Why 1950s Computers Could Calculate But Couldn't Find Anythng

[00:00:00]In 1953, a computer could perform thousands of calculations in the time it takes you to blink. But if you asked it one simple question, how many refrigerator compressors were sitting in a warehouse in Indiana? It might take an hour to find the answer. Not because the answer was difficult, because computers had no idea where to look. It was a Tuesday morning in the autumn of 1953, and Walter Hedges was standing in the tabulating room of a midsized appliance distributor in Dayton, Ohio, waiting for a machine to tell him something he already needed to know an hour ago. A regional sales manager wanted a single number. How many refrigerator compressors were sitting in the Terraote warehouse? It sounded like the simplest question in business, but to answer it, Walter had to pull a tray of punched cards, feed them into a tabulating machine, and wait while the machine read every single card in the deck, one after another, from beginning to end. If the compressor record happened to be near the front, he might have an answer in a few minutes. If it was near the back, and today of course it was, he would wait while thousands of cards representing every other product the company sold rattled through the reader first. There was no shortcut. A computer in 1953 could not jump directly to information. It could only start at the beginning and move forward. Nobody considered this a flaw. It was simply how machines worked. Companies hired extra clerks because of it. Entire overnight shifts were built around it.

[00:01:36]Businesses learned to adapt themselves to the pace of the machine rather than expect the machine to adapt to them.

[00:01:43]Walter had long ago stopped apologizing for delays. The tabulator hummed at its own fixed rhythm card after card after card, and his job was simply to wait.

[00:01:55]But 3,000 m away, a small team of IBM engineers was trying to do something that many experts believed was either impractical or impossible. They wanted a machine that could reach into a mountain of information and retrieve a single fact almost instantly. No searching from the beginning, no reading every record first, just the answer. The device they were building would become the first hard disk drive and it would change the way every computer on Earth stores information. For the better part of three decades, fast and random simply did not belong in the same sentence when people talked about business machines.

[00:02:36]The problem of finding one record without reading everything around it first was not a mystery nobody had thought of. It was a limitation everybody had made peace with. Five forces kept it that way.

[00:02:48]The entire tabulating industry was built on sequential logic by design, not by accident. Herman Hollerith's punched card system developed for the 1890 US Census and commercialized through the company that became IBM in 1911 worked by physically feeding cards through a reader one at a time. Every calculation IBM's business had been built around for over 40 years assumed that data arrived in a line. Random access wasn't a feature nobody had added. It was a concept the entire architecture had no room for. Even the machines built to sort cards faster. IBM's electromechanical sorters and collators of the 1930s and 1940s were faster versions of the same idea, not a different idea. They shuffled the line quicker. They never eliminated the line.

[00:03:37]IBM's own revenue model rewarded the sequential card system rather than punishing it. By the mid 1950s, IBM was leasing tabulating equipment and separately selling the punched cards those machines consumed by the billions each year. A business so lucrative that punched card sales were treated internally as a dependable, almost self-renewing source of income. [snorts] Approximate illustrative figure C sourcing flags. A company profiting handsomely from the cards themselves had limited internal urgency to make the cards obsolete. Every card a customer bought and fed through a machine was in effect a small recurring payment to IBM.

[00:04:16]And recurring payments are not the kind of thing a business rushes to replace.

[00:04:20]The available alternative storage technologies were sequential too, just faster. Magnetic tape, which IBM and its competitors began deploying commercially in the early 1950s, could move data more quickly than punched cards, but it still had to be read from the beginning of the reel forward. Magnetic drums, used in early computer memory, stored data on a rotating cylinder, but only in fixed preassigned locations with limited total capacity, typically enough for a program's working data, not an entire company's customer or inventory records.

[00:04:51]Neither technology let a machine reach for one arbitrary record among millions, without a structural rework. Businesses that upgraded from cards to tape in the early 1950s were in a very real sense buying a faster line, not a way to skip the line altogether. The corporate structure slowed the search for a fix.

[00:05:10]IBM's engineering leadership through the late 1940s and early 1950s was concentrated on the east coast in Endicott and Pikipsy, New York, close to the company's punched card manufacturing and its most established customers.

[00:05:24]Radical departures from that architecture had to compete for attention and budget against product lines that were already profitable and already understood by IBM Salesforce. An engineer proposing to replace the card reader with something nobody had built before wasn't just proposing new technology. He was proposing to compete internally against a business model that was already working. And perhaps most powerfully, businesses had simply adapted. Warehouse managers scheduled inventory counts around overnight batch runs. Insurance companies staffed second shifts specifically to process policy updates while the tabulators worked through the full deck. Retailers accepted that a stock check requested in the morning might not have an answer until the following day. The weight wasn't treated as a defect to be engineered away. It was treated as the cost of doing business with a computer.

