In the last hours of December 31, 1999, John Koskinen got on a plane to New York.[1] He was President Clinton's Y2K czar, the official in charge of making sure America's computers did not trip over the new year, and he picked that flight on purpose. He wanted the public to see that the man in charge of the repairs was willing to be in the air at midnight.[1]
Twenty years later he explained the reasoning. "If nobody had done anything," he said, "I wouldn't have taken the flight."[1]
The flight was the last step of a repair job that had taken most of a decade and cost the world hundreds of billions of dollars. It ended in an argument that has never been settled, about whether all that work saved the world from disaster or whether there was never much danger in the first place.
Two digits to save a little memory
The problem began in the 1960s, when computer memory was expensive. To save space, programmers wrote years as two digits, so 1967 became 67.[1][2] Business computers were fed with punch cards that held only 80 columns, and every column counted.[2]
It was a sensible shortcut for machines that would be thrown out in a few years. The trouble was that the programs were not thrown out. IBM's System/360, the business computer that set the industry's standard in the 1960s, spread the habit.[2] Code written then was still running banks, payrolls and utilities in the 1990s.
When the clock went from 99 to 00, a computer could decide it was 1900. Anything that depended on how much time had passed, such as a mortgage calculation, could come out wildly wrong.[1]
The people who saw it coming
Bob Bemer, one of the people behind COBOL, the language much of this business software was written in, had given programmers a way to store four-digit years. Most of them ignored it.[2] He published warnings in 1971 and again in 1979, the second time telling programmers plainly not to drop the first two digits.[2] Later he said that if he and Grace Hopper, COBOL's best-known pioneer, were to blame for anything, it was for making the language so easy that anyone could use it.[2]
Peter de Jager, a young Canadian, ran into the problem in late 1977, in his first job as an IBM computer operator. He asked whether the machine would work in the year 2000. His manager laughed, asked how old he was, and told him it would not matter until he was 45.[2]
De Jager did not let it go. In September 1993 he wrote one of the first warnings about the problem in the trade magazine Computerworld, guessing the fix would cost $50 billion to $70 billion. He later called that guess very conservative.[3] By 1995 he was giving more than 85 talks a year on the subject.[2]
The repair job
Fixing it was simple in principle and miserable in practice. Somebody had to find every place in every program where a date was stored or used, and change it. By early 1999, Time put the amount of code that needed checking at 1.2 trillion lines.[2]
Many of the people who understood the oldest code had retired, so employers went and got them back. In June 1998, with about 570 days left, at least eight US government agencies were trying to fill at least 365 jobs with retirees.[4] The government payroll office in New Orleans, which paid 450,000 federal workers, rehired a 57-year-old programmer named Thomas Copfler at $71,000 a year and let him keep his full pension. The tax agency invited 200 recent retirees to apply for a few dozen temporary jobs. An online job bank for retirees held 10,000 résumés from people who knew COBOL.[4]
The estimates of the total bill grew as the work went on. The research firm Gartner put the worldwide cost at $300 billion to $600 billion.[3] Koskinen's own count, given in January 2000, was about $100 billion in the United States, including $8.5 billion spent by the federal government, and roughly twice that worldwide. He said $200 billion was the lowest figure anyone would give, and that the bigger ones did not match what had actually been spent.[5]
The mood in 1999
Outside the computer rooms, people got nervous. Some Americans stocked up on food, water and guns.[1] The Federal Reserve, in case people rushed to empty their bank accounts, arranged in 1998 to raise its stock of cash from $150 billion to $200 billion.[6]
Washington built a command center for the night, a few blocks from the White House in an old Secret Service building. It cost about $50 million and was expected to hold around 200 people over the New Year weekend, collecting reports on what was failing where.[7] It was there to gather information, not to make decisions.[7]
Even at the time, some of the press thought the reaction had swung too far. Time described the whole affair in January 1999 as decades of neglect followed by too much fuss.[2]
What actually broke
Midnight arrived in one country after another, and the lights stayed on. The problems that did appear were real, mostly small, and sometimes funny.
The most serious one was kept quiet at first. On New Year's Eve, the American ground station that processes data from the country's spy satellites went down for several hours because of a Y2K failure. A backup system came on shortly before midnight Washington time and handled a bit more than half of the incoming data, and some data was lost. The normal system was back by Sunday night.[8]
On the Friday evening, after the system had already broken down, Pentagon officials told reporters there were no problems. Deputy Defense Secretary John Hamre said they had not known about the failure yet. "It was a significant event, but fortunately it had insignificant consequences," he said. It was the first time the Pentagon had publicly admitted a breakdown in its spy satellite system.[8]
The rest of the American list was less dramatic. About 150 slot machines at three Delaware race tracks stopped working for a while. An Amtrak computer that tracked trains failed, and staff tracked them by hand until it was fixed. The US Naval Observatory, the government's official timekeeper, showed the date on its website as January 1, 19100. Its master clock was fine.[9]
Japan had a more nervous night. At the Shika nuclear plant, radiation monitoring failed, along with a system meant to send radiation and plant readings to the government in an emergency. At the Fukushima No. 2 plant, information about the position of the control rods did not reach a monitor, so the company checked the rods another way and kept the plant running. Officials said none of the problems affected power generation.[10]
Real danger or a scare
Afterwards, a lot of people decided Y2K had been a hoax and the repair a waste of money.[1] The strongest argument was that other countries spent far less and had no more trouble.[11]
The people who ran the repair had answers ready. Koskinen said most poorer countries had few computers to begin with, and that many of their vital systems turned out to run on manual or old-fashioned controls that had no dates to get wrong.[5] He also said he knew of no big bank, power company or phone company that felt it had wasted its money.[5]
Leon Kappelman, a University of North Texas professor who sat on several Y2K preparation committees, admitted that some things were replaced that did not need to be. He still reckoned that 80 to 90 percent of the spending was well aimed.[11] Writing in Slate in 2009, Farhad Manjoo added that American systems were bigger and more tangled than most, and that other countries did not skimp as much as the story suggests.[11]
De Jager, the man who had once been laughed at by his manager, put the case more bluntly. "The notion that nothing happened is somewhat ludicrous," he told Time.[1]
The next one
Many computers keep time by counting the seconds since the start of 1970. On older systems that counter runs out of room at 3:14 in the morning, universal time, on January 19, 2038, and the date can jump back to December 1901.[12] Most modern machines have switched to a bigger counter, but old systems have a way of sticking around. When a computer museum tested one of its old machines with the clock moved forward, it crashed in 2037, a year early.[12]