[time-nuts] Win XP and NIST time

[Hal Murray]

albertson.chris at gmail.com said:
> I think you can get Windows to run at the "few milliseconds" of error range
> with the standard NTP distribution.

I assume you are talking about getting time from the net rather than a local 
GPS/GPSDO or such.

The accuracy depends upon your network connection and/or the way you use it.  
This is not a Windows problem.  Any OS will have similar problems.

I have an old/slow 384 KB SDSL connection.  The downlink side has almost 4 
seconds of buffering.  That means ntpd can be off by almost 2 seconds.  (Yes, 
seconds.)

I don't think I've seen anything that bad, but it's easy to be off by 100s of 
ms if I download something big like a CD or a long video from YouTube, 
anything that that keeps the pipe full for long enough will confuse NTP.

The other evil player is Firefox and/or the way I use it.  It opens a zillion 
connections to download stuff in parallel.  (I think I found the way to limit 
that.  I'm not sure it always does the right thing.)

-----

Changing hats slightly.  NTP works by exchanging a pair of packets.  The 
client ends up with 4 time stamps: client side send, server side arrival and 
send, and client side arrival.  NTP assumes the network delays are 
symmetrical and calculates the time offset from client to server.  If you 
trust that the clocks are correct, you can measure the network transit times 
in each direction.


-- 
These are my opinions.  I hate spam.