[ntp:questions] Re: time

CBee I.Dont at want.your.spam
Thu Dec 1 17:07:16 UTC 2005

K wrote:

> Hi,
> What are some good ways to get time?
> I know about NTP v3 and v4 and use it on some systems.  I need a time
> accuracy of < 1 ms.

For what purpose? Or do the components just need to be in sync within 1 ms?

> NTP over gigabit Ethernet is down to what time of precision?   1ms?   Does
> anyone have any data on this?

gigabit ethernet is just a clock frequency of 1 gigabit and yes, the bits in a 
package travel each on that clock. However, switches, routers and also nics do 
insert gabs between the packages just to keep up. It is depending on the quality 
of the devices how small these gaps are and when and how they come.

For timekeeping with ntp, the speed of the network is not directly related to 
the ultimate time quality you can get. Things like latency and a stable latency 
are things that do produce a better time quality: It is better to get a 
guarantee that the pakage takes 100 ms to travel than that it travels at 10 to 
20 ms. If you do know the package takes 100 ms, ntp can (and will) calculate 
with that. If it travels at a speed between 10 and 20 ms, it cannot rely on it 
and the time quality will be less.

> I know that the linux os is compiled for something like 1 ms precision

Yep, it might even do much better. However, it uses the hardware for that. Since 
linux is available on many platforms, it would be nice to see which platform can 
produce the best precission. If I read a 3 GHz machine actually runs at 2.94 
GHz, it has an offset of 2 %. Who tells me it is accurate in its time?

Once again, yep, linux has an internal clock that counts the miliseconts and it 
travels all of them. However, who tells me each milisecond actually takes 1000 

As far as I know (well, expect actually, update me if I'm wrong) is that this 
milisecond clock is used to skew the current time into place so the 
date-and-time clock is as accurate as possible.

> IRIG is a pain?  Is there anything better?

What application does need a clock more precise than 1 ms? I'm just curious.


More information about the questions mailing list