was moved so as to place the heads on any of the 100 tracks on each of the 100 disk surfaces at a speed that would match an accounting machine cycle, which was less than one second.  We had anticipated that there would be a need for as many as twelve access stations on each file.  A provision that proved to be excessive.  The maximum travel between addresses was one-twentieth of an inch.
    Two access drive systems were designed and modeled, one mechanical and one electronic-servo system.  We finally chose an electronic-servo system for the first file model.
    The tub file application led us to test out the disk performance by pairing it with a keypunch, because the keypunch could be used for entering information and for recording in punched holes the information read back from the disk file.
    On February 10, 1954, this first sentence was fed into and read back from the disk file – “This has been a day of solid achievement.”
    By March, 1954, tests of components and the card to file machine made us confident of being able to build a product.  Lou Stevens was made the manager with full development responsibility.
    He and his staff of very capable engineers initiated a program of re-design that started in mid-March 1954.  By November this design had matured into a “magnetic disk processing machine.”  RAMAC was on its way.  Hopes were high that his revolutionary concept would develop into an IBM product.
    The potential for large random access memory was attested to be activity among competitors who were using very large drums, drum arrays, tape loops and even the surface of a power station fly wheel as recording surfaces.
    The IBM vice-president for marketing, L.H. Lamotte, stationed his long-range planner, F.J. Wesley, in our laboratory mid-summer, 1954.  On October 8, 1954, Mr. Wesley sent a memorandum, which he called a “pontifical announcement,” to his boss.  In part it started, “we must immediately attack accounting problems under the philosophy of handling each business transaction as it occurs, rather than using batching techniques.”  Wesley’s memo was widely circulated among IBM management and, needless to say, in our laboratory.  The promise of developing a product for more than a file tub replacement led to a corporate decision in November 1954 to build at least five prototypes of our product to field test.
    Initial specifications for the RAMAC were prepared December 17, 1954.
    The non-RAMAC projects of the research and engineering laboratory moved to Julian Street to open more spaces for the RAMAC team.  At the Julian Street laboratory work continued on advanced ideas for disk files.  Gliding heads and multiple parallel access arms were developed and eventually transferred to Lou Stevens domain where these features were incorporated into disk file products that followed after the RAMAC 350.
    Lou Stevens and his engineers at 99 Notre Dame successfully demonstrated and operated the Model II file on January 16, 1955.  Debugging of this machine continued around the clock for months.
    Early in 1955, a corporate decision was made to build 14 RAMAC 350 machines for internal use and field-testing.
    On May 6, 1955, IBM held a press conference to announce that it was bringing out a new product “that takes information from a stored program using a multi-million character random access memory that makes it possible for the new system to do a whole job automatically without using batch processing.”  On August 25, 1995, the San Jose Mercury News published a short article while said, “IBM plans a giant new San Jose plant that may employ more than 5,000 here.”
    The first RAMAC was shipped to Zellerbach Paper Company in June 1956, by an outstanding San Jose manufacturing team.  Mr. Porter remembers the event.
    T.J. Watson Jr., President of IBM, announced the RAMAC on September 4, 1956.  He said in part, “This is the greatest product day in IBM’s history and I believe in the office equipment industry.”
    The RAMAC was featured in the United States exhibit at the World’s Fair in Brussels – and in the United States Technical Exhibit in Moscow.  Chairman Khrushchev of Russia came to San Jose in 1956 to visit the RAMAC plant.
    The American Society of Mechanical Engineers designated the RAMAC an international historic landmark on February 27, 1984.
    The first four years of disk file history have been reviewed in my remarks.  In the 33 years since the RAMAC 350 left the laboratory, tremendous advances have been made in product development, manufacturing and research, as you all well know.