37

I've come across a handful of GPS NTP servers, as well as some inexpensive solutions using off-the-shelf receivers and software. Right now I'm just using NTP with a list of servers over the Internet. What is the advantage to using GPS instead (considering the alternative is free)?

6
  • 5
    Geek bonus points?
    – MathewC
    Jun 18, 2009 at 19:54
  • It is amazing to me how much BS was pumped in this thread. Stratum 0 indeed. At the OP. If you have an operation that depends on a highly accurate time source then invest in a good Stratum 2 / 3 clock, based on the accuracy required. Pay attention to holdover specifications.
    – dbasnett
    May 10, 2010 at 14:58
  • 3
    @dbasnett Erm, I think you are actually misinformed... There's no such thing as a "Stratum 2 / 3 clock", all hardware sources are inherently Stratum 0. Perhaps a quick read through RFC 958 would be useful.
    – Chris S
    Oct 16, 2012 at 17:08
  • See RFC 1305. You might want to check out the USNO site and any number of manufacturers (symmetricom.com) They obviously did not get the memo about there not being stratum 2/3 clocks.
    – dbasnett
    Oct 17, 2012 at 13:07
  • @dbasnett I think you might want to have another look at the RFC 1305: Stratum (sys.stratum, peer.stratum, pkt.stratum): This is an integer indicating the stratum of the local clock, with values defined as follows: 0 unspecified; 1 primary reference (e.g., calibrated atomic clock, radio clock); 2-255 secondary reference (via NTP) which indicates that stratum 2+ is not a clock source per se, but rather clock that references other NTP clock. Whatever you saw on symmetricom page is taking time from NTP and passing it down the net. Section 3.2.1 of rfc-editor.org/rfc/rfc1305.pdf Mar 17 at 16:35

12 Answers 12

13

I've used them at a gambling operation so we could keep accurate time but still have a true air gap between the production network and the internet/corporate network.

With the revenue generated per hour and that on one occasion a single remote sales rep came into the office and plugged in his infected laptop to the corp network and brought it down within minutes (production was unaffected). The effort of using things like gps time to allow for an air gap was worth it.

2
  • If you are on an air gap, how do you obtain the GPS time? Wouldn't that require connecting to a network?
    – Pacerier
    Dec 20, 2014 at 15:22
  • 1
    GPS signals are one-way radio transmissions from satellites, so no external network connection is required.
    – Kaypro II
    Feb 17, 2016 at 21:20
12

With a GPS ntp server you'll be getting a stratum 0 clock source. the advantage is accuracy and lower jitter. The downside is cost. There are very few circumstances that require this level of accuracy. in general you want to make sure your systems are in sync with each other. So syncing all your servers against a pair of internal ntp boxes that sync against external lower stratum ntp servers is usually sufficient.

The only time I've been involved in using a GPS ntp box (for a PKI system), we never actually used them. We couldn't get permission to mount the antenna on the roof of the datacentre and get the cabling back to the ntp box.

3
  • Stratum 0??? Not likely.
    – dbasnett
    May 10, 2010 at 14:47
  • 7
    @dbasnett, it's considered Stratum 0 if the source is hardware based (as GPS is). That doesn't guarantee accuracy, but there's no inherent jitter from using a computer network like the Internet. Note: the computer attached to the hardware is still Stratum 1 (the lowest a NTP Server can be).
    – Chris S
    Oct 18, 2010 at 16:09
  • The Stratum is really the distance from the reference clock and I agree that it has nothing to do with quality. In the past we chose clocks based on Stratum and what would happen if we lost the reference(GPS). The clocks we chose all had Rubidium backups.
    – dbasnett
    Oct 18, 2010 at 21:45
8

We do have a pair of stratum 1 time servers using GPS for NTP.

Does it make a difference in our day to day business vs when we had no NTP servers? A little, for example, on the client support side, the phones, both digital and voip and all the desktops now have the exact same time. Which cuts down on all the why do all these clocks have different times and can I fix it questions.

On the server side, when we aggregate logs we know that our timestamps are all the same. So if something happened at 12:37:15.34 on two servers we know that it really was at the SAME TIME, not 30 seconds to a minute appart.

Did we need GPS based NTP to do this? Probably not, we could have used any NTP source. We know everything is synced perfect and the time never drifts. Also if our internet connection is down, or laggy it's still perfect. Any NTP would have been an improvement over the lack of NTP we had though.

1
  • 3
    "A man with a watch knows what time it is. A man with two watches is never sure." With two clocks it is impossible to identify a false ticker. But don't take my word for it, Section 5.3.1 of the ntp faq: "This is actually the worst possible configuration -- you'd be better off using just one upstream time server and letting the clocks run free if that upstream were to die or become unreachable."
    – dfc
    Nov 11, 2012 at 7:38
4

While not used for time sync so that all devices are using the same time for logs, GPS time sync is commonly used in carrier class wireless transmission hardware. It allows access points to sync and all transmit/recieve simultaneously, so that they won't be causing interference or raising background noise when listening for transmissions from clients.

3

The cost is not small (relatively) and the benefits are few. The only places I've seen running their own stratum-0 time source had extremely high expectations for the quality of the input, and a lack of trust in external time sources.

