[time-nuts] Timing performance of servers

[Magnus Danielson]

Dear Edgardo,

On 10/27/2012 07:41 PM, Edgardo Molina wrote:
> Dear Magnus,
>
> I do not have a reference for the performance of Windows as NTP server.
> This has been a busy week and long working nights. It is a logical
> workload after being absent for nearly two weeks : )
>
> I will browse the web and NIST to find some solid references to this
> issue. I am still enthusiastic about doing the side by side comparison
> for measuring various parameters between a Windows and a Linux box
> running NTP. It will require time to set the test but the contribution
> could be interesting, specially if no work has been done previously.
> I will use my spare time this weekend to search for information on the
> subject.

A wealth of information has already been show in this thread. I'm sure 
there is more out there.

I wonder to what degrees the different methods to illustrate errors have 
been used. Frequency stability for traditional white, flicker and random 
noises we illustrate with (modified) Allan Deviation, but it is maybe 
not the best method for illustrate temperature shift variants as well as 
the noise of packet networks. Similar for phase stability, where TDEV is 
being used. Typical way to illustrate time effect of systematic noises 
in telecom networks is the MTIE measure, which aids in showing the 
buffersizes and clock recovery PLL bandwidth needs, which also the 
traditional sinusoidal tolerance curves does. There are also new methods 
like MAFE for the packet world.

What will happen on lost of reference and the hold over properties can 
also be of interest. Then systematics will surely dominate.

Cheers,
Magnus