[ntp:hackers] Re: NTP and leap-seconds

David L. Mills mills at udel.edu
Wed Jul 5 00:38:44 UTC 2006


Very tiny amendments. First, IBM Sysplex 9037 is listed at $100,000. 
Second, it does the leap, but the operator has to tell it when 
beforehand. IBM's suggestion for operators that really want local time 
is to turn the machine off during the Fall leap backward interval. In a 
global system, that would mean dark computing for one day.


Tim Shoppa wrote:

> "todd glassey" <todd.glassey at worldnet.att.net> wrote:
>> I have suggested that NIST and USNO and the other National Timing
>> Laboratories submit secured notice - through signed emails, to the 
>> Satellite
>> Timelords of the Non-Federally Operated leaves of the global NTP 
>> Services.
>> I am a Professional Auditor, and as such assure you folks that the US
>> Government already uses the exact same thing for Notices of 
>> Electronic Court
>> Activities... So it does work.There is an entire practice model which if
>> this group is interested, I can submit later this week.
> All of this seems like extreme overkill considering that few of the
> OS's that underly NTP, and very few syscalls/libraries/applications,
> support the concept of leap seconds in a Good Way. Usually
> we just want to supply non-leapsecond aware applications
> with a monotonically increasing time that isn't off by more than
> a second or so in a leapsecond event.
> There are a very few OS's that do support TAI-like timescales (e.g.
> a base clock that always ticks forward one second every second no matter
> what) and use OS-supplied tables to do leapsecond and 
> timezone/calendar magic.
> But as far as I know none of them use NTP nor do they support automagic
> net propogation of leapsecond/timezone tables. An example is z/OS.
> I could be wrong and maybe parallel sysplex operation now allows NTP 
> supplied
> time; I am not a Mainframe Guru although I do bump my horns up against
> their horns at meetings :-) Last time I checked a refclock for a
> parallel sysplex rang in in the $30000-$60000 price tag.
> Of course a similar tick-forward-a-second-every-second-no-matter-what
> timescale is the GPS timescale. And even there, there are a lot of
> embedded timing devices that do NOT properly interpret the leapsecond
> corrections distributed (e.g. Z3801A's) in all circumstances and
> we end up with NTP drivers for those GPS refclocks that apply their
> own correction to the leapsecond-pending flag correction!
> Tim.

More information about the hackers mailing list