"Ron Frazier (NTP)" <[email protected]> wrote in message news:[email protected]...
I'm forking the subject line, which didn't really seem relevant any more.

more below
[]
 http://www.satsignal.eu/ntp/Win-8+Internet.html
[]
From my point of view, those look pretty good. Those peerstats graphs come in pretty handy.

It is an unloaded system, but may show what NTP can do at best using just Internet servers in a domestic Internet-linked environment.

[]
What's the dangerous part? Are we just talking lots of CPU churning? If so, can I just shut down NTP and tweak it?

Sincerely,

Ron

Ron,

It's just about performance. Windows XP performed better when NTP tried to interpolate the ~15 millisecond timer ticks. Windows Vista performed significantly worse when that interpolation scheme was used, so it was conditionally disabled. In early tests of Windows-7, performance seemed better, but the OS code which changed the timer interval didn't start at the same time as NTP, so sometimes NTP made the wrong choice.

Here's what Dave Hart said at that time:
"I thought I understood the problem with ntpd interpolation on Vista/Win7 to be caused by the OS scheduling granularity being insufficiently finer than the native clock granularity, so that the interpolation thread's sampling of counter/native clock correlations was nearly always occurring at the same relative point between two native clock ticks, rather than being nicely spread around. Being spread around is important because the algorithm chooses the sample nearest the prior native clock tick when converting a counter value to an interpolated time."

So there's no problem with high CPU or anything like that - simply worse offset and jitter. You can also force interpolation off by setting:

 NTPD_USE_SYSTEM_CLOCK=1

or on with:

 NTPD_USE_INTERP_DANGEROUS=1

Delete the environment variables not in use. The actual value they are set to doesn't matter, it's only the presence or absence which is tested. After setting these SYSTEM environment variables, simply restart the NTP service to see the change. I see these settings as ones for experimenters, and they should not be needed in normal use.

Cheers,
David
_______________________________________________
questions mailing list
[email protected]
http://lists.ntp.org/listinfo/questions

Reply via email to