On the top of the Chilly Conflict, one very specialised laptop was so secret that the world didn’t realize it existed. It ran its jobs as much as 200 occasions as quick as every other laptop of its time. It was the U.S. National Security Agency’s most important cryptographic processor in operation from the time of the Cuban Missile Crisis in 1962 via the Vietnam War and on previous the 1975 Helsinki Accords. The machine stopped operating solely when its transferring elements lastly gave out.
The Harvest laptop mattered due to what it was in addition to when it ran. For 14 years, it was the engine processing the NSA’s most delicate intercepts at a time when signals intelligence was as near a strategic weapon as something in need of a warhead.
Designed and constructed by IBM for the NSA, Harvest was one of many first machines designed to use operations to huge datasets dashing previous, a precursor to the computer systems right this moment that handle steady video streams and security systems in actual time. It was additionally one of many first machines built as an add-on—a specialised helper supposed to do one job exceptionally properly, bolted onto a common laptop. Harvest’s modular design is sort of a Nineteen Sixties model of right this moment’s graphics chips that CPUs use to run intensive video-game and AI processing hundreds.
All that uncooked processing energy meant that Harvest additionally wanted nonstop rivers of knowledge to run on. And that led to a different pioneering achievement: the world’s first automated tape library that might robotically fetch any considered one of tons of of huge cassettes of magnetic tape from the machine’s racks.
Given Harvest’s unprecedented processing and storage capability, the machine’s designers naturally wanted to rethink how their system dealt with data. So IBM wrote a custom-made programming language referred to as Alpha to let code breakers rigorously describe cryptographic issues, simply as scientists on the time had been utilizing the emerging language Fortran to explain equations and data-processing algorithms.
In Fort Meade, Md., an NSA knowledge heart hosted one of many world’s quickest computer systems of its time—though not usually mentioned, due to its delicate, high-security code breaking and cipher looking work. Nationwide Cryptologic Museum
The story of Harvest, pieced collectively from declassified documents and contemporary manuals and technical overviews, gives a brand new and sudden vista on the history of computing. It additionally presents a case examine in how nationwide safety wants, particularly in the course of the Chilly Conflict, pushed laptop know-how past the far reaches of what unclassified, civilian computing might obtain. Harvest’s distinctive historical past reveals a visionary algorithmic, coding, reminiscence, and {hardware} structure often many years forward of its time. However this machine was additionally constructed solely as soon as, for one singular objective, after which in the end quietly retired.
The Coronary heart of NSA’s Secret Machine
IBM’s landmark 1960 transistorized mainframe, the IBM 7030, higher often known as Stretch, offered the entrance finish for Harvest (which was formally often known as the IBM 7950). IBM delivered Stretch to eight or nine customers, principally scientific analysis labs, from 1961 via ’63. Designed and prototyped throughout the second half of the 1950s, Stretch launched the now commonplace notion of an 8-bit byte. For its first three years of operation, Stretch was the non-classified world’s quickest laptop, though it failed to satisfy IBM’s aggressive purpose of operating 100 occasions as quick as Stretch’s predecessor, the IBM 704. Whereas IBM engineers in Poughkeepsie, N.Y., had been designing and constructing Stretch, the corporate was additionally quietly discussing a brand new system that may be constructed for NSA.
On the time, NSA’s present cryptanalytic computer systems—giant, batch-processing machines that required human operators to manually stage every tape run—had been struggling to maintain tempo with the sheer quantity of intercepted message site visitors coming in from across the globe. What the company wanted was a machine that might course of an unbroken river of incoming knowledge, mechanically, across the clock. That requirement alone profoundly formed Harvest’s design.
IBM’s Harvest system, custom-built for the NSA for code breaking, paired the IBM 7030 Stretch mainframe with a bespoke data-stream processor. Stretch dealt with peculiar computing and enter/output, together with the Tractor automated tape library. Each items shared two sorts of reminiscence: a big most important financial institution and a smaller, quicker financial institution. When Stretch switched to streaming mode, Harvest drew two streams of knowledge, P and Q, from reminiscence, processed them in parallel, and returned the outcomes as a 3rd stream, referred to as R. Chris Philpot
After two failed proposals to NSA, in 1958 IBM lastly landed the contract: a Stretch-based machine, augmented by a {custom} coprocessor, with a revolutionary tape-based storage system, referred to as Tractor.