[00:06:13]The same way an office in the 1950s accepted that a long-distance phone call required an operator and a wait. By 1952, sequential processing wasn't an obstacle standing in the way of American business. It had become the shape American business was built around. A set of habits, staffing decisions, and customer expectations, all quietly built on top of a limitation nobody expected to see removed. The pressure that finally broke this arrangement didn't come from inside a single overworked tabulating room. It came from the fact that IBM was starting to look for the first time in its history beatable. In June 1951, IBM's most serious rival, Remington Rand, delivered the UNIVAC 1 to the US Census Bureau, a fully electronic computer that generated national headlines and made IBM's punch card empire suddenly look like yesterday's technology. When UNIVAC correctly projected Dwight Eisenhower's landslide victory on live television during the 1952 presidential election, the machine became a household name almost overnight, and IBM's own executives understood exactly what that publicity coup had cost them in prestige. Inside IBM, the response was structural. In 1952, the company established a new research laboratory in San Jose, California, deliberately located roughly 2500 miles from the East Coast engineering culture that had produced the tabulator business. The mandate given to this new lab was blunt.

[00:07:41]Find a faster way to store and retrieve business data, one that could leapfrog the sequential logic IBM's entire product line still depended on before a competitor did it first. The commercial pressure was just as concrete. American businesses were scaling up rapidly through the post-war boom. More inventory, more customer accounts, more insurance policies, more payroll records, and the gap between how fast a company could grow and how fast its tabulating machines could search a card deck was widening every year. A retailer that wanted sameday answers on stock levels or an airline that wanted to check a single reservation without running an entire deck was being told in effect that the technology to do that on demand simply did not exist yet.

[00:08:25]Airlines in particular were becoming a visible symbol of the problem.

[00:08:30]Commercial air travel was expanding fast through the 1950s, and reservation desks across the country were still leaning on card files and switchboard operators to confirm whether a single seat on a single flight was still available. Every added route and every added flight multiplied the number of records a reservation clerk might need to search.

[00:08:49]And the sequential machines available to handle that search were falling further behind the pace of the industry they were meant to serve. IBM's leadership under Thomas Watson Jr. who had taken over day-to-day operations from his father, understood that the company that solved random access first wouldn't just win a product category. It would set the physical standard the rest of the computing industry would build around for a generation. Watson Jr. had watched Univac television moment closely, and he had no intention of letting his father's company be remembered as the one that missed the transition from mechanical tabulating to true electronic computing.

[00:09:23]There was also a quieter, more internal kind of pressure building. At the same time, IBM Salesforce, the same salesmen who had spent decades confidently telling customers that a tabulating machine was the only tool a modern office needed, were starting to feel a new and uncomfortable kind of question.

[00:09:42]Could IBM's machines look something up the way UNIVAC seemed to promise a computer eventually could? A sales organization built on confidence was suddenly being asked to explain account by account why the answer was still no.

[00:09:58]The race was on and it was being run in a nondescript lab building at 99 Notre Dame Avenue in San Jose, far from anyone's expectation of where IBM's future was being decided. Watson Jr. had given the West Coast operations something the East Coast labs rarely enjoyed, distance from the existing product line. and with it permission to build something that made the existing product line obsolete. The man IBM put in charge of that lab was not the profile most people picture when they imagine a computing pioneer. Reynold B.

[00:10:31]Johnson was 46 years old when he was asked to head IBM's new San Jose engineering operation in 1952. And his path there had started nowhere near a computer. Two decades earlier, Johnson had been a high school science teacher in Michigan, where he built a machine to automatically grade pencil marked multiple choice tests. A practical, unglamorous solution to a practical, unglamorous problem. He built it, notably with the help of two students assigned to him as punishment for stealing a school radio, turning their disciplinary sentence into an apprenticeship. IBM had shown no interest when he first pitched the test scoring device. Johnson persisted, eventually joined the company in 1934, and spent the following years as an engineer with a reputation less for theoretical brilliance than for stubborn, methodical tinkering, the kind of person who solved problems by building things with his hands and testing them until they worked, rather than by working the answer out on a blackboard first. By the time IBM sent him to California in 1952, Johnson was known internally as someone comfortable with failure as a working method.

[00:11:35]Colleagues described him less as a theorist than as a builder who kept a workshop mentality even inside a corporate research lab. Someone who would rather bolt together a rough prototype overnight and see it fail by morning then spend the same hours debating whether it would work. He assembled a small team of engineers around him among them Louis D. Stevens, Arthur J. Kitchau and several others recruited specifically for the San Jose labs mandate and gave them a problem statement almost embarrassing in its simplicity. find a way to reach any single piece of stored information in a fraction of a second without reading anything else first. Neither Johnson nor his team had a fully formed answer when they started. What they had was permission to try things that didn't work. That permission mattered more than it might sound. Johnson was reportedly proudest later in his career, not of any single mechanism his team built, but of the decision to keep concentrating the group's efforts on storing information on stacked laminated discs even after early versions of the idea underperformed. In a company built around the discipline of manufacturing precision, cards punched to exact tolerances millions of times over, Johnson was asking his engineers to spend years on an idea that for a long stretch simply didn't work as well as the technology it was supposed to replace. The San Jose lab did not arrive at its solution in a straight line.

