Study finds that a GPS outage would cost $1B per day | Hacker News Reader

Study finds that a GPS outage would cost $1B per day

arstechnica.com

70 pointsjonbaer67 comments

Nobody's talking about what might cause an outage. There are three basic parts to any GNSS: The control segment, the space segment, and the atmosphere on the way back down.

The control segment is run by the military so let's assume it's as secure as any other military network, and that attacking it would be seen as an act of war. And there are several constellations run by various countries, who do not represent a common target.

The space segment is vulnerable to space weather, but has proven pretty darn resilient so far. Even if a big CME or something knocked out a few satellites, there are ton of them up there, and most of the most critical applications only rely on timing, not positioning, and would work fine with a degraded constellation.

The atmosphere is where it gets interesting; an airburst nuke will make the ionosphere basically opaque to such signals for quite some time. How long depends on a lot of details, but assume several weeks. This would affect all signals from all constellations for all users, and is IMHO the only threat worth worrying about. Because an airburst over the ocean might not be construed as an attack on any specific nation, but it would exact a huge economic toll on the most developed and tech-reliant nations. It must be a tempting option for a spastic tyrant who has a nuke or two and feels backed into a corner.

What about space debris? Isn't there a non zero chance that a satellite could hit some debris and cause a chain reaction?

Isn't there enough redundancy in space that one sat going down means nothing for the system

No, the issue with space debris is that we are quickly approaching the point where a single unfortunately placed collision could chain using a sort of exponential “shotgun effect”, destroying virtually all satellites at a given orbit and rendering space travel impossible for decades if not centuries, setting human space exploration back significantly as launches have to deal with thousands of metal shards traveling at tens of thousands of mph.

See this video[1] by Kurzgesagt: “End of Space - Creating a Prison for Humanity.”

[1]: https://youtu.be/yS1ibDImAYU

Space debris is largely an issue for low earth orbit, through will all space activity has to pass through and is a pretty tight orbit. GPS satellites are in geosynchronous orbit which is over 30,000km from the surface of the earth so there is much more volume of open space, making debris a much smaller risk (for the same amount of debris, since 600^3 << 30,000^3).

It would be interesting to know of everything that uses GPS what % of use is purely for a very accurate synchronised time source vs uses that actually require the location aspect.

The article states 'The researchers found that the largest benefit, valued at $685.9 billion, came in the "telecommunications" category, including improved reliability and bandwidth utilization for wireless networks'. I wonder if location is actually required for these purposes or it's purely the synchronised time source (clearly a useful thing to have when building communication systems).

Edit: Ah the question is answered from another part of the article 'Wireless technology continues to evolve in ways that increase its reliance on highly precise timing, which in turn increases reliance on GPS.'

You might be able to use the 1Hz IRIG H code from WWV and WWVB to discipline a crystal oscillator to get a clean 10MHz.

AFAIK this isn't how it works; the GPS provides a reference so the skew of the crystal can be measured as it drifts.

That's literally what "discipline a crystal oscillator" means. You measure and then correct the skew.

The crystal provides a local reference with low phase-noise, and the long-time-constant disciplining provides reasonable holdover performance.

Thanks, I wasn't familiar with the jargon or the electronics.

The original post said "might" but it seems that's exactly what cell sites use (GPSDOs).

Right, a normal cell site would use the 10MHz or 1 pulse per second from a GPS receiver to discipline their own oscillators. GPS could go out for a few minutes and everything would still work just fine. Anything past that and it's all dependent on how stable their oscillator is over time when it's free running.

The WWV and WWVB atomic clock radio stations have 1Hz reference frequencies that might be stable enough to be your 1PPS reference if you designed your GPSDO to accept an external 1PPS.

Location may be used during setup, but only timing is used during normal operation.

Long version: When a new site is installed from nothing, it doesn't even have a name in the database yet, so it gets a sequential number. If several sites are being commissioned at the same time, the techs need a way to establish which one they're talking to. Querying the GPS timing module for its location is a good way to be sure you're not mixing something up.

But the site timing antenna is not necessarily on or even near the tower itself, they could be a hundred feet away. So the location used for location-critical services is actually surveyed in a separate step and programmed in as a site parameter that does not change. It's not dynamically derived from GPS during operation.

Well, you could consider things like the Google Spanner database, that uses gps timing to do atomic commits across datacenters. Since this powers many of googles services (Gmail etc.), the fallout would potentially be massive on IT infrastructure.

The article mentions only GPS. Most modern receivers can use multiple systems for redundancy; GPS, GLONASS, Galileo, BeiDou. They can also provide accurate timing. https://gssc.esa.int/navipedia/index.php/Time_References_in_...

