> ## Content Index > Fetch the complete content index at: https://huizhou92.com/llms.txt > Use this file to discover other available public pages before exploring further. # The Operating System That Ran Until 2000 — And Is Still Called a Failure - URL: https://huizhou92.com/multics-failure-that-spawned-unix-en/ - Published: 2026-10-23T02:26:22.000Z - Updated: 2026-10-23T02:26:22.000Z - Description: Multics is remembered as the bloated failure that gave us Unix. In reality, it ran for thirty-three years and earned the Pentagon's highest security rating. - Author: huizhou92 - Tags: #lang-en, Technology, Operating Systems, Unix, Computer History, Software Engineering ### Bell Labs Walked Away, Unix Grew from Its Wreckage, and the Original System Quietly Worked for Thirty-Three Years On October 30, 2000, inside a defense facility in Halifax, Nova Scotia, technicians for the Canadian Department of National Defense pulled the main breaker on an aging mainframe. It was the final native Multics installation on Earth powering down for good. That was the year the software industry held its breath over the Y2K bug, terrified that mission-critical infrastructure written in the 1990s would collapse at the date rollover. Meanwhile, inside that Canadian coastal base, an operating system whose core architectural blueprint had been drafted in **1964** quietly completed its final shift. From the arrival of the prototype in 1967, Multics had served on the front lines of industrial and military computing for thirty-three continuous years. Yet in the popular mythology of computer science, Multics is almost universally paraded as a cautionary tale: an over-engineered dinosaur, an academic exercise in runaway ambition, and the bloated wreck from whose ashes the nimble, elegant Unix was born. It makes for a satisfying morality play. But it obscures two foundational historical facts: Multics never failed as an engineering system, and the birth of Unix was neither a planned revolution nor an act of ideological defiance. ### The Computing Power Grid The project began in 1964 across the Charles River in Cambridge, Massachusetts. Fernando Corbató, a computing pioneer at the Massachusetts Institute of Technology, joined forces with General Electric and Bell Telephone Laboratories to launch Project MAC. At the time, the world interacted with computers through batch processing: programmers stacked punch cards into trays, handed them through a raised-floor service window, and waited hours—sometimes overnight—for printers to cough out results. A single misplaced comma meant an entire day went straight into the wastebasket. Corbató envisioned an absolute paradigm shift: **the computing utility**. In his vision, computing power should not be an expensive machine hoarded in a basement, but a municipal utility like water or electricity. A central plant would hum in the heart of a city, copper lines would branch into every office, and engineers at teletype terminals would tap into an inexhaustible stream of processing power whenever they needed it. To allow hundreds of remote users to type, compile, and execute code simultaneously without stepping on one another, the Project MAC team invented the architectural foundations that still anchor modern operating systems: They were the first to organize a flat storage medium into a **hierarchical tree-structured directory**, introducing the parent-child folder model every computer user navigates today. They invented the **Shell**, divorcing user-facing command interpretation from the delicate internal mechanics of the kernel. They pioneered **single-level storage through memory-mapped files**, eliminating the need for programmers to write explicit disk buffer routines; the operating system mapped disk files directly into the virtual address space. Most aggressively, General Electric built a custom mainframe, the **GE-645**, designed specifically around hardware-enforced **protection rings**. Eight concentric rings of privilege radiated outward: the microkernel sat in Ring 0, operating system services occupied the middle rings, and user programs ran on the perimeter. Any unauthorized attempt by a user process to read kernel memory was intercepted and aborted directly by CPU circuitry. ![General Electric GE-645 Mainframe, 1968](https://upload.wikimedia.org/wikipedia/commons/1/16/GE-645-mainframe-1968.png) *The General Electric GE-645 mainframe, custom-engineered for Multics and delivered to MIT in 1967\. Photo: Dóeltenga, CC BY-SA 4.0, via Wikimedia Commons* It was a brilliant blueprint for an electric power grid. But on day one of construction, reality struck. In an