Conference gyro::internet_toolss

And we thank you for your support. ([email protected]) wrote:
: Title: Network Time Protocol (NTP) question.
: Reply Title: (none)

:     We're planning to make available either two or three servers.  The
:     OpenVMS and UNIX clients will use ntp; the Windows NT clients will use
:     SNTP.  Given that, what is the right number of servers?  Is there
:     something magical about the 3 servers used in PA?

Nothing magical; it probably started as an homage to Monty Python and
becamse a tradition (three shall be the number of NTP servers, and the
number of NTP servers shall be three). 

Stephen
--
- -----
Stephen Stuart				[email protected]
Network Systems Laboratory
Digital Equipment Corporation
[posted by Notes-News gateway]

Three's also handy as a tie-breaker, if 1 and 2 disagree 
significantly about the time, then having a 3rd server
would help (or is this only w/ Digital's DTSS?).

    re: .-1
    
    DTSS and DCE DTS work like that, I think, at least when you have only
    local clock sources, but xntpd I think takes whatever it has and, by
    some fancy algorithm based on stratum, delay, dispersion, etc., picks
    the "most reliable" of the bunch.
       
    -Tom

The documentation recommends "at least three".  Interestingly, I suspect there
is little point in having multiple ntp servers that are themselves synchronised
back to the same stratum 1 time source, but I have never seen this written down
- the only thing it gains is a bit of resilience against a machine being down.

T

> Three's also handy as a tie-breaker, if 1 and 2 disagree 
> significantly about the time, 

	Reminds me of an old saying:

		A man with one watch always knows what time it
		is.  A man with two watches is never sure.

	I sure that can be applied to time servers as well...

    If I want 2 nodes to time sync just between them, how do I set up ntp,
    especially if there is a possibility that one or the other may not
    always be there (one system or the other may be taken down for maint.,
    etc)? Are they both peers, or what? This is for DUNIX V4.0B.
    
    Thanks,
    
    /jim
    

> 	Reminds me of an old saying:
> 		A man with one watch always knows what time it
> 		is.  A man with two watches is never sure.

	Well we are getting closer to the day when all watches (at least
	electronic ones) can be automatically set from various time
	sources (the TV, your computer, etc).

	I've been wondering for years why broadcasters and Cable TV
	stations/operators didn't get together with TV & VCR makers so
	your TV & VCR could set the time themselves.  Now this is
	actually starting to happen (another idea I should of come up
	with a patent for, but someone like Bill Gates would of
	stolen the idea anyways ....)

    I had a VCR with the automatic time.  It still did not agree with the
    rest of the world including the time on one of the cable preview
    channel :-(
    
    - Vikas

> I've been wondering for years why broadcasters and Cable TV
> stations/operators didn't get together with TV & VCR makers so
> your TV & VCR could set the time themselves.  Now this is
> actually starting to happen (another idea I should of come up
> with a patent for, but someone like Bill Gates would of
> stolen the idea anyways ....)

	And FWIW, I also expect some day that electric companies will
	one day distribute the time over the power lines and into
	your home.  This way any appliance that you plug into the
	wall (clocks, microwaves, coffee makers, TV's & VCR's, etc)
	can automatically set themselves.

>> I've been wondering for years why broadcasters and Cable TV
>> stations/operators didn't get together with TV & VCR makers so
>> your TV & VCR could set the time themselves.  Now this is
    
>	And FWIW, I also expect some day that electric companies will
>	one day distribute the time over the power lines and into
>	your home.  This way any appliance that you plug into the
>	wall (clocks, microwaves, coffee makers, TV's & VCR's, etc)
>	can automatically set themselves.
    
    Would that stop all those neon signs on 42nd St. from flashing?
    

    Re .15 (time over power lines):
    
    See http://www.digital.com/info/rcfoc/970224.htm
    
    PJDM

>     Re .15 (time over power lines):
>     See http://www.digital.com/info/rcfoc/970224.htm

	That article appears to talk about using your buildings wiring
	as a LAN, which like the article said, has been around.

	When I said "time over power lines", I was talking about
	having the power company broadcast the time over their
	entire infrastructure city, state, or more wide, not just
	intrabuilding.  I imagine this to be a much easier task than
	bi-directional high-speed (and bandwidth) communications than for
	even intra-building communications.

