[ntp:questions] Tighter regulation?

David Woolley david at ex.djwhome.demon.invalid
Mon May 20 13:12:30 UTC 2013


Mischanko, Edward T wrote:>
> The particular client I am monitoring is on an extremely stable corporate
> LAN.  The servers are on the same LAN.  Changes in network propagation
> are minimal.

On the other hand, it is on Windows, which doesn't have very stable time 
measurement.

> 
> The 8 stage minimum delay filter is exactly what I am seeing and exactly
> what I would like to disable.  When polling above 256 seconds the amount
> of time it takes to make an adjustment is excessive.  Setting maxpoll to
> 8 does help, but that defeats what server administrators like to have to
> reduce loading, on a grand scheme.

Defeating this will not have a large effect if you really do have very 
consistent measurement errors, as the polling over samples compared with 
the needs of the loop filter.  If measurement errors are variable, it 
will degrade the performance.
> 
> I am not sure what jitter will do, but what good is low jitter at the
> expense of poor offset correction?
> 
Increased jitter means that your offset measurements span a wider range 
of values.

I think you need to provide:

1) a tabulation of the raw offset values, including signs;
2) an analysis of the breakdown of errors between the crystal frequency, 
Windows load related errors, and network load related error.

It is important to understand that offset is time error + measurement 
error, and in the steady state, measurement error will be the dominant 
component.



More information about the questions mailing list