MRC@PANDA.PANDA.COM (Mark Crispin) (01/04/88)
This business of leap seconds makes me wonder. The last I read about times on the network, the entire issue of leap seconds was ignored; RFC 868 simply says "the number of seconds since 00:00 1 January 1900". None of the examples indicate any leap seconds (but were they any before May 1983?). So, what are the time servers returning, the 32-bit value with leap seconds or without them? I hope it's the latter! -------
Mills@UDEL.EDU (01/04/88)
Mark, Before 1 January 1972, when the international time scales were rationalized and leap-seconds introduced, the time standards were corrected at appropriate intervals to maintain time with respect to Earth rotation to within 100 ms or so. This fact, along with the fact that atomic clocks were rare near the beginning of the century, makes it meaningless to consider leaps before that date. There have been maybe a dozen leap seconds since then, all positive; however, I don't have the dates handy at the moment. In 1980, when IEN-142 was published, and again in 1983, when its successor was published, it didn't seem like such a big deal to tell time down to the milliseconds, especially since the precision provided by the protocol was only to the second. Since these documents were published, protocols and measurement techniques have evolved to the point that events can be synchronized with high reliability to within a couple of milliseconds over campus LANs and within a couple of hundred milliseconds over typical Internet paths. Thus, it comes as no surprise that the leap-second issue has now become important. See RFC-956, -957 and -958 for further discussion on timetelling in the Internet and a description of the Network Time Protocol (NTP) which makes all this possible. I considered the issue of whether to build in leap-second corrections for NTP, at least back to their inception in 1972 and decided against it. For just about any purpose, it is not necessary to order events with respect to what I will call archival time to better than a second; however, when that becomes necessary, the user would have to insert the leaps as necessary and using archival information telling when they were made. In fact, the time scale NTP uses really concides with International Atomic Time (TA-NBS). After leap-second corrections TA-NBS becomes UTC-NBS as broadcast. TA-NTP concided with TA-NBS just after the original correction was inserted prior to 1 January 1972. The rest is history. Dave
LAWS@rsre.mod.UK (John Laws, on UK.MOD.RSRE) (01/05/88)
Dave, Your extensive and timely(!) knowledge about time-keeping makes me wonder if you are thinking of applying for Swiss nationality. The Swiss have great respect for clocks. They have that great engineering principle - if its working don't muck it about. As far as I know they are alone in Europe in NOT changing there time + or - 1 hour each year; they must be miffed by all this 1 second leaping done at infrequent intervals. John