[ntp:questions] Help appreciated

Towli towli at oldboyz.dk
Wed May 13 09:06:32 UTC 2009

"Dave Hart" <davehart at gmail.com> skrev i meddelelsen 
news:eee8cec8-5004-49b3-b708-16c6dc1c50c4 at y6g2000prf.googlegroups.com...
On May 12, 8:41 pm, "Towli" <to... at oldboyz.dk> wrote:
> I think i may have the answer to my own question above...
> I read somewhere that a man with 2 watches never know what time it is...
> My switches sees the 2 external NTP-stratum1-servers as being closer to 
> each
> other in time (due to their physical placement close to each other and
> relative to my core switches).
> Therefore my core switches choose to trust the to more agreeing servers as
> opposed to my internal server, which is quite alone in its view on the 
> time
> and whose time differs from the 2, due to the closer proximity /smaller
> delay.
> Am i right in this assumption?
> I removed 1 of the external NTP servers and now my internal DCF is used as
> prefered Ntp server.

I think you answered your own question once again ;)

> Should i keep 1 of the external stratum-1-servers for backup or should i
> remove it all together? My Dcf77 sometimes looses connection to Frankfurt
> and when this happens it looses it stratum 1 value, and hence my core
> switches looses synch...

Yes, I would keep an external backup, probably even several, but
remote from each other to avoid the problem with nonoverlapping
confidence intervals you saw.  Also test the configuration with just
one local stratum 1 server even after you get your second one
operational, to be sure that losing one of those local stratum 1
servers will not cause behavior you don't want.

I would also suggest you consider using hosts running NTP version 4,
rather than switches running an adaptation of NTP version 3.  The
switches are designed and optimized to shuffle packets, not keep
time.  Precision of 2**-6 means the switches are stepping time 64
times per second, or every 15.6 milliseconds.  You shouldn't rely on
such crude machinery as the local distribution mechanism.  ntpd easily
coexists with most other software and has very light resource
requirements, even serving large numbers of clients, so it should be
possible to find a few suitable servers.  Alternatively, you can find
NTP appliances or build one yourself with a $200 Soekris or similar
embedded platform.

Dave Hart

Dave -your help is very highly appreciated - thank you! -I have talked to 
quite a few certified this-and-that and Googled a lot, but Ntp seems not to 
be a field very many people knows a lot about!
What i dont get ,is that the precison of 2**-6 seems to steem from which is the DCF77 appliance box (that is - not a switch) on 
my internal network. How come it has this low precision ? is it bad hardware 
could it be bad network connection/ bandwidth?

Below is an output of one of my client switches, getting its time from the 
core switches (which then in turn is synching with the mentioned aplliance 
box). -Here the precision is better (?).

Thanks again

/Towli - Denmark configured, our_master, sane, valid, stratum 2
ref ID, time CDB50AD3.E971F741 (11:01:07.911 CEST Wed May 13 
our mode client, peer mode server, our poll intvl 64, peer poll intvl 64
root delay 3.10 msec, root disp 10.41, reach 377, sync dist 13.687
delay 1.80 msec, offset -0.4984 msec, dispersion 0.49
precision 2**18, version 3
org time CDB50B39.98E64C23 (11:02:49.597 CEST Wed May 13 2009)
rcv time CDB50B39.9942006A (11:02:49.598 CEST Wed May 13 2009)
xmt time CDB50B39.98CAD852 (11:02:49.596 CEST Wed May 13 2009)
filtdelay =     1.80    1.74    2.67    2.99    1.43    1.59    2.11    4.24
filtoffset =   -0.50   -0.40    0.33    0.66   -0.02    0.42    1.12   -0.97
filterror =     0.02    0.99    1.97    2.94    3.92    4.90    5.87    6.85 

More information about the questions mailing list