	Though wireless time signal distribution and ultra-cheap receivers
	would be ideal (then anything with a clock, including wristwatches,
	not just appliances that plug into the wall that have clocks,
	could be set automatically).  Imagine evenyones, including the
	average joe's, syncronized to the same time.  It could have the
	same effect of large scale economic savings that the the invention
	of the clock originally had.  No more "my watch is slow or fast".
	Of course imagine the havoc if a hacker was able to set everyones
	clock to the wrong time (so security would be very important) :-)

	So what time is it? :-)

vaxcpu> rdate -v
birddog.zko.dec.com: Wed Mar  5 20:30:50 1997
brkhse.zko.dec.com: Wed Mar  5 20:31:30 1997
corvet.zko.dec.com: Wed Mar  5 20:28:52 1997
dicer.zko.dec.com: Wed Mar  5 20:31:30 1997
dirbase.zko.dec.com: Wed Mar  5 20:34:00 1997
dirobj.zko.dec.com: Wed Mar  5 20:33:30 1997
eeyore.zko.dec.com: Wed Mar  5 20:31:30 1997
hpobb.zko.dec.com: Wed Mar  5 20:30:23 1997
obany.zko.dec.com: Wed Mar  5 20:26:21 1997
obblue.zko.dec.com: Wed Mar  5 20:28:41 1997
oclass.zko.dec.com: Wed Mar  5 20:31:48 1997
otest3.zko.dec.com: Wed Mar  5 20:31:30 1997
sunlite.zko.dec.com: Wed Mar  5 20:40:45 1997
thp02.zko.dec.com: Wed Mar  5 20:25:06 1997
tsun02.zko.dec.com: Wed Mar  5 20:22:26 1997
tsun03.zko.dec.com: Wed Mar  5 20:35:02 1997
vaxcpu.zko.dec.com: Wed Mar  5 20:36:05 1997
yung.zko.dec.com: Wed Mar  5 20:35:14 1997
Network time is Wed Mar  5 20:31:54 1997

    Signaling over the power grid has a major problem in that the grid is
    full of transformers with large ammounts of iron in them, and to any
    signal higher in frequency than  the power line frequency they look 
    like huge impedance.
    
    Local to a building, where all the wiring is on the same side of one
    transformer works fine, as with the X-10 stuff for household control.

    >as with the X-10 stuff for household control.
    
    And we all know how well *that* works!  I'm really looking forward to
    that!
    
    jeb (who still goes to the north side of the house to turn off a light
    in the south side!)
    

> Signaling over the power grid has a major problem in that the grid is
> full of transformers with large ammounts of iron in them, and to any
> signal higher in frequency than  the power line frequency they look 
> like huge impedance.

	Then don't use higher frequencies :-)

	I didn't say (nor assume) how the time information could be
	transmitted would be obvious or easy, or I would have patented
	something myself :-)

There was an article in Friday's Australian Financial Review about a consortium 
of power companies that is seriously considering using the existing power lines 
to provide Internet services.

PJDM

    I saw a real goofy horror movie where a house was "alive" and did all
    kinds of bizarre things using the power lines, etc.  Never thought such
    a whacko story might come true, but with intelligent appliances linked
    together via their power cords across the internet, hmmmm....
    
    jeb
    

There is a power company in the UK, Energis, who run the
National Grid, and provide a phone service over the grid.  There's
no reason that this couldn't also do Internet ...

T

Demon Internet use Energis to provide their VPoP's

-- Rob (at home via Demon Energis connection)

    Yes, but they don't use the electricity transmission cables for data.

    re: data over power-grid
    
    what was the Electricity Commission of New South Wales (Elcom) (used
    to) run(s) a lot of voice traffic over the power lines. The quality is
    pretty good - about normal for a "copper-wire" connection. I would
    guess that you could run up to about 9K6 data without too many
    problems. In fact, I would dare say that they have been doing that -
    although I have no "proof" ;')
    
    Not sure who I'd call there to ask... hmmm...
    
