[ntp:questions] Re: Good GPS for attic?

David Schwartz davids at webmaster.com
Sun Feb 13 02:22:15 UTC 2005

"Danny Mayer" <mayer at gis.net> wrote in message 
news:mailman.111.1108257465.583.questions at lists.ntp.isc.org...

>> > how about getting the atomic clock time via internet?
>> >
>> > Matt

>>     Sadly, this doesn't really work. The problem is that to request the
>>time, you have to send out a request and then wait for a reply. Unless the
>>request takes exactly as long to get to the clock as the reply takes to 
>>back, you have no idea what to do with the time when you receive it.
>>     Say you send out a request for the time and get a reply 30 
>> milliseconds
>>later. The time in the reply was correct sometime between when you 
>>it and 30 milliseconds before that. You can 'guess' 15 milliseconds,
>>assuming it took as long for the query to get to the clock as it took the
>>reply to get to you, but this is no more than a guess. Your time can be as
>>much as 15 milliseconds off.

> Dave Mills is much smarter than that. He originally wrote the code some
> 25 years ago when the internet was young and unreliable (relatively).
> One of the fundamental assumptions was that the time to send and the
> time to receive would probably NOT be symmetic. In fact it wasn't
> even assumed that if you sent a packet it would arrive or if it arrived
> you would receive the packet or that it would return on the same path
> that the original packet used.
> The algorithms are that good.

    I'm afraid that it is fundamentally impossible to detect or correct for 
fundamentally asymetric delay. I am well aware of the available solutions 
and their limitations.

    I am not saying "don't use NTP to synchronize to an atomic clock over 
the Internet". I am saying, however, that there are a huge class of problems 
for which sychronizing over the 'net is no substitute for a local reference 
source. And now that GPS is so cheap, so reliable, and so easy to implement, 
there really aren't any excuses not to use it for many situations.


More information about the questions mailing list