[ntp:questions] Proposed NTP solution for a network

Uwe Klein uwe_klein_habertwedt at t-online.de
Fri Mar 6 16:46:45 UTC 2009


Unruh wrote:
> Uwe Klein <uwe_klein_habertwedt at t-online.de> writes:
> 
> 
>>Unruh wrote:
> 
> 
>>>These Gbit
>>>switches seem to have, or trigger in the ethernet cards, additional and
>>>assymetric delays of the order of 100-200usec. (assymentric delays are of
>>>course the worst, since they directly affect the errors in ntp). 
> 
> 
>>My guess is you see Store-n-Forward <> Passthrough indeterminisms.
> 
> 
>>i.e. some packets go via Passthrough others goe via Store-n-Forward through
>>the switch dependent on direction, activity on other ports of the switch
>>and the weather.
> 
> 
> Does not really matter why, it raises havoc with ntp and the ability to get
> really good timing via the net. But thanks for the hint. On one system ( a
> 100Mb system) I get occasional 1 sec delays (occasional meaning >1%) in the
> ntp packets.

my guess is a whole second is not your switch push/poping a packet ;-)

Do you have the raw packet timing data? If you plot them a points clouds
you should see distinct "slots"  each bracket by the window of expected jitter.

uwe




More information about the questions mailing list