Summary

  • time.gov followed a popular NIST web clock from 1997. Its domain was registered in June 1999 and the site opened shortly afterwards.
  • The browser used its own computer's clock, corrected by an offset from the server. A separate readout exposed the delay involved in obtaining that reference.

The official time has an awkward delivery problem. By the time a website has told you what time it is, some more time has happened. Excellent clocks can supply the answer. They still have to get it past the Internet and into a browser with other things to do.

In a screenshot preserved in NIST's May 2002 technical manual, the grand digits come with a smaller admission: “Accurate within 0.5 seconds”. That was the estimate for the illustrated request. The tiny line did a useful job. It brought the journey into view.

A clock gets its own address

NIST's account of the early site traces it to a clock added to its division website in 1997. Popularity earned the clock a place of its own. The time.gov domain was registered on 23 June 1999; the website went online immediately afterwards, following a joint launch decision with United States Naval Observatory.

Visitors chose a time zone on a map. Behind that simple click, the server supplied a reference and a Java applet compared it with the computer's clock. The display then ran on that local clock plus the calculated difference, checking back every ten minutes. The user's machine supplied the ticking. The laboratory supplied the reference.

This was an economical arrangement with a very human nuisance: the software had to work where the person happened to be. The animated clock contacted port 8013, a connection that a firewall might block. The alternative was a still snapshot through the ordinary web connection, port 80. To see a newer time, the visitor refreshed it. A clock requiring a refresh button was a splendidly Internet solution to the passage of time.

Please allow time for time

By 2007, the engineers had another reason to simplify things. Each old-style request started a Perl interpreter on the server and involved several files and clock calls. Asking the hour came with a surprisingly busy entourage.

The redesign paper described a forthcoming Flash interface, a compiled timestamp programme and requests through port 80. It would show the US time zones together, saving the preliminary click. Tests on NIST's own network showed the new approach handling simultaneous requests with much less delay. The server had less housekeeping to do before answering.

One tempting shortcut was left out. The designers measured the round trip, but chose to display the delay rather than subtract half of it. They could not tell where along the journey the waiting had happened. If much of it came before the server read its clock, subtracting half could push the displayed time ahead. Arithmetic can be wonderfully tidy while solving the wrong problem.

The new design still used the local computer's ticking between checks. It still had a journey to account for. The impressive part of this early web clock was the care spent on that small gap between having the right answer and delivering it. The large digits told the hour. The little delay readout explained the transport bill.