[SunHELP] NTP design

[Jim Pennino]

On Fri, Jul 19, 2002 at 07:54:35AM -0700, komokwa at earthlink.net wrote:
> >A technical nit; if the GPS netclocks are >network ntp devices, they will be
> stratum 1 and the 4 solaris machines will be stratum 2, 
> 
> My understanding is the actual clock (source of time) is stratum 0...but
> probably a minor point.

In the olden days before GPS ntp appliance boxes, you hooked some sort
of time standard such as a WWV or CHU receiver to a system that ran ntp.

In that case, I guess you could say the receiver was stratum 0, but the
computer system was stratum 1.

Todays network connectable GPS ntp servers look like the systems of old,
i.e. they are stratum 1, but now everything fits in a 1" rackmount box.

> >The only practical reason I've found for setting >a preferred server is to
> manage network traffic 
> 
> Traffic management and time accuracy was the logic behind specifying local
> servers for preferred use at each data center.

After things have been running for a few minutes, and given stable
connections, clients will at worst case connect once every 64 seconds.

There will not be a lot of traffic if things are stable. On the networks I
deal with the local clients update every 1024 seconds across the WAN (T1).

The most important criteria for accuracy is stability in the path to the
stratum 1 server.

Even with a crappy network you are going to have sub-second accuracy.

> >Under normal circumstances, the Solaris
> >boxes will automatically pick the local GPS box >as its preferred server.
> 
> How does it do this?

The short smart-ass answer is "quite well". The technical answer is a bit
much for e-mail.

For a real in-depth answer, go to http://www.eecis.udel.edu/~ntp/ and poke
around. You will find more information on timekeeping and how ntp works
than you ever wanted to know.

> >The local clients need only point to the local >Solaris boxes 
> 
> Are you suggesting to not set the 4 NTP servers as peers?

There is nothing to be gained by setting the 4 Solaris boxes as anything
other than clients to the 2 GPS boxes, and nothing to be gained by setting
the local clients as anything other than clients to the 2 local Solaris boxes.

The 2 GPS boxes are the ultimate sources and, assuming they are the same,
there is no reason to prefer one box over the other. Any network issues
will be handled automatically by ntp on the Solaris boxes.

The sole reason to point all 4 Solaris boxes to both GPS boxes is in case
one GPS box fails.

The peer and preferred server stuff is a leftover from the old days.

As an example, say at site A you had a system hooked to a WWV receiver, at
site B you had a system hooked to a CHU receiver and at site C you had a
system hooked to an "atomic standard", i.e. a cesium or some such thing.

Then you would do peering and make site C the preferred server since the
"atomic standard" would normally be a better time source.

So, yes, unless you have something better than a GPS box somewhere, don't
try to over think this, keep it simple. Like I said, even with a crappy
network, you are going to get sub-second accuracy with these GPS boxes.

Now, if you are doing timing in the nanosecond range, you have an entirely
different problem...

> 
> Thanks very much,
> -- Tim
> _______________________________________________
> SunHELP maillist  -  SunHELP at sunhelp.org
> http://www.sunhelp.org/mailman/listinfo/sunhelp

-- 
Jim Pennino