[LEAPSECS] presentations from AAS Future of Time sessions

[Poul-Henning Kamp]

In message <5036FB31-5CB7-46A6-949E-5534441FE972 at bsdimp.com>, Warner Losh write
s:


>> The development was concurrent, not sequential. Unix 1st and 2nd edition

>> had a 1971 epoch and 1/60th second resolution. 3rd edition moved the epoch

>> to 1972.


According to Dennis Ritchie what happened was that they _stopped_
moving the epoch all the time.

I don't think he told me exactly what representation they used
before time_t became 32bit*seconds, but prior to that, the wrap-around
of timestamps was prevented only by the kernel crashes.  I think
he mentioned that this changed caused all inodes to grow, so it can
probably be pinpointed in time by the appearance of the three
timestamps there.


>The precision of the time in Unix was sub-second, but given all

>accounts of the time, and my experience with systems a decade later,

>I'd be very much surprised if the accuracy of the system time was

>better than sub-minute since a synchronization Dennis' wrist-watch

>was involved. Given how well computers I used in the early 80's

>maintained time, I'd imagine the synchronization a decade earlier

>likely was a regular occurrence...


When we did credit-card processing on UNIX in the late 1989'es,
UNIX vendors stared blank at you, when you asked them about
things like clock stability and precision.

Granted, they seemed to stare blankly at you a lot if you asked any
questions, but I remember the one about the clock got a particular
blank stare.

Unisys were unique on this:  They proposed selling us their smallest
1100 mainframe, because that had a (hideously expensive) option for
attaching a WWV shortwave radio ("Spectracom" or "Spectratime" I
belive), and then we could send a request from UNIX, over an async
line and get a timestamp from the mainframe.

When I pointed out to them, that there were absolutely no way they
could design an antenna which would receive Shortwave WWV anywhere
near a computer room in Europe, they looked at each other and went
"Ohh, so that's why..."

Despite this level of competence, they got to sell us a UNISYS
7000/40, which was the Tahoe rebadged with a SVID-bastardized BSD,
because that was more or less the largest computer we could get
at the time.

In the end I we found out that we could "leech" on the accounting
timestamps on the X.25 lines, and we just slaved to those, trusting
that if it was good enough for telcos, we'd be clear also.

-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.