5
  • 1
    There ARE advantages. Otherwise no one would run them. Jun 18, 2009 at 20:24
  • 2
    I eluded to the two advantages in my post. A lack of trust in the input and a lack of trust in the external infrastructure. But, I've only seen those in .mil and financial services sectors.
    – dr.pooter
    Jun 18, 2009 at 20:29
  • 1
    Alluded ... and who's to say he's not in either? +1 for trying. Jun 18, 2009 at 20:34
  • 1
    I presume he's not in either of those sectors because he's currently syncing to NTP servers on the Internet and is only now considering this solution. If he were in those sectors, standards would have likely dictated that he not do that. Thanks for the spelling correction.
    – dr.pooter
    Jun 18, 2009 at 20:40
  • Unless you have toured the USNO it is not likely you have seen a Stratum 0 clock.
    – dbasnett
    May 10, 2010 at 14:53
2

Great answers so far! I can say that I've seen GPS time sources used on multiplexers, and certain types of crypto ... it was used for NTP, just because it was already there.

The advantage for using a local GPS NTP server would be

  • High Accuracy
  • Not dependent on an Internet connection

How much does one of these things cost anyway? It shouldn't be that expensive. GPS + Linux + NTP ...

Wow, $1000 - $3000, do you need it that bad?

4
  • 1
    I just really really like accurate time. So far I've had zero luck convincing TPTB that my accurate time fixation is worth that much.
    – sysadmin1138
    Jun 18, 2009 at 23:44
  • 2
    If it's just because you really like accurate time, your fixation is worth $0 in this sort of economy.
    – ceejayoz
    Jun 19, 2009 at 0:10
  • 2
    $3000 is about the cost of Windows 2008 Server Enterprise with 25 CALs, RHEL AP for one server with 24/7 support, or (fully loaded) a $100k/year employee for four days. If you require accurate time in a non-Internet-connected network, and your PHBs won't give Symmetricom what they're asking, please try to avoid slicing a wrist open on their pointy hair -- it's quite sharp. Dec 4, 2009 at 17:59
  • Well said! +1 For putting it into perspective. Dec 9, 2009 at 11:56
2

A couple years back I looked at whether it was possible to use a cheap GPS (~$200-$300) to get accurate time on a server. I'm sure that at the time Garmin or one of the other companies had software you could run on Windows to read the time on a computer to read the time from a GPS connected to a serial port. I was going to ask if anyone has used this as a cheap alternative to a purpose-built NTP server. It should be accurate enough and only a few hundred bucks.

2
  • Nope. The delays introduced with such equipment would give a lower precision than just using any ol' NTP server out there on the Net; the price tag of a purpose-built GPS receiver is not just for the pretty logo. Mar 11, 2012 at 11:59
  • Yes. Though Windows makes a poor NTP server because of the "tick" frequency (more or less analogous to kernel ticks in the *nix world). A Garmin 18x LVC with a little DIY soldering can be attached to a serial port for well under $100. The BSDs make the best choice of the widely used OSes (roughly nanosecond precision), though Linux is still within a millisecond (good enough for most applications).
    – Chris S
    Oct 16, 2012 at 17:21
1

A neutrino detection experiment I worked on used GPS for the primary time signal to our Data Acquisition system. They used a fairly expensive system supplied by someone in Japan.

This as part of a world-wide effort to provide high precision information about the different arrival times for the neutrino and photon wave-fronts from big supernovas. Lots of cool physics in that, if you can get precise enough data.

1

We use them because of their "deadman's" feature (won't correct more than a few seconds so that if something goes wrong your whole infrastructure does not move 12 hours into the past) and security; NTP is totally insecure; spoofing, man-in-the-middle, etc.

2
  • Have you actually tried faking ntp answers? At least with the unix reference implementation, that will be quite hard. NTP default is hardened and you can add cryptographic signatures to add authorisation. May 3, 2011 at 11:34
  • Note that GPS isn't "secure" either: there are constellation simulators out there, which would allow manipulation of the signal as desired. Mar 11, 2012 at 11:56
0

It gives you a Stratum 0 NTP server, under your control. You can then configure a few machines as Stratum 1 servers, syncing from your GPS-clocked server and have slightly less traffic needing to leave your network.

It's been around 10 years since I last had one of those (that I know of) in my network, but at that point, we didn't have a good antenna position and while it mostly saw a GPS signal, it did occasionally lose the signal, requiring a reset of the receiver equipment. With a better antenna placement, that would most probably not have occured.

2
  • 1
    There's no such thing as a "Stratum 0 NTP server". Stratum 0 is a hardware source outside NTP. The NTP server it's connected to is thereby Stratum 1 since it's 1 step away from the reference clock.
    – Chris S
    Oct 16, 2012 at 17:12
  • @ChrisS I've always interpreted the strata as being specifically only within NTP. But I am happy to change my understanding. Thank you.
    – Vatine
    Oct 17, 2012 at 12:15
0

The advantages are you can get real seqence of events logging.

An example is in critical infrastructure. All electrical utilities should be using stratum 1 time for their databases. Stratum 1 time is critical to working out what made something fail on the scale of local utility downtime, to figuring out who caused the Northeast Blackout of 2003. NTP is often passed up in favor of PTP (IEEE 1588) for synching to the microsecond level.

0

At my old job we used GPS for timing a lot of different equipment. each piece of equipment had their own GPS receiver, which allowed perfect sync where latency was not an issue. Using GPS practically eliminates network-factors in the accuracy.

You must log in to answer this question.