When a GPS disciplined time server or server appliance encounters a condition where the GPS signal becomes unavailable or unreliable a local oscillator is used to keep time. The local oscillator is generally a TCXO, OCXO, or Rubidium standard. Such systems are able to deliver fairly accurate and precise time in the event of a short outage like this. A quality server with a local Rubidium oscillator should be able to keep accurate time within 10 μsec in the event of a five day outage.

GPS signals can become unavailable for numbers reasons that are far more mundane than being jammed such as someone disconnecting the antenna, lightning strike, a rat chewing through the cable, meth heads stealing the copper cable and/or antenna, etc.

A few months ago one of the satellites was still broadcasting while it was down for routine maintenance, reporting some time that was off by a third of the distance to the sun.

There are enough weird GPS errors that I'd hope they have a backup plan already.

Wouldn't the GPS ephemeris/almanac data would still list the satellite as inop, even if it's still transmitting?

They did, so actual GPS location chips weren't thrown off. But apparently there are GPS-based clocks out there that don't actually bother to read the almanac.

(I'm a few layers removed from the error report where I saw this so I don't really know the details.)

How many NTP servers run off GPS? My understanding was that most people just piggy backed off pool.ntp.org and didn't bother running their own stratum 1.

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

  Grandmaster Clock [...] equipment designed for deployment
  in the mobile backhaul network, at macro base station
  sites, or small cell aggregation locations. [...]
  meet the stringent timing and synchronization
  requirements of 4G/LTE networks

In other words: Applications that need sub-millisecond-level precision, such as mobile phone base stations, where multiple towers coordinate their transmit/receive windows to avoid interfering with one another.

I'm currently building one for fun using a Raspberry Pi and a cheap GPS module from Aliexpress. I don't think a lot of hobbyists do it, but it's pretty common for stuff like datacenters to have their own local GPS disciplined NTP server.

It's not that hard to make one yourself by the way, and the hardware costs about $15. At the pro level you can get rack mount units with GPS + an on board TXCO/OXCO/rubidium frequency reference. You can probably get one with a cesium reference too, but I don't think that's much use outside a metrology lab (and will cost a fortune).

https://spectracom.com/products-services/precision-timing/en... edit: if someone knows about how much these cost please comment, I'm curious but I don't want to request a quote.

https://developers.redhat.com/blog/2017/02/22/how-to-build-a...

Using a raspberry pi as a GPS NTP timeserver greatly amuses me since the raspberry pi doesn't ship with an RTC (or at least last time I checked). As a follow up question, how does the system time get set on one of these time servers? Does the NTP process have rights to update local system time or does the system rely on requesting NTP from itself?

System time gets set either from the GPS signal if not connected to the network or another NTP server if you are at boot. The satellite gives you a coarse time directly (accurate to milliseconds I think), plus a high precision frequency reference driven by the onboard atomic clock, which NTP uses to keep time. Once setup, this thing is supposedly capable of stability in the microsecond range, which is better than you'd get relying on an external server (dunno if true yet however).

NTP can set your system time I believe.

Lack of a RTC doesn't matter in this case because you wouldn't want one, they typically aren't stable enough. You're not supposed to turn it off very often anyway. If you were more serious about it than me you'd probably add a rubidium oscillator to serve as a secondary reference when GPS is offline and hook the whole thing up to a large battery, so that it never turns off. I don't care that much though, I'm just having fun with cheap toys.

Interesting, I’m doing the same but using WWVB 60khz radio signal for my time source

Also cool. I opted for GPS because I had a spare Pi and you can get dubious but perfectly adequate GPS modules for $3 off of Aliexpress, it's kind of an impulse buy. I also picked up a MAX7219-based LED display for another $2, so it's also going to be an over-engineered desk clock for $15 + some 3D printed parts (the other $10 is for a Pi Zero W I had left over from a different project).

They’ll be sending fake signals, i.e time shifted. Since the speed of light is enormous, we’re probably talking micro- to millisecond disturbances. No worries for regular NTP servers I’d imagine.

if what they are doing is jamming, then if you are getting timing information from a gps receiver and your receiver is in range of the jammer you will get the timing information from the jamming signal, so yes, however the USAF _might_ be including the correct timing information in their jamming signal in order to reduce any side effects of their exercise...

Isn't timing the primary signal you get from GPS? Including a correct time would defeat the point of jamming.

SuperNAP is in that area. Only one of the largest colo’s in the country....no big deal

hopefully they have the necessary details from the USAF or have recently installed an atomic clock