Internet dies if GPS dies? Nah.

[Bill Stewart]

Variola quoted and commented:
 > > [If GPS dies] "Internet activity would slow to a crawl, because many
 > >   backbone operators rely on precise GPS time stamps to route data. "
 > > http://www.wired.com/wired/archive/11.05/start.html?pg=6
 > Sounds like bullshit to me, data clocks should be able to run without
 > being fully synched externally, constantly.

There are at least three interesting GPS failure modes
- Whole system fails at once
- Whole system becomes less accurate
- Local areas get jammed
The article's talking about a failure mode where the satellites
gradually wear out over a couple of years and nobody replaces them.
 From a telco perspective, this is annoying, because GPS is cheaper
and better than what we used to do before we adopted it about 5 years ago,
and we've got a few hundred billion dollars less ready cash then we did then,
and the people who really knew synchronization well have mostly retired
or moved to dotcoms, but we could always hire enough of them back to do the job
for a lot less money than it would take to redesign GPS,
on the off-chance that we haven't replaced enough of the old phone 
infrastructure with VOIP
for it to simply Not Matter by then.  It's not like this is Air Traffic 
Control.

[Summary:  If you blew GPS out of the sky without warning,
major telco synchronization would degrade a bit but not much,
voice calls would be more likely to get noise, but not much,
internet connections might have a fractional percent more TCP retransmits,
and a random number of things that didn't have other timing sources would break
until people fixed them.  Cell phone systems are the big telecom uncertainty,
and of course Air Traffic Control, nuclear missiles, and similar apps.]

As Tyler said, SONET synchronization is a complex topic,
and telco synchronization is more complex than that.
There are two basically separate problems
- timestamping / NTP Network Time Protocol, used by lots of Internet stuff,
	because ISPs sometimes need to know what time it is.
- synchronization used in synchronous telco transmission media,
	which doesn't care about time of day, just about phase.
	ISPs care about this because they don't like lots of dropped bits,
	but it takes a lot of dropped bits to really bother TCP.

Everybody uses GPS for everything these days, because they _can_,
but high-end GPS equipment is about 5 orders of magnitude more accurate
than the most demanding NTP things ISPs use it for,
and cheap GPS equipment usually gets you millisecond precision,
which is about 3 orders of magnitude more than most things care about,
and most of them don't care very much.  WWV radio clocks were Really Just Fine,
and ISPs who needed better clocking could build decent NTP systems.

Routers don't care much about phase, because each T1 or fiber interface
usually gets timing from the transmission line independently.

Voice transmission equipment cares more about phase,
because it uses Time Division Multiplexing to switch the voice bits,
so a whole voice switch runs on the same clock source -
it it's talking to somebody else whose clock is different and
drifts by more than the line buffers can fix, it drops a frame of bits
or adds a frame of 0s, which will cause brief noise on voice calls
or trash one or two data packets.  Telcos care about this because they
not only care about voice quality, they have SLAs with customers
for the number of errored seconds per day/month/etc.
that cost money that cost money if they don't meet them,
but it takes a lot to get down to cell-phone-in-traffic quality.

Telecom sync equipment is categorized in different strata,
and normally a box will have a clock that accepts timing from outside
and has a Stratum-N-quality holdover clock inside that's guided by the feed.
There are lots of different ways to measure the accuracies of these things.
A Stratum 1 clock is whatever master timing source you're using,
and a Stratum 2 clock is supposed to be able to hold its own for about 5 days
before the first slip and not do more than a slip every couple hours after 
that.
A Stratum 3 clock can slip about 10/hour the first day and 132/hour 
free-running.
SONET needs somewhere between Stratum 3 and Stratum 4 to stay connected.
The best timing feeds between buildings run on copper T1s, not fiber.
GPS is Stratum 1 quality, as long as you've got a good holdover clock,
typically Stratum 2 quality for telco offices.

A decade ago, a main AT&T, MCI, or Sprint office would have
a feed from a Stratum 1 source, and a Stratum 2 clock of its own,
and would feed a bunch of smaller offices which had their own clocks.
MCI and Sprint tended to use Loran for some of their timing sources;
AT&T used really expensive clocks in main offices, and had more offices.
An important concept for AT&T is the Building Integrated Timing Source -
there's a master clock for each building that feeds all the hardware.
Most of the other professional telcos do something similar;
I'm not sure if all the upstart newcomers do.
Since then, most AT&T offices have their own GPS clock,
except a few percent with Bad Radio Magic or building issues
which get fed by other nearby offices.
I'm not sure how local telcos get their timing - presumably a similar
combination of GPS, local clocks, and feeds.  They have more small offices
which are closer together, but also the clocks keep getting cheaper.

Cellular systems are the main telco case I don't understand;
some of the newer systems care much more about timing,
but the application that needs the most precision is 911 location,
because that needs to triangulate between multiple offices
as opposed to just handing off calls at the optimum time to avoid drops.
It's possible that something could go wrong here,
but I'd guess that cell towers are also more likely to be fed by copper T1s,
which would give them excellent sync if they need it.
(GPS is also useful for finding the precise location of a tower,
which phone-locator ICBM-delivery apps care about,
but you only need to do that once; the things don't move except in 
earthquakes.)
at least with some of the newer systems, but I'd be surprised.