Re: [Last-Call] [EXTERNAL] Re: Genart last call review of draft-ietf-dtn-bpbis-17

Joe Touch <touch@xxxxxxxxxxxxxx> · Fri, 8 Nov 2019 07:40:41 -0800

Unix time != the Posix time API. One is a time scale; the other is the interface to obtain potentially many different time scales.
I dove into these issues here: draft-touch-time
Joe

On Nov 8, 2019, at 7:34 AM, Burleigh, Scott C (US 312B) <scott.c.burleigh=40jpl.nasa.gov@xxxxxxxxxxxxxx> wrote:

I disagree.  From Wikipedia:

Unix time (also known as Epoch time, POSIX time,^[1] seconds since the Epoch,^[2] or UNIX Epoch time^[3]) is a system for describing a point in time. It is the number of seconds that have elapsed since the Unix epoch, that is the time 00:00:00 UTC on 1 January 1970, minus leap seconds. Leap seconds are ignored,^[4] with a leap second having the same Unix time as the second before it, and every day is treated as if it contains exactly 86400 seconds.^[2] Due to this treatment Unix time is not a true representation of UTC.

Now, of course, this is Wikipedia, which everyone knows to be useless and unreliable.  But the summary is at least concise and clear, and in this particular case it is supported by several source documents, noted at the bottom of the article.

For example, “The Open Group Base Specifications Issue 7, Rationale: Base Definitions”, section A.4 (General Concepts) says “Coordinated Universal Time (UTC) includes leap seconds. However, in POSIX time (seconds since the Epoch), leap seconds are ignored(not applied) to provide an easy and compatible method of computing time differences.”

Going back a bit, page 166 of https://www.tuhs.org/Archive/Distributions/Research/Dennis_v5/v5man.pdf Dennis and Ritchie (who ought to know) say “Time returns the time since 00:00:00 GMT, Jan. 1, 1970, measured in seconds.”

I am happy to agree that an operating system’s implementation of the time() function may typically obtain accurate UTC time (from GPS or NTP) and subtract the leap seconds out of that value to obtain Epoch time, and in this sense the implementation of the time() function is typically dependent on information about leap seconds.

But that is an implementation expedient, which BP does not care about.  What BP cares about is expressing time as a number of seconds that have elapsed since the Epoch (minus an offset, the number of seconds elapsed from the Epoch to midnight 1 January 2000 UTC).  The manner in which that value is generated doesn’t matter to the protocol.

Scott

From: Stewart Bryant <stewart.bryant@xxxxxxxxx> 
Sent: Friday, November 8, 2019 6:33 AM
To: lloyd.wood@xxxxxxxxxxx
Cc: Magnus Westerlund <magnus.westerlund@xxxxxxxxxxxx>; last-call@xxxxxxxx; gen-art@xxxxxxxx; draft-ietf-dtn-bpbis.all@xxxxxxxx; dtn@xxxxxxxx
Subject: [EXTERNAL] Re: [Last-Call] Genart last call review of draft-ietf-dtn-bpbis-17

On 8 Nov 2019, at 13:03, lloyd.wood@xxxxxxxxxxx wrote:

http://lloydwood.users.sourceforge.net/Personal/L.Wood/publications/wood-ieee-aerospace-2009-bundle-problems.pdf

The critical text in that paper is:

"The Bundle Protocol uses Coordinated Universal Time (UTC), where leap seconds are added at irregular, unpredictable, intervals to reflect slowing of the Earth’s rotation. For nodes ‘in the field’ for a long time (decades), some way of communicating newly-decided UTC leap seconds will be required to prevent clock drift over long time scales that would eventually lead to bundles expiring before delivery. This is most likely to be a significant issue for real-time traffic with very short bundle lifetimes."
That says that leap-seconds need to be transmitted to the remote site, but the ID does not say anything about that, indeed it silently implies that this is handled.
The draft text says
Like TAI, Unix epoch time
   is simply a count of seconds elapsed since a standard epoch.  Unlike
   TAI, the current value of Unix epoch time is provided by virtually
   all operating systems on which BP is likely to run.

Which is not quite right. The TAI is a continuous count of the number of seconds since the epoch. The UNIX tine is the continuous count + leap seconds since the epoch. Unix knows how many leap seconds have happened by a background process and adds them in, but for that to work it has to have a method of knowing the current leap second state. BTW leap seconds can be removed as well as added.

- Stewart

-- 
last-call mailing list
last-call@xxxxxxxx
https://www.ietf.org/mailman/listinfo/last-call

-- 
last-call mailing list
last-call@xxxxxxxx
https://www.ietf.org/mailman/listinfo/last-call