[Ntp] Antw: [EXT] Re: comments on draft‑mlichvar‑ntp‑ntpv5‑03

>>> Dan Drown <dan-ntp@drown.org> schrieb am 26.11.2021 um 03:43 in Nachricht
<20211125204300.Horde.zdIONiOUK0wqEW0NDfZvexa@mail.drown.org>:
> Quoting Steven Sommars <stevesommarsntp@gmail.com>:
>> Root dispersion should have the same resolution as the time stamp.
>> With today's 16‑bit field the smallest root dispersions are 0,   15.3,
>> 30.5 ... microseconds.
>> What should an NTP server having 1 microsecond uncertainty return?
> 
> I agree that NTPv4 root dispersion does not have enough resolution.
> 
> With this NTPv5 spec, root dispersion is 32 bits with 28 bits assigned  
> to the subsection fraction. This gives 1/(2^28) or 3.7 nanosecond  
> resolution. I don't see the point in taking root dispersion from 32  
> bits to 64 bits for a smaller resolution than that.

I wonder: Should a "virtual infinity" value (probably for every special
numeric type) be defined?
I mean: we are limiting the range for root dispersion rather arbitrarily
(through the number of non-fractional bits), while in theory (through math) the
value could be exceeded.
I think some of the algorithms might be more clear when using a symbolic
infinity value, even if that's far from infinity.

MAXDISP (16) somehow is that, as the value cannot actually be reached.
But it seems to have some double-role: The infinity value and some
protocol-related value.

Example:
if r->rootdelay / 2 + r->rootdisp >= MAXDISP 

Regards,
Ulrich

> 
> _______________________________________________
> ntp mailing list
> ntp@ietf.org 
> https://www.ietf.org/mailman/listinfo/ntp 