[ntp:questions] Stratum 1 GPS Network Time Servers

Chris Albertson albertson.chris at gmail.com
Thu May 12 15:11:44 UTC 2011

I think there are two issues that people confuse

1) Availability.  This means that your clients are able to access an NTP
server.  For this I think two servers are enough because you can always fall
back to Internet pool servers.  You cgain some here if the two servers are
not clones of each other in the same rack were one problem can take down
both at the same time.   Maybe even put them in different buildings.

2) Assurance that the UTC time you are serving is correct.  You can have 10
working servers all of which server the time wrong time.  The best way to
assure time is collect is to get it from multiple (at least three)
independent sources.  I think three different brands of GPS receiver would
do.  Two is not enough.  If they differ which is wrong?   Three is the
minium, others will argue for more so that you still have three after a

I think you can do the above with three servers each with their own GPS
receiver but make the two servers use different hardware and different
brands of GPS  Each client is then configured to look at three of your
servers.  The clients them selves will determine if one of your servers is

You don't need an expensive computer.  An older computer works fine even an
old 1Ghz Pentium is over kill.  Just make sure it has a realRS232 serial
port andnot a USB/Serial converter.  Older PCs are more likely to have this.
 Good GPS receivers cost well under $100.  These are cheap enough you could
stand up three.   I found some new computers that use the Atom CPU and pull
about 10W of power.  I put a small notebook disk drive in, no heat and no
noise and I'm running at the handful of microseconds level with a oldet
Motorola UT+ GPS Receiver.

On Thu, May 12, 2011 at 12:14 AM, Alan Decelis <alandecelis at gmail.com>wrote:

> In the context of a requirement to set up a redundant configuration of
> internal (peered) GPS Network Time Servers that operate at Stratum 1, is
> there a best practice of the number of GPS Network Time Servers to set?
> Would it be risky to rely on two reliable (meaning rather renowned and
> expensive) GPS Network Time Servers each connected to a separate GPS
> antenna
> with a TCXO? We are talking of a configuration whereby the number of ntp
> requests to these GPS Network Time Servers will be low (10 to 20 nodes at
> Statum 2 with 10 to 20 requests every 5 to 15 minutes) but the context
> requires a high level of security (thus eliminating connections to the
> internet) and a moderate level of accuracy (millisecond).
> http://support.ntp.org/bin/view/Support/SelectingOffsiteNTPServers#Section_5.3.3
> ,
> in line with how NTP functions, would suggest a minimum of four upstream
> servers. My understanding of this is that this applies to a scenario
> whereby
> external time sources are being used, for instance official time sources
> over the internet.  Does this also apply to the scenario specified
> above? Would I thus require to look at the option of having 4 GPS Network
> Time Servers at Stratum layer 1
> Thanks in advance
> _______________________________________________
> questions mailing list
> questions at lists.ntp.org
> http://lists.ntp.org/listinfo/questions

Chris Albertson
Redondo Beach, California

More information about the questions mailing list