It's not clear if they use GPS as synonym for all available GNSS or is the study GPS only?

EDIT: I skimmed the document. The underlying assumption seems to be that US based industries don't have or don't use multi-constellation capability.

>In looking across the many sectors and applications that require GPS’s accuracy and precision, it becomes clear that GPS has some attributes of a utility. The signal is a public good and service provided by the U.S. government that enables productivity, quality, and efficiency benefits that would not otherwise be possible. For many years, it was the only comprehensive PNT signal available. Signals are now available from GLONASS (Russia), Galileo (Europe), and BeiDou (China). The global marketplace means that many devices are increasingly capable of receiving signals from multiple constellations. In the United States, however, critical infrastructure, industries, and applications leverage the GPS signal.

Products like [1] consider concurrent GPS+GLONASS tracking a paid upgrade.

However, they also have 'holdover performance' allowing them to remain accurate for several hours without a GPS signal.

[1] https://www.microsemi.com/product-directory/carrier-grade-nt...

Similar product from Oscilloquartz has optional GNSS kit with GPS/GLONASS/BEIDOU/GALILEO

And a ton of older GPS-only hardware is still perfectly reliable and relevant. The recent week number roll-over forced some of the more decrepit junk to be replaced, but those with passable WNRO support are still in service despite being single-constellation.

It'll be a long time before multi-constellation receivers are the norm, and even longer before they displace the majority of critical timing receivers and reference stations and stuff.

Note that GPS and "GPS technology" are different things. When people say "GPS" they normally mean the US system and that is the one the article is talking about. But there are several satellite networks that each act as a global positioning systems using "GPS technology".

We are seeing more and more devices that can access many different networks. This is a good thing. Not only does it increase redundancy in case of a widespread outage of one network, it diminishes any one nation's ability to disrupt positioning. No longer can the US simply turn off or obfuscate location data (they used to blur it. Google "GPS selective availability"). The existence of competitor networks allows users to quickly identify and ignore such actions.

GPS is specifically the name of the US system. GNSS, Global Navigation Satellite System, is a general name that also covers other systems.

https://en.m.wikipedia.org/wiki/Satellite_navigation

NAVSTAR is specifically the name of the US system.

GNSS is the generic term.

GPS could be used to mean either, and sucks as a term.

If we want to be really specific, the US program is now called "NAVSTAR GPS" and it was once manged by the "Interagency GPS Executive Board" ... so the people running NAVSTAR still call it GPS, internally at least.

"Collins Aerospace is coordinating with safety regulators and its equipment clients after a GPS connection outage on its parts resulted in the cancellation of hundreds of flights, especially flights on aircraft operated by US regional carriers." [June 10 2019] https://www.flightglobal.com/news/articles/collins-gps-outag...

This seems pretty low. People depend on electronic navigation almost exclusively these days.

My thoughts exactly. How many millions would be unable to get to work (or stranded at work)? So many people have lost the ability to navigate without it.

Your phone can do a lot with just the cell towers for navigation. The towers themselves would lose GPS signal but they are stationary, they should be able to pump out location despite the loss of signal. The guy out in the middle of nowhere might be screwed but there always should be a backup navigational aid when you are that far out.

A GNSS blackout would be more of an issue for ships and planes. Both can get by in most cases with a GPS blackout using beacons, dead-reckoning, maps, and other old-school techniques. However, a major solar storm could cause issues with all of the navigational instruments (radios, compasses), you might be limited to just using a sextant and a precise clock.

Cell towers do not "pump out" location. Your phone queries Google/Apple/Mozilla/OpenCellID/whatever with the IDs of the towers it sees, the service returns approximate location based on its database of towers' locations. Same is done with WiFi SSIDs.

As a metaphor for triangulation, "cell towers pump out location" is a passable first-order approximation, even though the infrastructure is far more complex.

Really, is that a thing?

I'm completely unable to redo a nontrivial GPS route without the GPS after a handful of times, but I can definitely redo routine routes. At first I'd expect so can anyone else.

What about all the people that started at a job recently?

They would get a paper map? You can buy, print, or draw maps.

google maps works without GPS you just need to remember the directions and follow the signs. Even if someone uses GPS even for the easiest routes doesn't mean that he's not capable of doing it without.

Not everyone.

I've always had a horrible sense of direction. I can follow a road or a map, but if I take a wrong turn or otherwise get off track, I get completely disoriented, and it's extremely stressful to me. I also seem to have a poor working memory that makes it hard to remember many steps at a time. I often need many drives over many months along the same route to finally etch it into my brain.

With GPS, I'm much calmer and less anxious, and more willing to take risks with alternate routes.

