On Mar 8, 2011, at 11:28 AM, Chris Albertson wrote: >> And exactly what is that difference? While ntp is perhaps too slow to >> respond to local frequency changes, how do you see the difference >> between keeping a computer's idea of local time accurate from keeping a >> telecom's idea of local time accurate? > > The telecom equipment needs to know the time with less then one > microsecond error and to do that requires a clock that is at least 10X > better. NTP typicaly works at the microsecond level and has error > 1000X more than is required.
It depends on which telecom equipment we're talking about; for example, the CDMA spec (IS-95, IS-2000 aka CDMA2000) requires the cell towers to be sync'ed to better than 10 microseconds. Without making any special efforts (ie, just using random NTP servers from the pool), NTPd does typically offer around a 1 millisecond accuracy. People who care a bit might configure nearby time sources and set up local peers, which will probably give accuracy around the +/- 100 microsecond level, and anyone who seriously cares about good timekeeping will find a way of using a PPS signal, in which case with kernel PPS_SYNC discipline, you can get accuracy better than microsecond level, depending on the quality of the PPS signal. For example, PHK measured +/- 120 nanosecond accuracy using relatively inexpensive Soekris hardware: http://phk.freebsd.dk/soekris/pps/ Regards, -- -Chuck _______________________________________________ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions