[ntp:questions] Keeping NTP Honest

Unruh unruh-spam at physics.ubc.ca
Mon Jul 13 23:05:18 UTC 2009

"Richard B. Gilbert" <rgilbert88 at comcast.net> writes:

>Evandro Menezes wrote:
>> On Jul 12, 3:21 pm, Unruh <unruh-s... at physics.ubc.ca> wrote:
>>> The way ntp works, faster polling also means worse rate estimation and
>>> more annoyance of the providers of the time. The current setup is done
>>> that way to try to minimize the rate error, so if your sconnection to
>>> ntp goes down, your system can freewheel with the greatest accuracy.
>> But that's the issue: NTP allows for good freewheeling if it comes to
>> that, provided that the system maintained in STP and in a vacuum.
>> In the real world, ambient temperature changes frequently even in
>> conditioned environments, network load affects packet jitter, etc.
>> And all this also affects a system's peers, compounding the issue of
>> NTP's slow reaction time.
>> Thanks.

>You are certainly at liberty to write your own version of NTP and have 
>it behave as you think best.

>For most of us, NTP works quite well.  I suspect that an NTP equivalent 
>designed to react instantly to events such as: the opening of the 
>computer room door, the laser printer starting or finishing a print job, 
>  or the starting and stopping of the refrigerant compressor, would do 
>the job far worse than we what we have now.

Sure. And if you are happy with it, there is no reason to change to
anything else. It is a well supported system, and has millions using it. 
On the other hand if you really want a system which keeps time really
well, get a GPS PPS system, and use chrony or get an ntp version which
uses the temperature to compensate for rate changes. 

(I have seen it but have no references to give you. Ie, a group altered
ntp to use the computer temperature to predict the rate changes, and
found that they got a significant improvement in the clock discipline
doing that. The biggest problem is finding which temp on your system is
the best proxy for the temp of the clock crystal on the motherboard. 

Or you could glue a thermistor onto the crystal and interface it to the
computer somehow.

More information about the questions mailing list