In a recent note, Edward E. Jaffe said:
> Date: Fri, 29 Jul 2005 05:59:28 -0700
>
> Farley, Peter x23353 wrote:
>
> >PMFJI here, but isn't NIST time available via NTP to the microsecond or
> >better? ISTR they use an atomic clock to synchronize NIST, and the accuracy
> >is quite high (or so I was led to believe).
> >
The NIST web page says little about accuracy. But perhaps I haven't
looked far enough. The first thing to consider is that the speed
of light is 300 meters/microsecond. So unless you know the path
length from NIST within that accuracy, the time accuracy is no
better. But you may require nothing better than consistency within
your site.
> z/OS systems in a sysplex must have synchronized TOD clocks. My
> understanding was that the precision of NTP was insufficient for this
> purpose I'm certainly no expert in this area, so I'll defer to IBM to
> explain the intricacies of STP when the time comes.
>
NIST says more about their dialup service:
File that you are currently viewing
Linkname: Synchronize your computer to NIST Time
URL: http://tf.nist.gov/timefreq/service/time-computer.html
Link that you currently have selected
Linkname: By Telephone
URL: http://tf.nist.gov/timefreq/service/acts.htm
Phone number Location
(303) 494-4774 Colorado
(808) 335-4721 Hawaii
Since the OTM is delayed as it travels from NIST to your
computer, ACTS sends it out 45 milliseconds early. This always
removes some of the delay. Better results are possible if the
user's software returns the OTM to ACTS after it is received.
Each time the OTM is returned, ACTS measures the amount of
time it took for the OTM to go from ACTS to the user and back
to ACTS. This quantity (the round-trip path delay) is divided
by 2 to get the one-way path delay. ACTS then advances the OTM
by the one-way path delay and the OTM changes from an asterisk
to a pound sign (#). When the # sign appears, the time code is
synchronized within a few milliseconds of UTC(NIST).
I remember reading more in the past, perhaps from the "HELP"
command on the service itself. The half-path delay functions
only if the measured round-trip delay is consistent with an
even number of satellite hops. In the unlikely event that
the round trip path is unsymmetrical, NIST can't tell whether
it's land inbound and satellite outbound, or vice versa. The
one-way satellite delay (up+down) is about 1/4 second.
The time is correct at the center of the stop (IIRC) bit of
the OTM.
Accuracy is best with a 300-baud modem; higher speeds have
greater uncertainty because of DSP latency.
My own musings:
Of course, the effective accuracy depends on the interrupt latency
of your system's processing of serial transmissions, particularly
on the latencies involved in echoing the OTM. Ideally, this should
be done on a bit-by-bit, not a character- by-character basis.
I suppose the best one could do is record the digitally sampled
ACTS carrier with precise time markers from the TOD clock and
unwind the mess with DSP after the fact. All this would require
very specialized hardware. And software.
I'd not like to see the HMC used for STP. I value the reliability
that comes from simplicity in such a component. Would you like
to see the HMC running Windows?
Offhand, I'd expect GPS to be 5 orders of magnitude better: a few
tens of nanoseconds, as opposed to a few milliseconds. But GPS
has political and military entanglements: DoD could fuzz it again,
or even shut it down if that were deemed strategic. (Hmmm. Why
does maps.google.com show a strangely whited out area about 80
miles NW of Las Vegas?)
One of the documented functions of the ETR has been to provide
a precise frequency reference to drive the TOD clock. This
works even on systems (Linux, VM) with no software support for
the ETR. How can this be duplicated with no hardware? Upgrade
to embedded microcode?
Will STP relax the irritating 15-second maximum clock adjustment
limit? Will it support a greater than 2 sec/day slew rate?
And on Ed's earlier comments:
I avoid access to NDA materials unless I have a need-to-know in
my own duties. It's my first line of defense against the temptation
to make unwarranted public comments. And my own duties have little
to do with hardware, so I don't know how one might attach modems
do z/Series; evidently not to the OSA.
-- gil
--
StorageTek
INFORMATION made POWERFUL
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
