Re: [ntpwg] [dhcwg] Fwd: New Version Notification for draft-ogud-dhc-udp-time-option-01.txt

[Warner Losh <[e-mail]>]

On Dec 2, 2013, at 12:58 PM, Harlan Stenn wrote:

> Warner Losh writes:
>>>> The reason for time_t is the dhcp-client can feed that directly into
>>>> following call: settimeofday(const struct timeval *tp, const struct
>>>> timezone *tzp)
>> 
>> During a leap second, the value of time_t is not well defined because
>> by definition leap seconds don't happen in time_t/POSIX time scale
>> (don't happen -- well don't consume space in the time_t timescale is
>> more pedantically correct). Obscure point likely worth noting...  But
>> using a NTP time stamp doesn't save you, since during a leap second
>> its value is well defined, but the same as another second (ntp
>> distinguishes in bits set in other parts of the header). Assuming the
>> tolerance for this application is >> 1s, this won't matter, but might
>> be worth specifying that during a leap second, the value returned
>> shall be either the value of the prior second or the value of the next
>> second to make it conform to most current usage.
> 
> It's actually fuzzier than that, and this situation (which is handled
> differently with ntpd under unix, ntpd under windows, and on systems
> that "smear" the leap second over a longer period) is one of the cases
> handled by the General Timestamp API project at Network Time Foundation,
> described at http://nwtime.org/projects/timestamp-api/ .

Actually, in all those cases it is within a second of the 'real' time, hence specifying that it should be one second or the other seems appropriate. The timestamp APIs are for sub-second resolution, while the time_t in this protocol has only a second resolution... The main point is that it should be (a) specified to be what's going to most likely happen (or keep bounds on the error)  and (b) something sane that doesn't introduce errors > tolerance of the protocol. Having it be completely unspecified is a mistake, since -1 could be returned...

Warner

_______________________________________________
ntpwg mailing list
[e-mail]
http://lists.ntp.org/listinfo/ntpwg