Hi Ok, at least now we are down to some numbers. Ever taken a look at NTP on a Windows box running serial? It's not anywhere near a microsecond….
Bob On May 23, 2012, at 6:47 PM, Chris Albertson wrote: > On Wed, May 23, 2012 at 2:59 PM, Bob Camp <[email protected]> wrote: >> Hi >> >> Ok, to be 1000: 1, you would take the 0.2 to 0.5 ms that you see on the LAN >> and take it up to 200 to 500ms. That's *way* worse than anything I have ever >> seen for a serial server over a LAN. > > Yes, 200 ms would be insanely poor. NTP with a direct connected GPS > can run the micro second level. The 100x to 1000x worse I'm talking > about puts NTP at the 0.5 to 1.0 ms level. > > With PPS connected directly to the DCD line on a Linux based server > the hardware counter is captured with a typical error of 1 micro > second. Over the LAN or USB we see this error at about 1 or 2 > milliseconds So what I'm saying is by using the LAN you give up > micro seconds for milliseconds or about a factor of 1000. You are > right. It is never anything at all like 200 ms. We get better than > 200 ms error over the Internet from servers 1,000 miles away. > > In general you should expect a 1 uSec level error from a direct > connect to from a GPS's PPS to the DCD line of a serial port and a > mSec level error from USB or LAN and tens of ms error from an Internet > connection. > > Where NTP really shines is extracting 10ms level timing from an > unreliable connection that has much longer then 10ms delay. It > really is not so impressive that it can get u-sec level performance > from a directly connected Trimble Thunderbolt. The bottle neck is the > PC hardware. You really never can do better then 1 uS with a generic > PC. > > > >> >> Bob >> >> On May 23, 2012, at 5:15 PM, Chris Albertson wrote: >> >>> On Wed, May 23, 2012 at 2:02 PM, Bob Camp <[email protected]> wrote: >>>> Hi >>>> >>>> What ever degradation the serial stream sees on the LAN, the resulting NTP >>>> output will see once it's on the same LAN. It's unlikely you will see more >>>> than a 2:1 net degradation no matter what is going on. The flywheel in the >>>> NTP algorithm will likely help you in this case to actually improve things >>>> a >>>> bit. >>> >>> Have you actually tried this and measured? 2:1 is very optimistic. >>> Typically it is 1000:1 or worse >>> >>> But you are right that it may not matter. For most uses if the >>> computer's clock is correct at the 0.1 second level they are happy. >>> but this is a "time nut" mailing list and some of us like to get NTP >>> to run at the uSecond level. Useless as that might be. >>> >>> >>> >>> Chris Albertson >>> Redondo Beach, California >>> >>> _______________________________________________ >>> time-nuts mailing list -- [email protected] >>> To unsubscribe, go to >>> https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts >>> and follow the instructions there. >> >> >> _______________________________________________ >> time-nuts mailing list -- [email protected] >> To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts >> and follow the instructions there. > > > > -- > > Chris Albertson > Redondo Beach, California > > _______________________________________________ > time-nuts mailing list -- [email protected] > To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts > and follow the instructions there. _______________________________________________ time-nuts mailing list -- [email protected] To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there.