When I go to a new city, I follow my satnav device for the first few days, then start altering my route -- deliberately missing turns, or starting out in a different direction, to learn where it takes me as a result of those changes. I'll study the map before I depart, and plan which turns to blow, and predict what reroute I think it'll give me.

After a week or two of this assisted learning, I tend to have a pretty good base map in my head, so I start doing without. I'll leave the device face-down on the passenger seat, not giving me any hints, but still tracking my location and there for a quick peek if I get turned around. After a few days of this I can typically do away with it entirely.

Dunno why you were downvoted. I worked with a guy who literally used his GPS every morning for months to get from the same hotel to the same job site. The route consisted of something like three turns. One of which had a big sign clearly advertising the exit.

Some folks would be incapacitated if GPS was out for long enough.

Whilst that three-turns example sounds a bit extreme, I often plug in routes I could drive without GPS to get traffic / road closure notifications & automatic re-routing around them. There are benefits aside from not knowing where you're going :)

I assume Galileo would keep on working? If you trust GLONASS then that one two. I seldom see Chinese satellites somehow.

The assumption here was "no geolocation capabilities". But yeah, diversified satellites mean fewer eggs in one basket, even though the eggs are all very similar, so to speak.

I suspect the distribution of possibilities considered across our level of ignorance is very, very wide. Even if $1B/day is the peak (which I wouldn't consider this anything more than very weak evidence for; no offense to the study authors, this is a freaking hard problem and I respect their attempt to answer the question), I'd expect it to be a very shallow peak in a very wide distribution. There's a lot of unknown unknowns in this space, and a lot of ways that things could cascade surprisingly out of control (e.g., "oops, it turns out that in this particular surprising GPS failure mode that this common model of airplane wants to land half-a-mile west of the actual runway, and while that isn't a problem on its own the modern avionics software is really insistent about that and literally won't let the pilots override, and now the entire air transport industry is on the verge of collapse because nobody wants to fly anymore").

Last year, I would have discounted that as impossible, but, well, B737MAX.

Planes should be able to fall back completely on VOR/ILS systems, GPS is used pretty minimally in aviation. The new ADS-B Out transponders won't be able to transmit their location, but the traditional radar systems should be able to locate them the old way.

Yeah, but after a few days of confusion, people would adapt. Driving around town would be the least of worries, I think.

However, a ton of modern app-based things like ride-hailing apps and scooter rentals rely on the availability of GPS for their operation. Cars could still drive, scooters could still scoot, but reserving them and paying for them don't seem to have a fallback option, and without those parts I don't expect the operators of such services to have any incentive to operate those services.

A basic map is all people need if GPS were to go offline. I have driven across the country, and around cities (both in the US and non-US) without using a GPS. Not really noteworthy nor does it say much about a lot of people who can't function without one.

I would understand if it was a full GNSS outage, must most "GPS" equipment nowadays is multi-constellation, using as much as four (GPS, Galileo, Glonass, BeiDou). The loss of one constellation should go undetected by most end users.

What disaster would knock out only one constellation out of four (perfectly one) and also go unnoticed to users?

Politically driven shutdown. But the point I just made makes it unlikely, it's not the weapon it once was.

The phrase that should strike fear into the heart of any engineer who has to work at scale (any kind of scale, not just software): "Correlated failures". A software bug that affects only one of them. The political possibility mentioned elsewhere. Some computer virus never "intended" for a satellite manages to get into one of them for some reason, most likely a very bizarre one that nobody would have ever guessed could happen but still somehow did. Some hardware component that turns out to fail all at the same time (more or less), used by all of a particular constellation. Or most likely of all, something I can't even think of to put here as an example, but still it somehow turns out that a particular failure ends up having as a side effect greatly increasing the chance of failures in the rest of the system despite all the efforts to prevent that. (The "ones we can't even think of" end up with elevated probabilities precisely because we don't think to prevent them and thus put no engineering into preventing them.)

All of these outcomes are low probability, sure, but they are all vastly higher probability than sufficient simultaneous independent failures to take down the system. It's why despite all the incredible software engineering that goes into Google and AWS, they still go down sometimes. Almost no conceivable set of independent events can take those things down, but correlated failures still can, do, and will continue to do so.

On the contrary, I have a hard time imagining an event that would knock all of them out and was not a relatively minor side effect of a far bigger problem, e.g. global thermonuclear war or a hypermassive solar eruption.

In theory - but if there's one thing leap seconds and GPS rollovers and suchlike have taught me, it's that once-every-few-years situations aren't always robustly supported even if they're entirely foreseeable.

Maybe most equipment purchased in 2019, from top-tier vendors, with all the software options unlocked.

But that's several steps away from what's actually deployed in the world right now.

