REMINDER: Leap Second

This June 30th, 235959UTC will be followed immediately by 235960UTC.

What will /your/ devices do?

http://www.marketplace.org/topics/world/leap-second-deep-space-and-how-we-keep-time

Cheers,
-- jra

Hi,

Java had some issues with 100% CPU usage when NTP was running during
the additional second in 2012.
http://blog.wpkg.org/2012/07/01/java-leap-second-bug-30-june-1-july-2012-fix/

Google did something different to get the extra second in:
http://googleblog.blogspot.de/2011/09/time-technology-and-leaping-seconds.html

Most devices probably don't even know about the leap second coming as
that would require a firmware upgrade.

Karsten

I think devices would likely be fine, unless they're concerned with reconciling
a leap-second updated ntp source and one that's not. Who wins?

For most NTPs I would guess they're slaves to whatever feed and just 'believe'
whatever they're told. (Sounds like a security hole waiting for high frequency
trader types, q.v.
http://www.theverge.com/2013/10/3/4798542/whats-faster-than-a-light-speed-trade-inside-the-sketchy-world-of
)

Can't we just subscribe to a leapsmeary NTP feed if we care to have no
big leap (I dont mind)? Isnt NIST offering this?

/kc

In article <201501251019290550.005C05BC@smtp.24cl.home> you write:

I've always wondered why this is such a big issue, and why it's done
as it is.

A lot of people don't think the current approach is so great.

In UNIX, for instance, time is measured as the number of seconds
since the UNIX epoch. imo, the counting of the number of seconds
should not be "adjusted", unless there's a time warp of some sort.
The leap second adjustment should be in the display of the time,
i.e., similar to how time zones are handled.

It shares with time zones the problem that you cannot tell what
the UNIX timestamp will be for a particular future time. If
you want to have something happen at, say, July 2 2025 at 12:00 UTC
you can guess what the timstamp for that will be, but if there's
another leapsecond or two, you'll be wrong.

Life would be a lot easier for everyone except a handful of
astronomers if we forgot about leap seconds and adjusted by a full
minute every couple of centuries.

It shares another problem - that doing calculations across a boundary is
difficult. If you have a recurring timer that pops at 23:58:30 on June 30,
and you want another one in 2 minutes. do you want a timer that the next pop
is at 00:00:30 - or 00:00:29? The operating system can't tell whether the
desired semantic is "as close to every 120 elapsed seconds as possible" or
"as close to the half-minute tick as possible".

And of course doing interval math across several years where you cross multiple
leap seconds is even more problematic - for some corner cases that have an
endppoint nearmidnight, doing a naive "timestamp in seconds +/- 86400 * number
of days" can land you on the wrong *day*, with possibly serious consequences...

the quote from the GNU coreutils manpages on Date Input Formats:

"Our units of temporal measurement, from seconds on up to months, are so complicated, asymmetrical and disjunctive so as to make coherent mental reckoning in time all but impossible. Indeed, had some tyrannical god contrived to enslave our minds to time, to make it all but impossible for us to escape subjection to sodden routines and unpleasant surprises, he could hardly have done better than handing down our present system. It is like a set of trapezoidal building blocks, with no vertical or horizontal surfaces, like a language in which the simplest thought demands ornate constructions, useless particles and lengthy circumlocutions. Unlike the more successful patterns of language and science, which enable us to face experience boldly or at least level-headedly, our system of temporal calculation silently and persistently encourages our terror of time. ...

It is as though architects had to measure length in feet, width in meters and height in ells; as though basic instruction manuals demanded a knowledge of five different languages. It is no wonder then that we often look into our own immediate past or future, last Tuesday or a week from Sunday, with feelings of helpless confusion."

-Robert Grudin, Time and the Art of Living.

http://www.gnu.org/software/coreutils/manual/coreutils.html#Date-input-formats

/kc

I spoke on time hacking and ntp 3 years ago at shmoocon.

I have automation code for a "classroom" full of Cisco routers, and the
way I deal with that sort of issue is to say that anything that has to
be synchronized to the wall clock uses cron(8), but for actions tied to
a fixed interval I use sleep(3) and let the operating system sort out
the issues, if any. I did this to sidestep the issues with Daylight
Savings Time (DST) cusps; it also works for leapseconds, as the OS
interval timing is not tied to the real-time clock.

Funny you should mention ticks versus elapsed time. In every
specification I've written since 1970, I've differentiated between the
two. I got started doing that because in the computers of that era the
real-time clock was tied to power-line frequency, while the interval
timers were based on counts on a crystal oscillator. The crystal was
using good for 1000 parts per million, good enough for short intervals.
The power-line clock was pulled back and forth by the power company as
said company would fine-tune the time so electric clocks would stay at
the right time long-term as the expense of short-term jitter.

Today's computers don't use clocks derived from 50- or 60-hertz
power-line frequency. The last computer I remember seeing with such a
clock was the IBM System/360. The System/370 used a motor-generator set
for the power supply, so it had to get its real-time clock time source
another way.

The 360/95 and /91 also used 400 Hz from a motor/gen. Water cooled, too.
One of my fondest war stories is when the power was turned off for July 4th
weekend but the water was left on.

That made the transformers smaller/cooler and more efficient. I seem to remember a 195 as well but maybe it is just CRS.

Google says the 360/195 did exist. But my baby was the 360/95,
where the first megabyte of memory was flat-film at 60ns, which
made it faster than the 195 for some things. It was incredibly
expensive to build - we heard rumors of $30 million in 1967 dollars,
and sold to NASA at a huge loss, which is why there were only two
built. I used to amuse myself by climbing into the flats memory
cabinet, and was amused again some years later when I could have
ingested a megabyte without harm. Ours sat directly above Tom's
Restaurant, of Seinfeld fame. Very early climate modeling was done
on that machine, along with a lot of astrophysics.

The /95 was a /91 with a megabyte of thin film memory, which was both
much faster than core (120 vs 780 ns cycle time) and much more
expensive (7c rather than 1.6c per bit.)

The /195 was a /91 reimplemented in slightly faster logic with a 54ns
rather than 60ns cycle time, and a cache adapted from the /85. I can
easily believe that for programs that didn't cache well, the /95 with
the fast memory would be faster. IBM lost money on all of them and
eventually stopped trying to compete with CDC in that niche.

See alt.folklore.computers (yes, usenet, reports of its death are
premature) for endless discussion of topics like this.

R's,
John

I'm pretty sure University College, London (UCL) had a 360/195 on the
net in the late 1970s. I remember it had open login to I guess it was
TSO? I'd play with it but couldn't really figure out anything
interesting to do lacking all documentation and by and large
motivation other than it was kind of cool in like 1978 to be typing at
a computer in London even if it was just saying "do something or go
away!" I guess you had to be there.

        -Barry Shein

The World | bzs@TheWorld.com | http://www.TheWorld.com
Purveyors to the Trade | Voice: 800-THE-WRLD | Dial-Up: US, PR, Canada
Software Tool & Die | Public Access Internet | SINCE 1989 *oo*