[ntp:questions] Re: Getting good NTP tracking
Eino-Ville Talvala
quantumet at gmail.com
Mon Jun 26 21:18:00 UTC 2006
Richard B. Gilbert wrote:
> Eino-Ville Talvala wrote:
>
>> Ok,
>>
>> I've been messing with this for a few weeks, and haven't yet managed
>> to get NTP to where I'd like to be (and where I think it should be
>> able to be).
>>
>> I'm on an academic network, which has both a university-wide stratum 2
>> server (actually, a pool of a few servers), and a departmental stratum
>> 3 server. I'm trying to set up a pair of machines that I want
>> synchronized to each other at < 5ms consistently.
>>
>> Given that the delays between the machines and the servers are on the
>> order of 0.3 ms, I'd expect to be able to maintain offsets at less
>> than 5 ms consistently. However, while my average offset values are
>> usually in that range, I'm seeing RMS offsets on the range of 10-30
>> ms, with daily peaks around 90 ms for some servers.
>>
>> Both machines run Centos 4.2, and I've now disabled (temporarily) both
>> the firewall and SELinux protection for ntpd in an attempt to figure
>> out the problem. Here is the ntp.conf file for one machine (with
>> commented out bits removed, and anonymized)
>>
>> ---------------------------------------------
>>
>> restrict default nomodify notrap
>> restrict 127.0.0.1
>>
>> server <UNIVERSITY>
>> server <DEPARTMENT>
>> peer <SECOND MACHINE>
>>
>> # Let's throw in servers from the public pool
>>
>> server 0.north-america.pool.ntp.org
>> server 1.north-america.pool.ntp.org
>> server 2.north-america.pool.ntp.org
>>
>>
>> driftfile /var/lib/ntp/drift
>>
>> statistics loopstats
>> statsdir /var/log/ntp/
>> filegen peerstats file peers type day link enable
>> filegen loopstats file loops type day link enable
>>
>>
>> -------------------------------------------------
>>
>> And here's a bit of peerstats (from before I added in the public pool
>> servers):
>>
>> peers.20060620
>> ident cnt mean rms max delay dist
>> disp
>> ==========================================================================
>>
>> <UNIVERSITY> 132 -4.089 90.922 986.193 4.340 939.038
>> 30.380
>> <DEPARTMENT> 137 18.467 27.483 114.025 1.615 939.420
>> 32.474
>> <SECOND MACHINE> 137 24.696 29.110 59.226 1.236 938.365
>> 25.751
>> peers.20060621
>> ident cnt mean rms max delay dist
>> disp
>> ==========================================================================
>>
>> <UNIVERSITY> 84 5.482 21.357 89.828 7.645 45.069
>> 14.830
>> <DEPARTMENT> 85 10.523 10.672 34.443 4.559 23.647
>> 14.829
>> <SECOND MACHINE> 84 5.883 5.845 18.113 1.458 31.992
>> 20.572
>> peers.20060622
>> ident cnt mean rms max delay dist
>> disp
>> ==========================================================================
>>
>> <UNIVERSITY> 84 -2.083 115.705 931.227 7.427 46.999
>> 14.830
>> <DEPARTMENT> 84 4.506 7.546 41.621 2.652 39.808
>> 14.828
>> <SECOND MACHINE> 85 1.590 3.878 14.613 1.384 32.507
>> 18.043
>>
>> ----------------------------------------------------
>> And here's the driftfile:
>>
>> -96.459
>
> I'm a little puzzled by the low values of "cnt". With 84-85 samples
> per day, it looks as if your system is polling the servers at the
> maximum poll interval of 1024 seconds. Did you, by any chance, tamper
> with the default values of MINPOLL and MAXPOLL? It's generally a poor
> idea. Ntpd will adjust the poll interval upwards and downwards as
> conditions change and as limited by MINPOLL and MAXPOLL. The defaults
> allow for the conditions usually found. By setting MINPOLL to 10, you
> force the longer polling interval even when ntpd needs a shorter
> interval to achieve good synchronization!
>
I've certainly not intentionally changed it anywhere - if that's an
ntp.conf setting, I haven't changed it.
I should clarify a bit - I just added the 3 public pool servers and
turned off SELinux protection for ntpd this morning, so I don't yet know
if those changes may fix my problem.
More information about the questions
mailing list