The IBM Mainframe Era: The Computer That Ran the World

Zusammenfassung

For three decades, International Business Machines Corporation did not just dominate the computer industry — it was the computer industry. IBM’s System/360 announcement in April 1964 was the most consequential business decision in the history of computing: a $5 billion gamble on a single, unified architecture that could span machines from a $30,000 departmental computer to a $30 million scientific processor. It standardized the industry, created the software business as a separate commercial activity, and established the template for platform competition that technology companies still follow. Understanding the mainframe era is understanding how computing became infrastructure — the utility that organizations could not function without, which made whoever controlled it extraordinarily powerful.

Before System/360: A Tower of Babel

In the early 1960s, IBM sold a range of mutually incompatible computers: the 1401 for business data processing, the 7090 for scientific computing, the 7080 for commercial workloads. A program written for one machine ran on nothing else. When a customer upgraded, programmers rewrote everything. Peripheral devices — tape drives, printers, card readers — were incompatible between product lines. IBM serviced each machine with separate teams of engineers trained on that specific hardware.

This was not unique to IBM. The broader industry — the BUNCH (Burroughs, UNIVAC, NCR, Control Data, Honeywell) — each had their own incompatible product lines. The computer industry of 1963 resembled the early automobile industry before standardized parts: capable and growing, but requiring enormous duplication of effort.

The System/360 Gamble

Thomas Watson Jr., IBM’s CEO, approved a project in January 1961 that internal documents called “the most difficult decision I ever made.” The goal was a single instruction set architecture — System/360 — that would run across an entire range of machines, from the smallest to the most powerful. A program compiled for the smallest System/360 would run without modification on the largest.

The architectural work was led by Gene Amdahl (who would later be best known for Amdahl’s Law, the theoretical limit on parallel speedup) and Fred Brooks, who would later write The Mythical Man-Month based on his management experience leading the OS/360 software project. The System/360 was announced April 7, 1964. It was simultaneously announced in the United States, Canada, Britain, France, West Germany, Italy, Japan, Switzerland, and the Netherlands — a global launch unprecedented in the industry.

The financial risk was staggering. IBM invested $5 billion in System/360 development over four years — roughly twice its annual revenue at the time (about $3.2 billion in 1964) — building new factories and hiring tens of thousands of new employees. If the architecture had failed technically or the market had rejected the compatibility promise, IBM would have been destroyed.

The 360 in System/360

The name referred to 360 degrees — a full circle, suggesting comprehensive coverage of all computing needs from scientific to commercial to real-time. The architecture introduced 8-bit bytes as a standard unit (before the 360, byte sizes varied by machine), a unified I/O interface for peripheral devices, and a memory model with 24-bit addressing (later extended to 31-bit and then 64-bit in successor architectures). The 360’s memory addressing model shaped every subsequent IBM architecture and influenced processor design broadly.

OS/360: Software as the Hard Problem

The System/360 hardware shipped roughly on schedule. The software — OS/360 — was a disaster that Fred Brooks would document exhaustively in The Mythical Man-Month (1975). The project employed thousands of programmers, cost far more than budgeted, delivered far later than promised, and when it arrived, contained thousands of known bugs. OS/360’s difficulties demonstrated something the industry did not yet understand: software complexity does not scale linearly with software size. Adding programmers to a late software project makes it later. These observations became “Brooks’ Law” and remain among the most durable empirical findings in software engineering.

The OS/360 experience also revealed that software was a distinct discipline from hardware engineering, requiring different management approaches, different metrics of progress, and different skill sets. The mismatch between how IBM’s hardware-focused management thought about software and what software actually required was the first large-scale instance of a problem the industry would replicate for decades.

IBM Dominance and the BUNCH

Through the late 1960s, IBM held approximately 70% of the global computer market by revenue. The industry shorthand for its competitors was BUNCH: Burroughs, UNIVAC, NCR, Control Data, and Honeywell.

Each company survived by carving out specializations that IBM did not dominate:

Burroughs (later merged with UNIVAC to form Unisys in 1986) focused on financial computing. Its machines ran the banking systems of major American banks and the transaction processing for the Federal Reserve’s check-clearing system. The Burroughs architecture used a stack-based processor design that was theoretically elegant and practically entrenched in its customer base.

Control Data Corporation (CDC), under William Norris, targeted scientific and technical computing with the CDC 6600 (1964) — designed by Seymour Cray — which was the fastest computer in the world at its introduction, running at roughly 3 MFLOPS, about three times the speed of the previous record-holder, IBM’s 7030 Stretch. IBM’s response to the CDC 6600 was documented in a memo by IBM president Thomas Watson Jr.: “I understand that in the laboratory developing the 6600 there are only 34 people including the janitor. Of these, 14 are engineers and 4 are programmers, and only one person has a PhD … Contrasting this modest effort with our ‘vast development activities,’ I fail to understand why we have lost our industry leadership position by letting someone else offer the world’s most powerful computer.” Control Data later sued IBM for anticompetitive practices; IBM settled for $80 million in cash and a ten-year computer service contract.