Stretch’s forte was floating-point math for scientific computations. IBM had designed it primarily for labs engaged on frontier analysis like nuclear weapons design and weather prediction. Against this, the {custom} coprocessor to be constructed atop Stretch would assist NSA analysts sift via alphanumeric characters—that’s, basically integer knowledge.
Harvest’s coprocessor was the other of a general-purpose system. It was, moderately, a streaming computer. As a substitute of executing lengthy sequence of directions, it adopted one mounted sequence of steps and utilized that very same sequence to each pair of characters as they streamed previous. Harvest shared reminiscence with the main Stretch processor and ran in bursts. Both Stretch was working, or else it suspended itself whereas Harvest’s coprocessor shot via knowledge in reminiscence at excessive speeds.
Stretch and Harvest had been among the many first giant computer systems constructed fully from transistors packaged in circuit playing cards and housed in giant, refrigerator-size frames. A 1962 technical manual about Stretch describes the machine’s CPU as divided into practical sections—the instruction unit, the look-ahead unit, the (parallel and serial) arithmetic unit, and the reminiscence bus unit. Harvest inherited Stretch’s fundamental circuit design however then added one thing unconventional: Its streaming items processed knowledge in overlapping levels referred to as a pipeline. So whereas one pair of knowledge bytes was being in contrast, the following pair was being fetched from reminiscence.
Harvest’s coprocessor operated by fetching two streams of knowledge, referred to as P and Q, from the system’s reminiscence, performing operations on them, then writing the outcomes to reminiscence as a 3rd stream, R. Every stream could possibly be anyplace from 1 to eight bits large. Harvest’s reminiscence was bit-addressable, which means phrase boundaries could possibly be ignored fully. As an illustration, it might fetch simply 5 bits moderately than filling out a complete byte. Streams P, Q, and R included versatile provisions for looping and addressing knowledge in advanced patterns—permitting, for instance, repeated fetching of quick strings from reminiscence.
Information from P and Q fed into two practical items. The easier was the logic unit, which carried out fundamental, bitwise operations—the identical operations any programmer would acknowledge right this moment—and wrote its outcomes again to reminiscence. The extra advanced was a table-lookup unit. It mixed incoming knowledge from P and Q to type an tackle in reminiscence, which might then be used to advance a counter by one, set a particular bit, or retrieve a saved worth. The latter unit functioned, in impact, just like the rotor wheel inside a cipher-encoding/decoding machine of the period, the sort that electronically substituted one worth for an additional based on the cipher machine’s wiring.
Harvest’s complexity baffled some on the NSA. Throughout worker excursions, based on James Bamford’s 2001 NSA historical past, Body of Secrets (Doubleday), officers would level to the machine and scoff, “It’s lovely, but it surely doesn’t work.”
Not everybody on the company was postpone by the monumental system, nevertheless. One of many few documented examples of Harvest at work, recounted by Bamford, describes the machine looking 3.5 billion characters of textual content for any of seven,000 goal phrases, in just below 4 hours.
In unclassified remarks from 1972, NSA analyst Robert Looney mentions one job Harvest had tackled—although he didn’t specify the tip purpose or the code-breaking effort behind it. Codenamed “Moretown,” the job concerned sifting via 11 million messages spanning 16 years of intercepted site visitors in opposition to an inventory of some 8,000 search phrases—all in about ten hours.
IBM’s Frances Allen helped design Alpha, Harvest’s custom-built programming language.IBM
IBM’s James H. Pomerene was chief engineer of Harvest, supervising its custom-designed circuits that’d been optimized for algorithms utilized in many cryptographic jobs.IEEE
IBM’s Fred Brooks Jr. was a key co-architect of Harvest’s {hardware} system. Computer History Museum
As a unified system, Harvest—that’s, Stretch plus IBM’s custom-built streaming processor add-on—streamed 1 byte each 0.3 microseconds, and it boasted about 800 kilobytes of addressable reminiscence.
“Right here you see one financial institution pulled out of its oil tub,” Looney stated in his 1972 remarks celebrating Harvest’s tenth anniversary of operations. He held up a photograph of Harvest’s magnetic core memory banks—six of them, submerged in oil for cooling.
Issue within the time calls for of assorted knowledge fetches from Tractor’s tape archives, and a single Harvest “instruction” typically carried on, with no need any human intervention, for hours.
