[ntp:questions] A Suggestion For Abolishing the Leap Second

mills at udel.edu mills at udel.edu
Wed Jun 6 18:26:27 UTC 2007


The "kernel clock" is an atomic variable that increments once each 
second. A leap is implemented following the increment at second 23:59 at 
which time the kernel clock is decremented, with the net effect the 
second remains the same as during second 23:59. That's how the kernel 
works, and I apologize for whatever confusion has been caused.

The issue about stepping backware is more subtle. Sometimes the clock 
discipline needs to correct the clock backwards a smidgin. In the code 
that leaves here the clock is stopped if the backward correction is less 
than one second and stepped if more than that. In this context a step 
would only occur if a backwards step more than one second was made by 
the Unix settimeofday() was called.


Guy Macon wrote:

> David L. Mills wrote:
>>Guy Macon wrote:
>>The kernel clock reading routine...
>>>The NTP clock...
> Just to make sure I am not confused, do the clock kept by the 
> kernal and the clock transmitted by the NTP protocol track
> each other (ignoring small errors/offsets) as they traverse 
> the leap second?  I need to study the references given further,
> but the descriptions of NTP seem to describe a clock that jumps
> back a second while the descriptions of the kernel seem to 
> describe a clock that is monotonic and never jumps backward.
> That's how I read the figure at 
> [ http://www.eecis.udel.edu/~mills/leap.html ].
> Could it be that I read a post about "the clock" stopping and 
> assumed from it being posted in comp.protocols.time.ntp that 
> it refferred to the NTP clock, followed by my posting about 
> "the clock" and getting a reply that assumed from the fact that 
> the post I replied to was about the kernel clock that I was 
> also referring to the kernel clock?  Of course I, being a hardware 
> engineer and in no way an expert in this area, may be totally 
> misunderstanding things.  If so, please forgive my ignorance.

More information about the questions mailing list