[ntp:hackers] ITU and Leap second elimination

Warner Losh imp at bsdimp.com
Thu Oct 3 17:59:28 UTC 2013


On Oct 3, 2013, at 11:53 AM, Magnus Danielson wrote:

> On 10/03/2013 12:40 AM, Warner Losh wrote:
>> On Oct 2, 2013, at 4:35 PM, juergen perlinger wrote:
>> 
>>> On 10/02/2013 02:04 AM, Danny Mayer wrote:
>>> <snip>
>>>> Just to clarify something. NTP disciplines the clock by adjusting the
>>>> frequencies of the ticks. The goal is to keep the clock synchronized
>>>> to the earth's rotation. If you want uniformity then the frequency
>>>> needs to be kept constant and NTP is not the tool to use. These are
>>>> very opposite goals. You cannot have both. 
>>> I'm not sure if i got this straight. AFAIK, NTP assumes UTC as the time
>>> scale, which is why the leap second announcement is important to NTP(D).
>>> Not that NTP woudn't work as well with TAI, but it's based on UTC. I
>>> might stand corrected, though. It wouldn't be the first time in this thread.
>>> 
>>> But I think there is a misinterpretion here. The frequency is not
>>> synchronised to earth rotation. The whole idea of leap seconds was to
>>> make the duration of a second in UTC the same as in TAI -- simply a SI
>>> second. The one that is now based on a Caesium time normal. (Sorry, NIST
>>> references not available due to government shutdown, I won't comment
>>> further on that.) The tricks employed by various NTP implementations to
>>> avoid disruptive behaviour around a radix change (aka leap second) are
>>> an artifact that is based on amelioration strategies on systems that do
>>> not support variadic radix representations. Most of them try to keep to
>>> the motto 'natura non saltat'. But in my understanding the duration of a
>>> second in NTP is a second in UTC is a second in TAI.
>> No. Seconds are the same length in UTC and TAI.
> Nowdays. 1961-1972 they could be different.

Ntp didn't exist in that time period.

>> Leap seconds are only to correct phase errors arising from differences between the frequency of Earth's rotation and that of the more constant cesium atom vibration...
> It's really not vibrations, and you know it.

Vibrations sounds so much nicer than "the duration of 9,192,631,770 periods of the radiation corresponding to the transition between the two hyperfine levels of the ground state of the caesium 133 atom."

Warner



More information about the hackers mailing list