“It was fairly a tremendous laptop,” recalled IBM Fellow Emerita Frances Allen in a 2001 oral history. “One instruction, for instance, might do types, and do statistical evaluation of the info that was streaming by it.… All the things we had been doing at the moment was on the leading edge. There was no query about it.”
Allen, who obtained the A.M. Turing Award in 2006, was one of many builders who labored on each Stretch and Harvest. On the time she began engaged on Harvest, Allen famous, the Fort Meade, Md.–based mostly NSA was largely unknown exterior of categorised intelligence circles. So she at first assumed she was engaged on an unspecified naval undertaking. “We considered ourselves as working for the Bureau of Ships, as a result of that was the code title for NSA within the finances!” recalled Allen, who died in 2020.
Different key Harvest designers and early builders wound up turning into influential figures over the course of computing historical past. Frederick Brooks Jr., recipient of the 1999 Turing Award and a serious contributor to the {hardware} and software program for IBM’s System/360, additionally helped develop Harvest. And James Pomerene, previous to his involvement with Harvest as its chief engineer, had beforehand helped construct the pioneering IAS computer alongside John von Neumann.
How Tractor Saved a World of Information
IBM constructed the Tractor tape system (IBM 7955) to connect to the identical Stretch machine that hosted Harvest, as a result of no present knowledge storage applied sciences might sustain with the pc’s staggering throughput. Stretch dealt with the enterprise of staging tapes from the library to the drives—utilizing Tractor’s automated cassette handler. Stretch additionally coordinated studying knowledge in from Tractor and writing outcomes again out from Harvest. Harvest, in flip, did all its precise computing on the system’s shared most important reminiscence.
Within the early Nineteen Sixties, and even after Tractor and Harvest had been put in, hard-drive knowledge storage was in its infancy. For code-breaking jobs of the scale Harvest was taking over, disk storage would have been impractical when it comes to each price and sheer flooring house. So Tractor needed to be based mostly round tape storage.
Every tape was sealed inside a case constructed like a boombox—twin encased reels below a window, carried by a deal with—and, at 6 to 7 kilograms, about as heavy as a bowling ball. Consider a Tractor cassette as an outsize predecessor of the audiocassette, which might come alongside a decade later, and holding some 120 megabytes of knowledge on a reel of tape 550 meters lengthy. Every storage unit housed as much as 160 of those cassettes.
An IBM technician holds one of many knowledge cassettes used with Harvest’s automated Tractor tape drives. IBM
When Harvest launched in 1962, it had three automated cartridge items, every serving two drives. So the accessible on-line storage throughout the three Tractor items totaled a surprising 44 gigabytes. That’s greater than 190 occasions as a lot capability because the IBM 2314 disk storage system, introduced in 1965, which held 233 megabytes throughout its full complement of eight drives.
Tractor needed to run repeatedly, swapping cassettes out and in, 24 hours a day, seven days every week. The system’s tape-handling pace was tuned to maintain tempo with Harvest’s personal urge for food for knowledge. The custom-built robotic mechanism for retrieving the cassettes was a servo-driven arm that traversed the system’s storage racks. It fetched a cassette from its slot and delivered it to a handler or obtained a cassette from one of many handlers and returned it to storage.
Working at 6 meters per second, Tractor’s tapes zipped previous the learn/write heads quicker than the attention might observe. Software program operating on Stretch dealt with the cassette shuttling in addition to studying and writing. For considered one of Tractor’s drives to maneuver from the completion of processing one tape to studying the following took about 18 seconds, assuming it had already been fetched and was able to mount. Robotically fetching a cassette from the storage unit and getting ready it for studying required no human dealing with or enter by any means.
Along with Tractor, the system had commonplace reel-to-reel tape drives hooked up to Stretch. Harvest’s technicians usually used the traditional drives for importing and exporting knowledge to and from different methods; there was no different sensible method to get giant datasets into or out of Harvest. Tractor might additionally retailer everlasting information and retrieve them immediately from its tape libraries when a job required them. In different phrases, Tractor’s substantial cassette libraries acted each as everlasting knowledge storage and as a spot to carry transient knowledge for processing by Harvest.
No system within the industrial computing world of 1962 got here near Tractor’s gigabytes of concurrently accessible knowledge. At most laptop facilities on the time, “accessible” knowledge meant bodily racks of tape standing someplace close to its drives—accessible solely as quickly as an operator might manually pull a reel and thread it onto a machine, one by one, over the course of a shift.
