[ntp:questions] Re: [Bug 177] Clock stepping messes up frequency. (fwd)
David J Taylor
david-taylor at blueyonder.co.uk.not-this-bit
Wed Nov 12 10:44:13 UTC 2003
"David L. Mills" <mills at udel.edu> wrote in message
news:3FB1D500.27A39972 at udel.edu...
> Not true, but close. The clustering algorithm is much more sneaky than
> you assume. The metric decisions are intricate and probably boring to
> others on this list, but there is firm principle involved. Take a look
> at the architecture briefing at the NTP project page. I warmly welcome
> discussion on the principles therein. The selection and clustering
> algorithms have a twenty year history, but I would be thrilled to adjust
> them in view of modern experience.
I took a look at the PowerPoint version of the briefing:
- slide 8, Clock filter performance, claims that a reduction in median
error from 192us to 112us is 18.3dB and "impressive". There's an error
To be honest, I'm not sure I am now any the wiser, but I still feel that
there should be something that says: "over a certain delay (say 120ms)
this is a satellite link, and therefore the result should be given less
precedence than a land link".
Perhaps others will agree or disagree?
More information about the questions