Re: [Ntp] Antw: [EXT] Timescales, leapseconds and smearing

[Martin Burnicki <[e-mail]>]

Kurt Roeckx wrote:
> On Tue, Dec 08, 2020 at 10:56:24AM +0100, Ulrich Windl wrote:
>> Hi!
>>
>> I think if we agree to use anything other than UTC in NTPv5, we absolutely
>> need to specify some use-level API how to get the current time:
>> time()
>> gettimeofday()
>> clock_gettime()
>> localtime()
>> strftime()
>>
>> They all assume UTC... It's rather useless if NTP has the exact time, but no
>> ("standard") program can use it.
> 
> They don't use UTC, they use POSIX/UNIX time. POSIX time, just like
> NTP, ignores the leap seconds. POSIX actually says that the length
> of a second is not specified, an implementation can then say it's
> SI seconds. POSIX has support for a leap second in struct tm, so as
> long as you don't convert it to a time_t you can work in UTC.
> strftime() works with a struct tm, so can work in UTC.

Right, and the problem is that protocols like NTP or PTP (and their
implementations) have to cope with the limitations of those specs.

I don't believe that this can be handled at the time synchronization
protocol level. That only shifts the reason of the problems to some
slightly different location.

> The linux kernel has support for TAI. You can set a TAI offset, so
> that applications making use of the POSIX API still get the POSIX
> time. The clock_gettime() function you mentioned has TAI support.
> The information to convert from TAI to UTC is available, so it's
> possible to work in TAI or UTC on Linux, it's just that most
> applications don't.

But in any case you first need to have a trusted source for the TAI
offset, so some program can send it down to the kernel.

By the way, if you start a Linux kernel and the TAI offset has not been
set by some program, the kernel reports TAI offset 0.

However, when the kernel receives a leap second warning from an NTP or
PTP daemon, and it inserts a leap second, the kernel increments its TAI
offset when the leap second is inserted.

To the kernel reports TAI offset 1 after it has observed a leap second,
which is obviously wrong, but it shows that you can't simply check if
the TAI offset is 0 if you need to know if the kernel knows the real TAI
offset.

Tricky, all this stuff.


Martin
-- 
Martin Burnicki

Senior Software Engineer

MEINBERG Funkuhren GmbH & Co. KG
Email: [e-mail]
Phone: [phone]
Linkedin: https://www.linkedin.com/in/martinburnicki/

Lange Wand 9, 31812 Bad Pyrmont, Germany
Amtsgericht Hannover 17HRA 100322
Geschäftsführer/Managing Directors: Günter Meinberg, Werner Meinberg,
Andre Hartmann, Heiko Gerstung
Websites: https://www.meinberg.de  https://www.meinbergglobal.com
Training: https://www.meinberg.academy