NCR (National Cash Register) dominated retail point-of-sale terminals and later ATM machines. Its hardware was optimized for transaction throughput in high-volume retail environments. NCR shipped its first ATM, the NCR 770, in 1977, and its NCR 5070 (1983) became the dominant ATM platform for a generation of banks.

Honeywell targeted medium-size business computing with machines that competed on price against IBM’s mid-range products. Honeywell acquired GE’s computing division in 1970, briefly making it IBM’s largest competitor in absolute revenue terms before the combined company failed to integrate the acquisition and began its long decline.

None of these companies competed with IBM across its full range. All of them were profitable. IBM’s dominance coexisted with a viable competitive ecosystem as long as IBM controlled compatibility.

The Antitrust Case and the Birth of the Software Industry

The U.S. Department of Justice filed an antitrust lawsuit against IBM in January 1969, alleging monopolization of the general-purpose computer market. The complaint cited IBM’s practice of bundling software and services with hardware at no additional charge — making it impossible for independent software companies to compete, since IBM’s “free” software undercut any price a competitor could offer.

IBM’s pre-emptive response, in anticipation of the litigation, was unbundling: beginning in January 1969, IBM began pricing software separately from hardware. Programs that had been included with the machine now had explicit price tags. Field engineering services were billed separately from equipment.

This single decision — made defensively, to demonstrate competitive rather than monopolistic intent — created the commercial software industry. When IBM charged for software, the market established that software had value. Independent software vendors could now sell programs for IBM machines without competing against “free.” Companies like Computer Sciences Corporation, Informatics, and Applied Data Research that had existed at the margins of IBM’s ecosystem could now build businesses on IBM compatibility.

The pattern would recur: when Microsoft licensed its BASIC interpreter to PC manufacturers in 1980, when Oracle sold databases for IBM minicomputers in the early 1980s, when SAP built enterprise resource planning systems on IBM mainframe and later x86 hardware — all of this was downstream of the 1969 unbundling decision. The software industry that produced Microsoft, Oracle, SAP, Salesforce, and every subsequent SaaS company descends directly from IBM’s decision to separate software pricing from hardware pricing.

The antitrust case itself became one of the longest and most expensive proceedings in American legal history: IBM produced millions of documents, consumed hundreds of millions of dollars in legal costs, and distracted management attention for thirteen years. The Reagan administration dropped it in 1982 as “without merit.” The irony was complete: the case that IBM spent a decade defending had already produced the outcome the plaintiffs wanted — a competitive software market — before the case concluded.

The z-Series: Unexpected Immortality

System/360 begat System/370 (1970), which begat System/390 (1990), which begat the z-Series (2000). The basic architecture introduced in 1964 — with extensions to address space, instruction set, I/O, and virtualization — remains in production as of 2024. IBM z-Series mainframes run approximately 68% of the world’s production workloads by transaction volume. Every ATM withdrawal, every SWIFT bank transfer, most airline reservations, and most large-scale insurance processing runs on hardware that is architecturally descended from the 1964 announcement.

This is not nostalgia or inertia alone. Mainframes offer capabilities that distributed systems struggle to match: hardware-enforced transaction integrity across billions of transactions per day, physical security (a z-Series machine processes encryption in tamper-resistant hardware modules), and reliability measured in minutes of downtime per year. The cost of migration away from mainframes for institutions that have been running IBM architectures since the 1970s is astronomical — not because change is impossible, but because the software systems are correct, and correctness is hard to rebuild.

Dead End: Plug-Compatible Mainframes

Gene Amdahl left IBM in 1970 after disagreements over the 370 architecture and founded Amdahl Corporation, which in 1975 shipped a mainframe that was fully compatible with IBM’s System/370 instruction set but cheaper and faster. IBM’s response — aggressive price cuts, lease arrangements that penalized customers for switching — became the subject of private antitrust litigation by Amdahl and others. Plug-compatible manufacturers (PCMs) including Hitachi, Fujitsu, and Amdahl competed vigorously through the 1980s. When IBM shifted from mainframes to RISC workstations (the RS/6000 line, 1990) and lost its technical performance lead, the PCMs lost their model: you cannot undercut on price if you cannot match on performance. Amdahl Corporation was acquired by Fujitsu in pieces through the 1990s and dissolved in 2000. The plug-compatible strategy worked as long as IBM led technically. Once IBM stumbled, there was no architecture left to be compatible with.