[ntp:questions] Side effects of using prefer on server?

Markus Schöpflin no.spam at spam.spam
Wed Sep 14 12:20:25 UTC 2011


Hello,

the following behaviour has been observed using ntpd 4.2.6p3 on all involved 
servers and clients.

Given the following configuration on the client:

---%<---
# driftfile
driftfile /var/lib/ntp/drift
# keys
trustedkey 1
requestkey 1
keys /etc/ntp/keys
# server
server WRB1 version 4 iburst minpoll 5 maxpoll 5 prefer
server WRB2 version 4 iburst minpoll 5 maxpoll 5
--->%---

I'm observing that when the preferred server becomes unreachable, it takes a 
really long time to switch to the second server:

ntpq> peer
      remote           refid      st t when poll reach   delay   offset    disp
==============================================================================
*WRB1            ntp1             3 u  365   32    0     0.08   -0.219 7940.09
+WRB2            ntp1             3 u   19   32  377     0.08    0.081    2.33
ntpq> peer
      remote           refid      st t when poll reach   delay   offset    disp
==============================================================================
  WRB1            ntp1             3 u  424   32    0     0.08   -0.219 15937.5
*WRB2            ntp1             3 u    9   32  377     0.08    0.198    1.52

When removing the prefer keyword, I get the following:

ntpq> peer
      remote           refid      st t when poll reach   delay   offset    disp
==============================================================================
*WRB1            ntp1             3 u  238   32  300     0.08   -0.047  442.04
+WRB2            ntp1             3 u   32   32  377     0.08    0.040    2.84
ntpq> peer
      remote           refid      st t when poll reach   delay   offset    disp
==============================================================================
+WRB1            ntp1             3 u  247   32  200     0.08   -0.074  942.00
*WRB2            ntp1             3 u    6   32  377     0.08    0.040    3.10

This surprises me a lot, as per 
http://www.eecis.udel.edu/~mills/ntp/html/prefer.html#prefer using prefer has 
an effect on the cluster algorithm (phase 4), but shouldn't phase 2 result in 
the preferred server being already excluded because of its zero reachability 
and horrendous dispersion?

Regards,
Markus




More information about the questions mailing list