[00:12:55]Between 1952 and roughly 1954, Johnson's team burned through a series of approaches that each solved part of the problem while creating a new one. And each dead end taught the group something they would eventually need.

[00:13:10]Rods and strips. The San Jose team's earliest experiments tried to adapt the idea of individually addressable storage elements directly. narrow rods or strips, each holding a small amount of magnetized data that a mechanism could select and read individually. In principle, each rod or strip could be reached without touching the others, which solved the random access problem on paper. In practice, the mechanical systems needed to select one specific rod out of thousands quickly and reliably were bulky, slow, and prone to jamming. a filing cabinet with thousands of tiny individually moving drawers, none of which held enough data to be worth the mechanical complexity.

[00:13:52]Magnetic tape reconsidered.

[00:13:55]The team also revisited magnetic tape, hoping that faster tape transport mechanisms or clever indexing might make sequential media feel random enough for practical use. It didn't. No matter how quickly the tape moved, a machine still had to physically wind past everything between the start of the reel and the record it wanted.

[00:14:15]Speeding up the weight was not the same as eliminating it. And for a customer who wanted one number out of a million records, even a fast sequential search was still a sequential search.

[00:14:26]Flat plates and early disc experiments.

[00:14:30]Before settling on their eventual design, the team also tried flat individual magnetic plates, essentially single discs, tested in isolation without the stacked architecture that would ultimately define the project.

[00:14:43]These early experiments demonstrated that a magnetic surface could store retrievable data reliably. But a single plate held far too little information to be commercially useful on its own. And multiplying the number of separate plates without a way to access them from a shared mechanism just recreated the same selection problem the rods and strips had run into.

[00:15:04]By the time the team had worked through rods, strips, tape, and single plates, roughly two years had passed with no product IBM could sell, and the pressure from Watson Jr.'s office to show results was not getting any lighter. Each failed attempt narrowed the problem rather than solving it, though. And that narrowing was its own kind of progress, even if it didn't feel like it inside the lab. By process of elimination, the team arrived at a question that would define the project's breakthrough. What if instead of thousands of small storage elements, each needing its own selector, there were a smaller number of large surfaces and a single moving arm smart enough to find its way to any point on any of them.

[00:15:43]It was in its way the same insight a librarian might have offered for free.

[00:15:48]Don't build a thousand tiny cabinets, each with its own lock. Build a few large shelves and train one librarian to walk directly to any book on any shelf without searching the ones beside it.

[00:15:59]The team's task from here forward was mechanical, not conceptual. Building an arm precise enough and discs reliable enough to make that librarian's walk take a fraction of a second rather than a fraction of an hour. The answer the San Jose team settled on was almost defiantly simple, and it borrowed quite openly from a piece of everyday household technology, the record player.

[00:16:22]Instead of storing data on one flat plate or thousands of individual rods, Johnson's team stacked 50 aluminum discs, each 24 in in diameter, on a single rotating spindle, spinning together at,200 revolutions per minute.

[00:16:37]Each disc was coated with iron oxide paint, the same basic magnetic material used in recording tape. And each surface could hold thousands of tiny magnetized spots representing data arranged in concentric tracks the way grooves circle a vinyl record. The genuinely novel part wasn't the stack of discs. It was the arm. A single mechanical arm fitted with read and write heads could move in two directions at once. Up and down the spindle to select which disc to read and in and out along that disc surface to select which track held the record it needed. Instead of reading through everything in order, the ARM simply moved directly to the coordinates where the requested data lived and read it on the spot. On average, that whole operation, find the disc, find the track, read the record, took about 600 milliseconds. What had taken Welter Hedges the better part of an hour, now took roughly the time it takes to blink twice. Think of it less like a filing cabinet and more like a jukebox. A jukebox doesn't play every record in its rack in order to reach the one a customer picked. A mechanical arm swings directly to the selected disc and drops the needle. Johnson's team built essentially the same idea at industrial scale and turned to a serious purpose.

