IEEE Spectrum breed · Technologie

IBM Built the Cold War’s Most Powerful Code Breaker for the NSA

During the Cold War, the NSA operated a highly secret and powerful cryptographic processor called Harvest, designed and built by IBM. This machine, operational from 1962 through 1975, was significantly faster than any other computer of its time and served as the NSA's main cryptographic engine during a critical period of global tension.

Harvest was a pioneering machine for processing massive datasets in real-time, predating modern computers used for continuous video streams and security systems. It was also one of the first specialized add-on computers, designed to excel at a single task, similar to how modern CPUs utilize graphics chips for intensive processing.

The computer's immense processing power required a constant flow of data, leading to the development of the world's first automated tape library, named Tractor, capable of robotically retrieving magnetic tapes from storage racks.

To manage its complex operations, IBM developed a custom programming language called Alpha, allowing code breakers to precisely define cryptographic problems. This system, housed at an NSA data center in Fort Meade, Maryland, was a testament to how national security needs pushed the boundaries of computer technology.

Harvest, officially known as the IBM 7950, was paired with the IBM 7030 Stretch mainframe. Stretch handled general computing and input/output, including the Tractor tape library, while Harvest's custom coprocessor focused on sifting through alphanumeric characters at extreme speeds.

Harvest's coprocessor was a streaming computer that executed a single, fixed sequence of steps on data as it streamed past, a departure from general-purpose systems. It operated in bursts, suspending Stretch while it processed data in shared memory at high speeds.

Both Harvest and Stretch were among the first large computers built entirely from transistors, utilizing a pipeline architecture where data was processed in overlapping stages, allowing for simultaneous operations like fetching and comparing data.

The coprocessor fetched two data streams (P and Q), performed operations, and wrote results as a third stream (R). Its memory was bit-addressable, offering flexibility in data handling. The system included a logic unit for bitwise operations and a table-lookup unit that functioned similarly to cipher machines of the era.

Despite its complexity, Harvest demonstrated remarkable efficiency. One documented job involved searching 3.5 billion characters for 7,000 target terms in under four hours. Another task, 'Moretown,' processed 11 million messages against 8,000 search terms in about ten hours.

The Tractor tape system, designed to support Harvest, stored data on cassette-like cartridges, each holding about 120 megabytes. At launch, the system offered 44 gigabytes of online storage, significantly more than contemporary disk systems.

Tractor operated continuously, with a robotic mechanism fetching and mounting cassettes for Harvest's data needs. The system's tapes moved at 6 meters per second, and software managed the cassette shuttling and data reading/writing.

The Alpha programming language, developed jointly by IBM and NSA, was specifically for Harvest's dataflow engine. It allowed programmers to define character sets and included unique characters representing unknown or null values, providing flexibility for code-breaking tasks.

By 1971, Harvest was operating at peak utilization but was beginning to be replaced by newer, general-purpose machines for routine tasks, leaving Harvest to focus on its specialized, large-scale jobs. It was decommissioned in 1976 when a critical custom component in the Tractor tape library failed, and its manufacturer was out of business.

Harvest was ultimately replaced by the Cray-1 supercomputer. Although Harvest didn't have direct successors due to its secrecy, its pioneering ideas in automated storage, pipeline architecture, and pattern-detecting logic influenced later computing technologies.

AI-samenvatting op basis van de bron.