David, I don't understand the question. The algorithm to calculate the select dispersion has not changed since rfc1305. The rf1305 description has been rewritten several times in many places, soI will not comment on rfc1305. There are much more modern descriptions in the places I named. The code itself is rather straightforward and worth looking at. See the description, code fragment and explanation in the NTPv spec; in particular, the clusteriing algorithm whit is computed.
Dave Russell, David wrote: > > >Dave, > > > >I am sure that the algorithm is great, really. It is just that I can't figure >out from RFC 1305 how it gets calculated: "which can be calculated in a manner >similar to the filter dispersion described previously". > > > >My first attempt was too similar and I ran the same data points through the >calculation but with a different "w" but that always gives a result larger >than the filter dispersion since ¾ is always larger than ½. :-) That is >obviously not right. > > > >I am suspicious of a particular implementation, especially since one problem >has been found already. I want to calculate the dispersion value >independently. > > > >Regards, > > > >David R. > > > > > >>Russel, >> >> > > > > > >>he detailed calculations are in the refeence implementation document on >> >> > > > >>the NTP project page and in the NTPv4 Internet Draft. However, if the >> >> > > > >>select dispersion is excessive, it isn't because the algorithm is in >> >> > > > >>error; it's because you really do have large differences in the server >> >> > > > >>population. >> >> > > > > > >>Dave >> >> > > > > > >>Russell, David wrote: >> >> > > > > > >>I am trying to validate an NTP implementation that is exhibiting a lot >> >> > > > >>of churn in the cluster algorithm. Unfortunately I don't think that I >> >> > > > >>been able to properly calculate the select dispersion based on RFC1305. >> >> > > > > > > > > > >>Can anyone refer me to a more detailed description of the calculations >> >> > > > >>or even better some example calculations with test data? >> >> > > > > > > > > > > > > >>Regards, >> >> > > > > > > > > > > > > >>David >> >> > > > > > > > > > > > > > > > > > >David R. Russell >CCIE #5751 > > >Dow Jones and Company > >GTS - Performance Engineering Group >Office 609-520-4458 >Cell 610-909-1129 >[email protected] > > > >_______________________________________________ >questions mailing list >[email protected] >https://lists.ntp.org/mailman/listinfo/questions > > _______________________________________________ questions mailing list [email protected] https://lists.ntp.org/mailman/listinfo/questions