Alpha Was Harvest’s Customized-Constructed Programming Language
Created collectively by IBM and NSA, the Alpha language existed solely to program Harvest’s streaming dataflow engine for code-breaking work. In accordance with a declassified Pentagon history of NSA computers, Alpha stood for Superior Language for Programming Harvest.
Alpha allowed the programmer to outline the alphabet by which code-breaking knowledge can be processed. The language additionally included two uncommon characters with no equal in standard computing till years later, when Multics and Unix launched wildcard characters. A “scab” (which was represented on Harvest’s enter keyboard, a repurposed early IBM Selectric typewriter, by a “?”) stood for a personality that was actual however unknown. And a “pad” (represented by a clean house) was a null or spacer. These characters offered flexibility of illustration for code breaking jobs, by which unknown or unsure characters had been commonplace.
A Harvest operator sorts on one of many most important system consoles, a repurposed IBM Selectric typewriter.IBM
The principles governing Alpha’s operations on strings anticipated different fashionable rubrics, like “not a number”—a designation describing an unknown worth in a dataset that may propagate via calculations, moderately than silently corrupting them. Strings in Alpha may be aggregated into cords, and cords into ropes, giving cryptanalysts a hierarchical vocabulary for describing advanced intercepts.
Allen wrote a last technical report on her part of the Harvest software program when her a part of the undertaking concluded—and simply as promptly misplaced entry to it. “I spent the nice a part of a summer time on that,” she recalled in 2001. “And it simply disappeared into Fort Meade someplace.”
Changing an Irreplaceable Machine
By 1971, based on NSA analyst Looney, the machine was operating at its highest utilization ever—115 hours of manufacturing every week, or greater than two-thirds of the time. But the variety of jobs it processed had been dropping since 1967. Unusual data-processing work, Looney famous, was by 1972 migrating to newer, general-purpose machines, leaving Harvest to focus on the very giant, specialised jobs no different system might deal with.
At its tenth anniversary of operations, Looney concluded, Harvest was a machine “conceived within the fifties, born within the sixties, and irreplaceable within the seventies.”
He received the final half mistaken.
On 27 February 1976, operators shut down Harvest for the final time. A {custom} mechanical part within the Tractor tape library had worn out, and the producer of the half was not in enterprise. By then Harvest had run repeatedly for almost a decade and a half—via the roughest shut name within the historical past of mutually assured destruction and into the age of détente—processing intercepts at a price no civilian machine might contact. By the point it retired, Harvest had outlived a number of generations of economic computing.
A placard commemorates the 1976 decommissioning of IBM’s Harvest laptop on the NSA’s headquarters in Fort Meade, Md. Nationwide Cryptologic Museum
Any person on the NSA determined to commemorate the machine with a mock telegram, written below Harvest’s title on the machine’s final day (and now preserved within the company’s archives). “I first started operations at NSA. Though not broadly recognized, I used to be in all probability the most important, quickest, and most technically superior laptop system on the earth,” the telegram stated. “And now, fourteen years later, the time to retire has come. The price of my maintenance and operation has been overtaken by extra fashionable equipments and the newer applied sciences.”
The NSA in the end changed Harvest with the landmark Cray-1 supercomputer. The Cray-1 was constructed from quicker, extra tightly integrated circuits that might outperform Harvest’s growing older transistors at almost any process, together with textual content processing. Though the Cray was designed primarily for numeric and scientific computing, it offered throughout many fields—which in the end made the supercomputer win out as soon as Harvest’s custom-built text-processing {hardware} was not well worth the maintenance for only one buyer.
The secrecy that shrouded Harvest meant it might declare no lineage of quick successors. However the concepts it pioneered didn’t disappear—they resurfaced, many times, within the years that adopted.
Tractor’s automated tape library was the forerunner of the robotic storage silos that may become standard in enterprise data centers about 20 years later. Harvest’s pipeline structure prefigured the dataflow computing movement of the 1980s. The continual pattern-detecting logic of its match items finds direct echoes in fashionable hardware packet-inspection intrusion detectors and programmable network switches that right this moment route site visitors via the web at wire pace.
Harvest didn’t discovered a dynasty. However, in its time, it steadfastly pointed towards the long run—in a number of instructions directly.
This text seems within the September 2026 print situation as “The Lost History of IBM’s Cold–War Code Breaokayer.”
From Your Website Articles
Associated Articles Across the Internet
