[ntp:questions] Confused about pool rotation

unruh unruh at invalid.ca
Thu Feb 7 04:41:51 UTC 2013


On 2013-02-07, E-Mail Sent to this address will be added to the BlackLists <Null at BlackList.Anitech-Systems.invalid> wrote:
> Robert Scott wrote:
>    the Android SDK.
>
>   It would seem like what you could do in a short amount
>    of time, would instead take several hours.

His concern is that rate, not the offset. A longer time leg is
advantageous, assuming that there is not a shift in frequency during that
long time (temp changes, etc). He wants 12PPM in the rate. 
That is 1ms in 80 sec, which he worries about achieving. (ntd packet
exchange over phone network may not be that good.). Note that ntpd will
take a lot longer than that to get down to 12PPM. 

>
>   ... Unless Android lies, and and the timer isn't monotonic,
>    or doesn't have nanosecond resolution.

Resolution and accuracy are different concepts. 


>
>
>   You do know that NTP server randomize the least significant
>    bits in the timestamps
>   (for the bits below the precision of the platform).

?? The precision is defined as the smallest difference between
consecutive calls to the timer function.

>
>   The 64bit fraction of seconds may resolve to .05 attosecond
>    (0.5e-18), however on a ntp server with precision of -21
>     anything below microseconds is randomized,
>     and precision of -15 below 100 microseconds is randomized;
>     {I guess I might be off by one digit}.
>
>
>   What precisions are you seeing for the servers you are querying?
>    Precision is in the ntp packet between poll and root delay.
>    {I'd expect -20 to -22 for most in the pool,
>     and I occasionally see a -24 or -29 in the pool,
>     but those are fairly rare.}
>     The ATT iPhone NTP servers have a precision of -19.
>
>
>   Also, do you query the same server(s) at both the start,
>    and end?  Otherwise the sum of offset errors between the
>    two might result in more jitter than you expect.

Again, the time even from two different sources should be better than
100ms which is what he is asking for over his 4 hours. 


I do not know whether the GPS on those phones is capable of giving time.

Also sntp uses the same round trip algorithm that ntp does. 



More information about the questions mailing list