[ntp:hackers] Time constant too large?

Brian Utterback brian.utterback at sun.com
Thu Jan 3 18:25:54 UTC 2008


Double checked. It fires each 10ms.

David L. Mills wrote:
> Brian,
> 
> P l e a s e check the tick interrupt frequency. I suspect it is higher 
> than 100 Hz, in which case as I said before, the frequncy gain is way 
> too high.. The Alpha Tru64 has been running the microkernel since 1992 
> with no ill effects. Note that its timer frequency is 1024 Hz and that 
> the frequency gain has been calibrated for that frequency.
> 
> Dave
> 
> Brian Utterback wrote:
> 
>> I have been testing the V4 code on Solaris Nevada, and have notice
>> something rather disturbing. On one system I have that has a measured
>> frequency correction of 28.5 ppm, it takes the version 3 xntpd about
>> an hour to get to plus or minus 2 or 3, and then settles on 28.5 in about
>> 10 hours. However, the V4 code jumps to something between 40 and 60,
>> and then slowly moved towards 28.5 over about a week and a half. This
>> is not acceptable performance and is a showstopper for upgrading to
>> the latest builds in Solaris.
>>
>> Now, looking at some of the documentation that Dave has published, there
>> is a mention that if you run a micro kernel with the V4 code the time 
>> constant
>> will be too big. Could this be the problem? Solaris is indeed using 
>> the micro
>> kernel ntp_adjtime. If it is, I would think that it would be possible 
>> to correct
>> for this at run time.
>>
>> Oh, and if :disable kernel" is configured, then the V4 code gets the 
>> right value
>> in just a few hours, just like the V3 code. Any ideas on how to fix this?
>>
>> Brian Utterback
> 
> 
> _______________________________________________
> hackers mailing list
> hackers at lists.ntp.org
> https://lists.ntp.org/mailman/listinfo/hackers

-- 
blu

"Mr. Jefferson, build up that wall!"
----------------------------------------------------------------------
Brian Utterback - Solaris RPE, Sun Microsystems, Inc.
Ph:877-259-7345, Em:brian.utterback-at-ess-you-enn-dot-kom


More information about the hackers mailing list