[ntp:questions] SNTP server + ntpd 4.2.4 client

Unruh unruh-spam at physics.ubc.ca
Wed Mar 19 17:12:21 UTC 2008

Noob <root at localhost> writes:

>Unruh wrote:

>> Noob wrote:
>>> Unruh wrote:
>>>> SNTP is a client protocol, not a server, according to RFC.
>>> You keep saying that. Which RFC are you referring to?

>Apparently, you forgot to answer this question ;-)

As I just wrote, I was wrong. I was misremembering the rfc. Sorry for the

>> The GPS timing claimed is 40ns, but the timestamp is only 10usec. How much
>> does this thing cost?

>Over 1000 euros (!!) AFAIU.

Ah, yes, that was exactly what I guessed ( see previous post). Sheesh. 

>> Are you really in a situation where this is a better
>> solution than say a cheap Garmin 18LVC?

>I've already told my boss several times about the GPS18LVC. I'm not 
>sure why he ignores my request to test a unit.

>Reference to self:

>The HEOL-T101 comes with a very stable XO (OCXO perhaps?) accurate 
>within +/- 30 ppb. How much does such an XO cost?

>Does the GPS18LVC provide an oscillator to serve time even when there 
>are not enough satellites in sight?

For a little while. But even with a bad off the shelf computer, the
internal crystal is probably good for 200ppb, and if it is temp controlled,
better than that. 

>>> Considering the answers I've been given by you and by others in this 
>>> thread, I believe there is a good chance that the setup outlined above 
>>> will work.
>> You have not told us what the requirements are, so what "will work" is
>> is unclear. Yes, I am sure it will discipline your computer's clock.
>> probably to better than ms accuracy.

>My goal was clearly stated in my original message: Will the new setup 
>provide better accuracy than the current setup.

I misremember your current setup. But from operating a 18LVC on one
computer distributed to a bunch around the department at a university, I am
getting what looks like 10us accuracy on most of the machines-- not all. It
depends on the machine load, network load, etc. 
On a machine querying that server over an ordinary telephone ADSL I am
getting about 200usec accuracy ( with a 100usec offset in quiet ADSL times)
Since I seem to recall that you were currently getting about 10ms accuracy,
yes, almost anything will give you better accuracy than that. 

IF your machine runs linux then you will get better accuracy if you use
chrony than if you use ntp ( by a factor of 2-3).(chrony is an ntp
implimentation but with a very different clock discipline algorithm than is
used by ntp). If you computer that you are trying to control is temperature
controlled and has a constant work load, both will probably be about the
same, but if the temp fluctuates due for example to changing work loads,
then chrony does a much better job of tracking those fluctuations. 

What accuracy do you need?

More information about the questions mailing list