Miroslav Lichvar <
mlic...@redhat.com> wrote:
> On 2021-08-03, Jim Pennino <
ji...@gonzo.specsol.net> wrote:
>>> In any case, you don't need a local GPS to get a 100ms accuracy. You
>>> just point an NTP client to
pool.ntp.org and you are done.
>>
>> How about when you are at some remote location with no Internet or
>> cell service?
>
> You didn't say this was a requirement. The vast majority of computers
> have Internet access and this is the NTP group, so you shouldn't be
> surprised when people suggest to use that instead.
NTP works whether you have an Internet connection or not if you have a
reference clock, which is the subject of the post.
When people suggest something other than what I am talking about, they
are making the supercilious assumption that I don't know about the other
things, which I find highly insulting.
I made a post that was informational only about a COTS product and
nothing more.
I did not ask any questions nor request suggestions on DIY projects
or anything else.
>> FYI the GNSS receiver that is the actual topic of this post provides
>> CONSISTENTLY better performance by an order of magnitude than any pool.
>
> No, that certainly is not always true. It depends on your Internet
> connection. The typical accuracy of NTP over Internet is few
> milliseconds, i.e. better than a typical PPS-less USB GPS.
You didn't read what I wrote below before writting that, did you?
>
> In my monitoring of the pool servers (from a single location) the median
> absolute offset is about 1.5 milliseconds. That includes all servers
> from the pool around the world, not just servers in the country I live
> in (which would be used by the NTP client).
>
> At home, the offset to the nearest server from the pool I see is below
> 100 microseconds.
Instantaneous numbers are useless for analysis of ntpd performance other
than a quick look to see if it appears to be working and offset alone is
also useless.
Median or average numbers are useless for analysis of ntpd performance
without standard deviations.
If you really want to know what ntpd is doing, you need to do averages
of several things with standard deviations and graphing of those things.
Right now, one of the GNSS receivers I am testing shows an average
offset of -0.85731 microseconds for a bit under 10,000 samples.
That sounds great until you look at the standard deviation which is
2003.41 microseconds. So now your 95% confidence level is about 4000
microseconds.
It becomes even worse when you look at the offset plot that shows random
instantaneous peaks of around 8000 microseconds.
And yes, I did all that for network servers.
>> The GNSS receiver that is the actual topic of this post provides
>> slightly better performance than my ISP ntp servers which are only a
>> couple of hops away.
>
> That suggests your Internet connection has an asymmetric delay. That is
> typical for DSL and cable.
I already know about the effects of asymmetric delay on ntpd.
I have actually read all the documents on
ntp.org that apply to anything
I am acutally doing, i.e. I couldn't care less about WWV receivers in
2021, though I did read it circa 1979 when I built one.
FYI I have one test machine that generates a nice plot that shows when
the air conditions cycles on and off to the microsecond level.