    H

Many public broadcasting stations send out a signal that is used by VCRs to
set the VCR clock.  The time to be sent is determined by the station operator
and because various networks skew the start of their show in an attempt to
gain viewer, the actual time setting used is generally a bit in advance of the
actual time.  Some networks start the hour with commercials, but others start
off immediately or a few seconds earlier with a "hook" and then after 90-200
seconds switch to 200 seconds of commercials.  (Now the cable channels usually
do the opposite; they generally delay the start of a program until 1-5 minutes
after the hour, but they also note this in the information they send companies
the supply the TV schedules.)  In almost all cases, network shows with 3 minutes
of commercials, so causing the VCRs to start a minute early is very unlikely
to result in any lost programming.

Bottom line: Set your watch to your VCR and you can arrive before everyone
else at that DEC meeting even if you show up at :01.

The cost of very precise pulse per second signals has now dropped to the
point where the largest part of the cost is the installation.  GPS (global
positioning system/satellite) receivers need a very accurate clock and
can easily output a precise pulse per second signal.  And with GPS systems
being sold in high volumes, the cost of the basic engines is down to well
under $300 and the cost of interface logic is around a hundred.  An active
antenna will run $100-150.  So the cost of the hardware is around $500 if
you are technically handy.  This is for quantity one.  Check out 
http://www.tapr.org/ for more information.

Unless the computer is on the roof, the big expense is going to be running
the cable to the antenna for "commercial" installation.  For your home, the
cost will be less than a day and maybe $50 of coax.

So, how precise can you get the time?  With some good analysis and good
record keeping (how many meters of cable did you actually use exactly)
you will be able to determine the time to within about 20 nanoseconds.
This signal is not affected by SA (selective availability) and in a few
years you will have your choice of satellite positioning data as more
recievers support both GPS and alternatives from countries who understand
that knowledge can't be chained and so you should aggressively sell it.
(Ie., knowing where you are is considered to be a weapon by the US government
just like being able to prevent the FBI from forging your signature.)


The problem is how to get this information into an Alpha with enough
predictability so that you can tie it to the cycle counter.  If you can
determine the delay from when the pulse generates the interrupt until you
read the cycle counter, and if this is stable from second to second, then
you would be able to correlate memory fetch traces on computer on either
side of the globe.

GPS will work for a while. From comp.risks:

The GPS user-equipment code is in even deeper trouble because of the Y2K
problem, and the breakage will occur well before Jan 1, 2000.  Date, in the GPS
signal standard, uses exactly thirteen bits (these bits represent a time-unit
offset from a conventional epoch date).  This allocation is burned into proms
on all existing GPS user equipment.  On about August 20, 1999, the actual date
value will overflow this 13-bit type, and the equipment will fail to produce
correct  time or position information.   Best estimate is that there are ~10^6
pieces of user equipment that will be immediately affected. Everybody who
depends indirectly on those pieces of equipment (meaning all the rest of us)
will also be affected. The GPS standards committee is desperately trying to
figure out what to do with the problem.

PJDM

In article <[email protected]_tools>, [email protected] (mulp) writes:
|> The problem is how to get this information into an Alpha with enough
|> predictability so that you can tie it to the cycle counter.  If you can
|> determine the delay from when the pulse generates the interrupt until you
|> read the cycle counter, and if this is stable from second to second, then
|> you would be able to correlate memory fetch traces on computer on either
|> side of the globe.

Note that a future version of Digital UNIX will probably have
simple support for PPS (pulse per second) inputs via one of the
modem-control pins on a serial interface.

I'm supposed to be putting together an IETF "Informational"
document on how the API for this is going to look, so that
there might be some chance of portability among multiple vendors.
But it's a "spare time" job for me, so it's a little delayed ....

Dave Mills (father of NTP) has already shown that this kind of
approach is good for ca. 10 usec accuracy.  The jitter in the
delays for handling an interrupt make it hard to do much better,
although his work was done with fairly old Alphas; newer ones
might do slightly better.

-Jeff
[posted by Notes-News gateway]

re .30:

>GPS will work for a while. From comp.risks:

The information from comp.risks is not quite correct.

