Recommended Number of NTP Servers

Hal Murray hmurray at
Mon Nov 4 11:32:18 UTC 2019

rlaager at said:
> Can you expand on the meaning of "should" here? Are you saying what you think
> the current state is, or what you think the desired state is? 

Here is the get-more chunk of code:

                if ((peer_associations <= 2 * sys_maxclock) &&
                    (peer_associations < sys_maxclock ||
                     sys_survivors < sys_minclock))
                        if (!dns_probe(peer)) {

peer_associations counts the number of peer blocks allocated.  That includes 
pool and refclocks.

sys_survivors is the count of not-way-out slots.  That won't include pool but 
will include refclocks.

The first line is a sanity check to avoid crazy cases with lots of bogus slots 
or lots of not-ready-yet slots.  (Read the comment in ntp_proto.c if you want 
the long story.)

So it should get at least minclock working clocks and won't get more if 
working plus in-progress plus pool is >= maxclock.

So it looks like we are both sorta right.  It uses working slots on the min 
test but counts pool on the max test.

The key idea is that there is an acceptable range between minclock and 

These are my opinions.  I hate spam.

More information about the devel mailing list