era when operating systems were handcrafted in raw assembly, the team made an audacious gamble: they decided to write the entire Multics kernel in an unproven high-level language called **PL/I**. The problem: a production-grade PL/I compiler did not exist. MIT graduate students had to write the compiler at the same time they were writing kernel modules. The compiled codebase ballooned unpredictably, and recompiling the system consumed hours of expensive mainframe time. At commercial time-sharing rates, every test compile burned hundreds of dollars. The GE-645 came with only a few hundred thousand words of physical memory. Multics ran on it like a behemoth wading through wet concrete. ### Bell Labs Stops Paying, GE Exits By 1969, five years had slipped past. Executive leadership at Bell Labs grew restless. As an industrial research enterprise accountable for shareholder returns, management saw no power grid—only an open-ended financial hemorrhage. Budgets were blown, hardware shipments were repeatedly stalled, and an operational system inside Bell Labs remained out of reach. In April 1969, Bell Labs announced it was cutting its losses and pulling out of Multics. The following year, General Electric’s board threw in the towel as well. Bruised by persistent computer division losses, GE exited the computer industry entirely, dumping its hardware manufacturing business into the hands of Honeywell. With two of its three institutional pillars gone, the tech industry assumed Multics was dead on arrival. ### A Bootleg Game on a Scrapped PDP-7 When Bell Labs withdrew, the GE-645 was unbolted from the floor and carted away. Overnight, a group of world-class computer researchers lost their interactive time-sharing environment. Among them was a thirty-year-old programmer named Ken Thompson. While working on Multics, Thompson had coded an orbital mechanics game called *Space Travel* on the GE-645\. With the mainframe gone, he tried running his simulation on a commercial GE-635 batch machine. He quickly discovered the computer center billed **$75 per run** in corporate compute time just to land an Apollo capsule on a jerky, batch-processed terminal screen. Refusing to abandon the game, Thompson scouted the shadowy corridors of the Murray Hill facility until he found an unallocated, dust-covered minicomputer: a **DEC PDP-7**. ![DEC PDP-7 Minicomputer](https://upload.wikimedia.org/wikipedia/commons/b/bb/Pdp-7-oslo-2004.jpeg) *The DEC PDP-7 minicomputer. Ken Thompson and Dennis Ritchie bootstrapped Unix on a discarded 8K-word machine. Photo: Matias Årje, CC BY-SA 3.0, via Wikimedia Commons* The machine was primitive by every metric: its core memory held just **8,192 18-bit words**, its display was a crude vector oscilloscope, and it possessed no operating system whatsoever. In the summer of 1969, while his wife and young son were away visiting family in California, Thompson enjoyed a month of uninterrupted solitude inside the laboratory. He recruited Dennis Ritchie and Rudd Canaday. They had no formal project authorization, no management committee, and no line-item budget. Their objective was singular: handcraft a tiny kernel in assembly language on that cast-off PDP-7, solely so *Space Travel* could run at smooth frame rates. They stripped down the carcass of Multics to salvage what was useful: they kept the hierarchical file directory, they kept the decoupled Shell, and then they threw out the rest. Gone were the eight nested protection rings, gone was the bloated PL/I runtime, and gone was single-level virtual memory mapping. Where Multics modeled every entity as an intricate, structured object, Thompson applied a brutally reductive simplification: **every resource, device, and file was flattened into an unstructured stream of bytes**. Colleague Brian Kernighan observed the modest, single-user system and coined a pun: **UNICS** (Uniplexed Information and Computing Service), intentionally lampooning the complex, multi-user multiplexing of Multics. Over time, the spelling hardened into **UNIX**. ![Ken Thompson and Dennis Ritchie, 1973](https://upload.wikimedia.org/wikipedia/commons/1/1b/Ken_Thompson_and_Dennis_Ritchie--1973.jpg) *Ken Thompson (left) and Dennis Ritchie in front of a PDP-11 in 1973\. Photo: Bell Labs, Public Domain, via Wikimedia Commons* The divergence in corporate capital was stark: Multics required a tripartite multinational alliance, millions in federal research grants, and a bespoke GE-645; Unix required a discarded minicomputer that nobody wanted, and a programmer looking for a game. ### Who Actually Failed? Popular historical retellings stop right here, packaging the