[ntp:questions] NTP Delay Measurement

Jason Rabel jason at extremeoverclocking.com
Tue Jul 10 16:53:50 UTC 2007

> Is there someone that have tested the delay of variously servers in the
> world? Is there an official document of the performances of the
timeservers ?

Performance of what exactly? You have the performance of the time server
itself, which is how accurate that machine keeps time. Then you have the
performance of the network, will vary greatly just from the usual flow of
Internet traffic. The further the server physically is you are trying to
contact, the greater and more erratic the delay will be (over the Internet).
Thus it is good practice to contact servers located near you. A traceroute
will reveal the path your packets take across the Internet, showing how many
hops from router to router, how many backbone jumps, and any points of
congestion... The fewer the better. But if the server you are contacting is
not that accurate to begin with then it's all moot.

> For me it would be enough to know the difference between a timeserver in
> Italy, in USA, in Japan (for example) contatting them from Italy.

Again, it is going to vary based on general Internet congestion, which
backbone(s) you are on, etc. What might work great for you could be horrible
for someone using a different provider across the street. I would not
recommend trying to contact servers half way across the world. Something
within the same city (over the Internet) is preferable. If you can not get
the accuracy that you want then a GPS, CDMA, GSM, WWVB, DF77, or whatever
receiver will probably fit your needs.

> In this moment i'm doing my tests (from Italy) with a Stratum 1 NTP server
> (GPS) in Francofort, but this server is located in my Company LAN.
> Is there a real difference between Internet and a protected LAN or the
> result could be similar?

If your stratum 1 servers are located on your LAN, then you can expect
usually around 10 ms or so of accuracy to that server. Over the Internet
(depending on your connection and servers locations) it could still be as
low as 10 ms (unlikely) but more reasonable would be around 50 ms, or even
as high as over 100 ms.

More information about the questions mailing list