[ntp:questions] Experimental ntpd for Windows

Dave Hart davehart at gmail.com
Sun Feb 1 17:52:12 UTC 2009


On Feb 1, 9:09 am, ma... at ntp.isc.org (Danny Mayer) wrote:
> Dave Hart wrote:
> > I didn't say no one ever uses timeBeginPeriod to raise the multimedia
> > timer precision on a server.  I said it is reasonable to know a
> > particular server will not run any software which raises the
> > multimedia timer precision.  If, on the other hand, you know you do
> > use such software, or you can't be sure what software will be run, -M
> > with the experimental ntpd is wise with Vista and WS2008.
>
> No, that's not correct. You cannot make any assumptions whatsoever as to
> what the system may be running.

I most certainly can make assumptions about what my system may be
running, your orders notwithstanding.

> It is a fallacy to assume that any admin
> or user will have any idea what any applications will be doing with the
> multimedia timer or any other code that affects the clock.

BS.  I know exactly what runs on my servers, and alone decide.  I am
an admin or user.  Your blanket assertion is wrong.  If they don't
know, the safe choice (and the Meinberg installer's default) is to use
-M.

> Furthermore
> as new applications get deployed with as a pure user application or as a
> server application you have no idea what its requirements might be. ntp
> is founded on the principal that the system clock needs to be correct no
> matter what the system or network is or is not doing.

Run with -M.

> I don't have the time right now to analyze your changes but you need to
> ensure that this works on a large variety of systems in a large variety
> of environments.

As I've stated repeatedly, it's better than what we have now.  Now
interpolation is used unconditionally.  With my change, it is avoided
if the scenario that breaks it is detected (fine-grained system
clock).  On Vista and later, with -M, it always detects that breakage
and disables interpolation.  There is no regression, only forward
progress.

Cheers,
Dave Hart




More information about the questions mailing list