story as a victory of minimalist agility over corporate bureaucracy. Yet inside the actual industrial sector, the narrative took an unexpected turn. Honeywell, having acquired GE's computer business, did not abandon Multics. Partnering with Corbató's remaining engineers at MIT, they upgraded the hardware architecture into the **Honeywell 6180**, embedding full protection rings into high-speed silicon. In 1973, Honeywell launched commercial Multics installations. It was not pitched to small consumer software outfits. A complete Multics operational center commanded a price tag running into **millions of dollars**. Yet the Pentagon, the United States Air Force, General Dynamics, Ford Motor Company, and major French banking conglomerates queued up with checkbooks open. Throughout the 1970s and 1980s, roughly **eighty premier computing installations** worldwide ran on Multics. They formed the operational nervous system for the Western alliance: from nuclear ballistic simulations and tactical air command to corporate payrolls spanning tens of thousands of employees. Why did these institutions spend millions on Multics when Unix could be obtained on tape for nominal administrative fees? The answer came down to one imperative: **security**. While system panics plagued early Unix, had no internal memory isolation, and allowed anyone who guessed a root password to seize total control, Multics delivered uncompromising structural defense. In 1974, the Air Force contracted Honeywell to harden the Multics kernel. The team implemented Mandatory Access Control (MAC) and formal multi-level security models. In 1985, the National Computer Security Center published the Trusted Computer System Evaluation Criteria, known throughout the industry as the Orange Book. Under this standard, the overwhelming majority of commercial enterprise systems—including versions of Windows NT released two decades later—scraped by with a baseline C2 security rating. At the apex of the Orange Book pyramid, a specialized system derived directly from Multics, designated STOP 2.1 / SCOMP, was awarded the highest possible rating: **Class A1 (Formal Mathematical Verification)**. ![Multics Command-Line Login Screen](https://upload.wikimedia.org/wikipedia/commons/d/d6/Multics-Login.png) *The Multics command-line login banner. The operating system remained in active production from 1967 to 2000\. Photo: MIT, CC BY-SA 3.0, via Wikimedia Commons* Its core kernel mechanisms were mathematically proven to prevent privilege escalation at the gate level. That architectural bloodline evolved directly into the military-grade XTS-400 operating system, securing classified intelligence gateways long into the twenty-first century. ### Defeated by Economics, Not Architecture Why did Multics eventually fade from public memory? It was not undone by poor engineering; the unforgiving economics of commodity hardware eclipsed it: **Multics was chained to proprietary iron**. To run Multics, an organization had to purchase a multimillion-dollar, multi-cabinet mainframe from Honeywell. Unix, by contrast, was rewritten by Dennis Ritchie in **C** in 1973\. In 1977, it was ported to an Interdata 7/32, decoupling operating software from specific CPU architectures. C operated as an architectural universal solvent. Whenever a semiconductor fabricator produced an inexpensive new microprocessor, compiling a C runtime allowed Unix to run on it almost immediately. When Moore's Law slashed workstation prices from hundreds of thousands of dollars to thousands, open commodity hardware crushed proprietary mainframes. The tide of cheap, networked microcomputers washed over the computing landscape, leaving bespoke mainframes behind as expensive museum pieces. In 2006, French computing giant Bull SAS released the complete Multics source code under a free software license. When contemporary systems programmers combed through the thousands of lines of historic PL/I code, they discovered an astonishing truth: Memory segmentation, dynamically linked libraries (`.so` and `.dll`), memory-mapped I/O, hierarchical access control lists (ACLs), and granular multi-level security—the hallmarks of modern computing—had all been worked out and deployed to production decades earlier. The young rebels who walked away from Project MAC built a pragmatic, loose, ubiquitous Unix world; Meanwhile, Fernando Corbató and those who stayed behind guarded an immaculate computing utility until the turn of the millennium. History stamped it with a casual label of failure; Yet on the front lines of defense and research, the machine showed up to work, every single day, for thirty-three years.