The problem is that the maximum number that can be inserted in the date field
is 1023. GPS time "began" when the system was initialized on January 6, 1980.
The date field will roll over to zero on August 21, 1999. Problems may occur at
this time because the almanac data stored in the GPS will no longer match the
real satellite positions for the "new" GPS time.

Whether or not this is a problem depends on the firmware in the GPS unit. All
Garmin units, for example, will have no problem with this date rollover, since
it was taken into account in the firmware. The problems with other
manufacturer's units vary from no problem, to slight time loss (minutes), to
requiring a cold start of the unit every time, to no functionality at all. In
general, units manufactured in the past few years will suffer no problem, while
early units will have some sort of problem.

There is an excellent article describing the problem, together with a list of
all GPS units and how they are affected, in the May 1997 issue of Cruising
World magazine.

 - Ken

    Right.  The problem is with GPS receivers.  Newer ones (and certainly
    ones manufactured after *today*) will have addressed the problem.
    
    The transmitters, the actual satellites, don't have a problem.  Hence,
    new VCRS that listen to GPS signals to set their time shouldn't have a
    problem.
    
    jeb
    

    Are there really any VCRs who sets their time via GPS ?

Re: .34

You are correct - the VCRs and TVs use a signal transmitted by PBS stations.

				Steve

all of the current (civilian) GPS receivers are single 
channel and won't be able to cope with the interference 
that the Sun will raining on the Earth.  We've gone through 
a Solar min (remember the 11 year cycle ?), and are heading 
back into more active times.


Does anyone know when the new 2 channel civilian receivers 
will become available ?

    The last sunspot maximum was 1989. Were there no civilan GPS receivers
    in use then?

    Also, there are finer distinctions than simply civilian or non (i.e.,
    military).  General aviation GPS receivers are multi channel.  At least
    two have twelve parallel channels.  These are definitely targeted at
    civilian users, albeit civilian users with private pilot's licenses.
    They range from $449 to $1299.
    
    jeb
    

In article <[email protected]_tools>, [email protected] (Jim, Digital UNIX Tech. Partners. Eng. Grp.,381-6199) writes:
|> Title: Network Time Protocol (NTP) question.
|> Reply Title: GPS receivers have a secondary problem
|> 
|> 
|> all of the current (civilian) GPS receivers are single 
|> channel and won't be able to cope with the interference 
|> that the Sun will raining on the Earth.  We've gone through 
|> a Solar min (remember the 11 year cycle ?), and are heading 
|> back into more active times.

Not all of them.  For example, the Garmin GPS 12 XL is
a 12-channel hand-held model, and seems to be available for
about $249.  Relatively new, though.

-Jeff
[posted by Notes-News gateway]

I forget the standard terms and TLAs, so this explanation is "accurate".

The US gps satellites transmit on two channels; I'll call them bands.

5 years ago the activity in handheld units was focused on getting them
to receive up to 8 satellites by switch a single receiver between them.
For the past 2-3 years the focus has been on building 8 channel units
at the low end and 12 channel units for aircraft.  This means that they
receive signals from up to 8 or 12 satellites simulateously and effectively
have 8 or 12 receiver backends.  If your on the ground, you're unlikely to
see more than 8 sats, but if your flying at 50,000 feet, you can see up to 12.
(Currently there are only 24 in the US constellation.)

The integration of the circuitry is complete for these multichannel units,
so the only way to have the best "specs" is to support 12 channels.

Or receive both channels from each sat.

Or receive both US and Russian sats.

I expect to see multiple units under $2K within the year that do use both US
and Russian satellites.

Dual band units using the US system are also within reach today ($5-10K).
The Clinton administration has committed to a dual band system by 2000.  This
is required to support instrument landing using gps.  (The proprogation
variation due to atmospheric and multipath differ by band so this allows
removing those errors).  The units that are currently available are used
primarily for precision surveying, or they are military units that can
unscamble the second channel.  There were a couple of proposals which would
have effectively added a third channel to the sats which would have been used
for civilian use, while the second channel would have evolved for military
use.  However, there was no agreement on the specifics for the third channel
by the deadline so the next 3-6 sats launched will NOT have a third channel.
This leaves converting the current second channel to civilian use and then
added a new military channel.

There are some other systems being proposed as well.