A tremendous amount of the telecommunications network uses BITS sync equipment deployed in the late 90s, because it was extraordinarily well built, extremely reliable, and supported over the long term. There's nothing wrong with it, but it's single-constellation.

Ditto the power generation and distribution industry, mostly deployed in the early 2000s. ISOs rely on frequency measurements that come from GPS-synced network monitoring hardware. Again, high-end industrial stuff, reliable and well-supported, but single-constellation.

Ditto WISP PTMP hardware, mostly deployed in the early 2010s. Heck, even the current stuff is still only single-constellation because that's all it needs to be.

Thinking about it further, I think your statement applies to cellphones and very little else.

My statement also applies to modern PTP servers, e.g. https://www.heoldesign.com/T108-GPS-GLONASS-BEIDOU-GALILEO-T... If you are in the market for one, IMHO nowadays multi-constellation should be a baseline requirement.

If devices have other radios such as WiFi and GSM these can function as a backup plan (I don't think BT would suffice, unless there would be P2P between Bluetooth devices with devices who can provide coordinates such as WiFi and GSM).

For example, the first iPhone used Apple WiFi positioning; it didn't have GPS. I still remember I had a an iPod Touch with a 3rd party GPS which worked with the predecessor of Cydia.

Check the sources UnifiedNlp [1] can use. Quoting:

> AppleWifiNlpBackend - Uses Apple's service to resolve Wi-Fi locations. It has excellent coverage but the database is proprietary.

> OpenWlanMapNlpBackend - Uses OpenWlanMap.org to resolve user location but the NLP backend did not reach release-quality, yet. Users interested in a freely licensed and downloadable database for offline use should stick with openBmap for now - Last updated in 2015

> OpenBmapNlpBackend - Uses openBmap to resolve user location. Community-created, freely licensed database that can optionally be downloaded for offline operation. The coverage varies from country to country (it's best in central Europe).

> MozillaNlpBackend - Uses the Mozilla Location Service to resolve user location. The coverage is OK. Only the cell tower database is free.

> LocalWifiNlpBackend - Local location provider for Wi-Fi APs using on-phone generated database.

> LocalGSMLocationProvider - Local opencellid based location provider backend. Has been surpassed by LocalGSMBackend which also has an OpenCellID option - Last update in 2014

> LocalGSMBackend - Local location provider for GSM cells. It works offline by downloading freely licensed database files from Mozilla, OpenCellID, or lacells.db.

[1] https://github.com/microg/android_packages_apps_UnifiedNlp

I think a lot of cellular technologies rely on GPS on their base stations as a timing reference. Most cellular capabilities might fall off a cliff if GPS isn’t available.

Right. They have a few hours of holdover before things start to lose sync, and when sync degrades, the towers may stay on the air but handoffs get real rough.

They may also take themselves off the air after too long in holdover, but that's a software setting. Conceivably this could be overridden by network operators (in close cooperation with their equipment vendor) in a prolonged GPS outage where it was determined that having the towers on-air as islands without handoff would be better than nothing.

Why though? If the celltower has networking (e.g. 4G), it could use NTP. The databases with GSM just use triangulation. As long as they broadcast, the triangulation should work.

Best case NTP is in the order of millisecond precision. Completely inadequate for this case.

Any idea why wifi isn't leveraged more often?

Wasn't there a huge deal about Google maps cars collecting wifi ssids? I never understood why this was a privacy issue.

> I never understood why this was a privacy issue

e.g. if you have a rather unique SSID, and someone knows it (say you posted a screenshot of your phone's settings), these databases of access points would pretty much point them to your home address

1. unreliable + imprecise 2. requires periodic map updates 3. when you actually need it, Murphy guarantees visibility to zero APs

OSM and Mozilla both allow to contribute easily to such databases.

That's orthogonal to the issue. GNSS gives you a fix anywhere; wireless triangulation gives you a guess in some populated areas (note all the three qualifiers). Moreover, GNSS gives you a 4D fix: [X,Y,Z,t]; triangulation gives you 2D at best: [~X, ~Y].

Worse, easy contribution to a database is still way off from ease-of-use: now the database needs to get to the end devices, which brings in a whole tangled ball of dependencies.

Also, last I checked, OSM doesn't deal in wireless maps.

I think the headline is trying to imply that this is a big number, but at 1.8% of the world's daily GDP it doesn't seem like it. I would have thought GPS would have a lot more impact than that.

1.8% of global GDP is huge!

It's 0.005% of GDP.

You said one day of GPS is 0.005% of one year GDP, but we're comparing a day of GPS to a day of GDP, so it seems proper to say it's 1.8% like your earlier quote, right? If we lost any length of time of GPS, it would be 1.8% of that period's GDP?

Reply on news.ycombinator.com