[ntp:questions] Stratum 0 server rejected by clients

[Danny Mayer]

Dirk Langner wrote:
> Hello everybody, 
> 
> We're facing a problem, that we have ntp-servers with synchronized
> system time by an external clock. The ntp servers are now distributing
> the (synchronized) time to all other servers. This works well at the
> first time, the ntp-servers are seen as peers on most of the clients,
> sanity checks are passed. But it seems that after a while the ntp
> clients are going to reject all their servers due to not passing the
> sanity checks, which means, that the time is not synchronized any more
> on those systems.
> 
> The paper "Using NTP to Control and Synchronize System Clocks - Part
> III" by Sun blueprints Online says in NTP Algorithms/Sanity Checks, step
> 6: "...that servers, which are not connected to a root time source, will
> be ignored." So is this now the reason why the ntpclients are going to
> reject the servers or am I missing something? And if so: is there any
> chance to get the system time (which is synchronized - see below)
> distributed via NTP?
> 

The proper source are the documents on the ntp.org web sites.

> Some words to the setup: 
> Four ntp-servers ntpserver[1-4] are synchronized via serial lines to a
> single external timesource.

Why would they be synchronized to a *single* external timesource? The
recommended guidelines is at least 3 and preferably 4.

 On these system a deamon (which does mainly
> other things) adjusts the system time according to these serial signals
> at least every 10 minutes.

What daemon? If it's not NTP then it's all over.

 I can't switch to other / external ntp
> servers, since I need this one and only time (broadcast environment) and
> the servers are sitting on an isolated network. The external device is a
> LTC timecode converted by a Miranda Littlered, the serial signals are
> processed by a deamon, which is exclusively connected to the serial
> port.

Is this supposed to be a refclock? Does it have a refclock driver?

> 
> ntpserver1 8# more /etc/ntp.conf
> # Generic ntp.conf file
> #
> # See documentation for more options
> #
> # add as many server as required
> server 127.127.1.0
> fudge 127.127.1.0 stratum 0 refid LTC1

DO NOT DO THIS. Stratum 0 is reserved for refclocks and your server is
not a refclock. The local clock like this is for systems that lose their
 connection to a higher stratum clock. Usually this is set to 10 for a
local clock.

> 
> broadcast 192.168.58.255 ttl 2
> 
> driftfile /etc/ntp.drift
> 
> # don't set local clock
> disable ntp
> 
> This last setting is necessary to disable the setting of the system
> clock by ntp, the system time of the ntp-server shall exclusively set by
> the deamon. 
> 

That's not necessary as it cannot set the clock anyway with this setup.

> The ntp clients listen to only these servers.
> ntpclient 14# more /etc/ntp.conf
> # Generic ntp.conf file
> #
> # See documentation for more options
> #
> # add as many server as required
> server ntpserver1 prefer

Why is this one preferred over all the others?

> server ntpserver2
> server ntpserver3
> server ntpserver4
> 
> driftfile /etc/ntp.drift
> 
> # monitor no
> # authenticate no
> #
> broadcastclient yes

The only argument that broadcastclient takes is novolley.

> # multicastclient
> 
> Especially if the ntp deamon is started on one ntp-server, this server
> is picked up as a peer (or at least a candidate). 

You didn't specify any peers so what do you mean by that?

> ntpclient 3# ntpq
> ntpq> pe
>      remote           refid      st t when poll reach   delay   offset
> jitter
> ========================================================================
> ======
> *ntpserver1 .LTC1.           1 u   20   64  377    0.502  -32.966
> 0.258
> +ntpserver2 .LTC1.           1 u   25   64  377    0.441  -32.922
> 0.155
> +ntpserver3 .LTC1.           1 u   51   64  377    0.511  -33.044
> 0.224
> -ntpserver4 .LTC1.           1 u   11   64  377    0.431  -33.129
> 0.098
> version="ntpd 4.1.1 at 1.786 Wed Jan 19 14:18:56 PST 2005 (1)",

That's a little old. The latest release version is 4.2.2.

> 
> After some time the servers are going to be rejected, although the
> sanity checks are passed. The associations are going to be 
> 

What makes you think that the sanity checks are passed? What evidence do
you have? All the indications are that they did not pass.

> ntpclient 2# ntpq
> ntpq> pe
>      remote           refid      st t when poll reach   delay   offset
> jitter
> ========================================================================
> ======
>  ntpserver1 .LTC1.           1 u   44   64  377    0.033  88339.7
> 28.006
>  ntpserver2 .LTC1.           1 u   48   64  377    0.064  88338.0
> 27.700
>  ntpserver3 .LTC1.           1 u   59   64  377    0.059  88333.4
> 27.613
> ntpq> ass
> ind assID status  conf reach auth condition  last_event cnt
> ===========================================================
>   1 60236  9034   yes   yes  none    reject   reachable  3
>   2 60237  9034   yes   yes  none    reject   reachable  3
>   3 60238  9034   yes   yes  none    reject   reachable  3
> ntpq> readvar
> status=00f4 leap_none, sync_unspec, 15 events, event_peer/strat_chg,
> version="ntpd 4.1.1 at 1.786 Wed Jan 19 14:18:56 PST 2005 (1)",
> processor="IP35", system="IRIX646.5", leap=00, stratum=16,
> precision=-18, rootdelay=0.442, rootdispersion=4115.402, peer=0,
> refid=ntpserver1,
> reftime=c9249b12.cc18c5c9  Sat, Dec  9 2006  3:16:50.797, poll=4,
> clock=c927cc57.727e846a  Mon, Dec 11 2006 13:23:51.447, state=2,
> offset=0.000, frequency=-259.475, jitter=13.014, stability=213.142
> 

Danny