[00:17:50]50 discs instead of a few dozen arranged so a single arm could reach any one of a 100 magnetic surfaces. Find any one of thousands of tracks on that surface and read back an exact answer before a customer standing at a counter had finished asking the question. The completed system called the IBM 350 disc storage unit was the storage half of a larger machine IBM named the 305 Raymac random access method of accounting and control. Announced on September 13th, 1956, the disc unit alone weighed over a ton, stood roughly the size of two refrigerators pushed together, and had to be moved by forklift and shipped by cargo plane. Its 50 spinning discs held 5 million characters of data. The equivalent IBM like to point out of 64,000 punched cards, all reachable in under a second without a single card ever needing to be fed through a reader.

[00:18:42]None of the individual pieces, spinning platters, a magnetic coating, a movable arm was exotic by 1956 engineering standards. What made the Rayax succeed where rods, strips, and tape had failed wasn't a single expensive innovation. It was a cheap mechanical idea borrowed from a record player applied at industrial scale. The team's real engineering achievement wasn't inventing a new physical principle. It was refusing to give up on an old familiar one, a needle finding its place on a spinning surface until it worked reliably enough, 50 layers deep, to run a business on. A company could lease the entire Rayac 305 system, disc included, for roughly $3,200 a month. A serious sum for a mid1950s business, but a price built around a machine that could be sold and installed nationwide rather than a one-off laboratory prototype.

[00:19:33]That distinction mattered enormously. A laboratory curiosity proves a concept. A leasable, serviceable, repeatable product changes an industry. IBM's manufacturing arm had to learn to produce these disc stacks to a consistent tolerance across dozens, then hundreds of units. The same discipline the company had spent decades applying to punch cards. Now aimed at a completely different kind of storage medium, IBM shipped the first production unit to the Zeerbach Paper Company in San Francisco in June 1956, and within a year, similar units were running inventory operations for Chrysler's Mopar Parts Division. Fittingly, the exact kind of parts inventory problem that had opened the story. A clerk at Mopar asking how many of a given part sat in a given warehouse no longer needed to wait for a deck of cards to finish its slow crawl through a reader.

[00:20:19]The arms simply moved to where the answer already was. The sequential era wasn't just being challenged. It was being replaced one lease agreement at a time. Once the RAM proved that random access disc storage could work commercially, its influence spread far beyond warehouse inventory counts. More than a thousand RAM systems were built before production ended in 1961 and IBM followed with steadily improved successors. The 1301 in 1961, the 1311 in 1962. Each generation shrinking the physical size of disc storage while multiplying its capacity. Approximate figures, sea sourcing flags. The applications multiplied just as fast. By 1960, a RAMX system was being used to track athlete standings in real time at the VI Olympic Winter Games near Lake Tahoe, California, spitting out up to the second results for a live international audience. A task that would have been unthinkable for a sequential punched card system working through a full deck between updates.

[00:21:17]airline reservation systems, banking, and eventually the relational databases that underpin nearly every modern piece of software all trace their architectural lineage back to the same basic principle Johnson's team proved in 1956, that a machine could reach directly for one piece of information instead of reading past everything else to find it. Reynold Johnson went on to become an IBM fellow, the company's highest technical honor, and continued inventing for IBM until his retirement in 1971. In 1986, he received the National Medal of Technology, one of the highest honors the United States government gives an engineer. Lewis Stevens and the rest of the San Jose team largely returned to the kind of anonymous corporate life that had defined their careers before RAM.

[00:21:59]Engineers who had solved a nationwide problem and then went back to solving the next one. The economic scale of what they started is difficult to overstate.

[00:22:08]The same San Jose facility that built the Rayax disc stacks by hand in 1956 became the anchor of a magnetic storage industry that within a few decades would be worth tens of billions of dollars annually, employing thousands of engineers across dozens of companies that hadn't existed when Johnson's team started tinkering with rods and strips.

[00:22:29]Approximate figure, sea sourcing flags.

[00:22:32]Every hard drive that followed, smaller, faster, cheaper, by orders of magnitude, was still at its core doing what the Ray did first, spinning a magnetic surface and moving an arm to exactly the point where the answer was waiting. The Ray itself, meanwhile, quietly disappeared from public memory almost as fast as it had disappeared from IBM's product line.

[00:22:54]The 305 was formally withdrawn from the market in 1969, made obsolete by machines that carried its own core idea forward at higher speed and smaller size. Few of the business people who benefited from same-day inventory checks or instant account lookups in the decades that followed ever knew the name Rayac, let alone the name Rainald Johnson. The spinning disc they were using inside a bank's mainframe, inside an airlines reservation desk, eventually inside a personal computer sitting on a desk, had simply become part of the invisible furniture of modern business, indistinguishable from air conditioning or electric light. That in the end is the pattern this kind of engineering follows every time. It doesn't announce itself as revolutionary while it's happening. It solves one clerk's Tuesday morning, then anothers, then a company's, then an entire economy's until the weight it eliminated is no longer even remembered as having existed. The best engineering becomes invisible. Thanks for watching and don't forget like and subscribe.