[time-nuts] NTP stratum 1 appliances with different (GPS, etc) cores...

Chris Albertson albertson.chris at gmail.com
Mon Mar 9 21:24:37 EDT 2015


The exact architecture you need depends a lot on how your company is
geographically distributed and the network that connects those parts.

If you have decided you want to run your own stratum 1 NTP servers
then you'd place a few of those in each plant or in each city.  Then your
next level servers, one server per local network can look at ALL the
stratum 1 servers plus some others out on the Internet.

The way to keep this RELIABLE is through e config files.  Make sure that
EVERY computer that uses NTP has at least FIVE reference clocks.  Some
company servers then some Internet pool servers.

Make sure to use authentication.  This will keep spoofers out

The other thing that can help reliability is use of orphan mode.  This is a
fall back mode where a group of NTP servers gets disconnected from the
outside world.  Then they can use each other as a reference clock and keep
and kind of consensus going.

Also remember when I say "NTP Servers" I mean anything that runs NTP which
should be just about anything including routers and file servers.  All of
these need at least five reference clocks that are widely distributed and
to be configured to fall back on Orphan Mode..

Reliability starts near the end users, right on their computers.  Configure
each one to look at many other clocks and those clocks to look at many
other.  You want the fan out be very wide from the start.   Orphan networks
can work well if part of your network is isolated from the rest of the
world if there are many computers.  The ones with good internal clocks will
all tend to agree with each other and drive the others.  Then when Internet
service is restored to the building or lab better clocks will become
available.

On Mon, Mar 9, 2015 at 1:31 PM, Robert Seastrom <rs-lists at seastrom.com>
wrote:

>
> Hi folks,
>
> Had a recent requirement at $DAYJOB to give our NTP architecture a good
> swift kick in the pants.  High precision is not a particular requirement (1
> ms is more than adequate) but high reliability is.  I'm about done with
> explaining the difference between the two to people internally who don't
> get it.  :-(
>
> Unfortunately, these days it's impossible to get sufficient out-of-band
> sources for stratum 1 clocks unless we're building our own constellation of
> primary frequency standards (Not Gonna Happen).  So I definitely intend to
> have our stratum 2 clocks (which customer and production devices will be
> pointing at) chiming off of plenty of off-site stuff too, hopefully things
> that are connected physically to a diverse set of primary sources (i.e. not
> gps or wwvb)
>
> This approach would probably be fine for our internal consumption,
> frankly, considering our need for reliability but not supreme accuracy.
> But there's still a desire internally to run a set of stratum 1 radio
> clocks.  In the course of this discussion, I invoked a particularly nasty
> "conspiracy" that happened back in 2002 wherein every TrueTime NTS-100
> (which were extremely popular in those days) simultaneously decided that it
> was 20 years in the future and warped the time.  That's not the only case
> of devices losing their cookies; I seem to recall a problem with certain
> Motorola cores having an issue with leap seconds.
>
> Anyway, I couldn't keep my mouth shut and suggested that at the very
> least, trying for multiple vendors (we have a strong preference for
> appliancey stuff here due to internal fiefdoms, ownership issues, and
> balkanization typical of a Fortune 200 company) with different GPS cores
> would be a really fine plan.
>
> So of course, now it's my job to come up with proper nominees for this
> list.  Any suggestions would be greatly appreciated.  Hopefully someone's
> already done the legwork for this.  I will likely send this same request to
> the ntp dev list...
>
> Thanks,
>
> -r
>
>
> _______________________________________________
> time-nuts mailing list -- time-nuts at febo.com
> To unsubscribe, go to
> https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
> and follow the instructions there.
>



-- 

Chris Albertson
Redondo Beach, California


More information about the time-nuts mailing list