[time-nuts] NTP and Windows 7

[Martin Burnicki]

Hal,

Hal Murray wrote:
>
> martin.burnicki at burnicki.net said:
>> Please note under Windows you should configure all upstream servers with  a
>> line reading
>
>> server aa.bb.cc.dd iburst minpoll 6 maxpoll 6
>
> There are lots of times when reducing maxpoll is reasonable, but I think an
> unqualified suggestion is not appropriate.

I don't think my suggestion was unqualified. ;-)

We have already discussed this in the NTP newsgroup/questions mailing list.

Here is a loopstats graph from a Windows XP system running ntpd 4.2.6p5 
without limitation of maxpoll:

http://people.ntp.org/burnicki/windows/ntpd-4.2.6p5-WinXP-no_poll_limit.pdf

And if you limit maxpoll to 6 in the same installation the results look 
like this:
http://people.ntp.org/burnicki/windows/ntpd-4.2.6p5-WinXP-poll_6.pdf


A loopstats graph from another test with ntp-dev running on Windows 7 
without limitation of the polling interval are here:
http://people.ntp.org/burnicki/windows/ntpd-4.2.7-Win7-poll4-max.pdf

And similarly, with limitation of the polling interval to 6:
http://people.ntp.org/burnicki/windows/ntpd-4.2.7-Win7-poll4-6.pdf

In the examples above I had minpoll set to 4, just to see how the fix 
for NTP bug 2328 behaves, and you can clearly see that time 
synchronization is degraded if the polling interval is *below* 6 since 
the workaround doesn't work properly, as mentioned in the bug report.

So I wouldn't suggest anyway to set minpoll below 6.

> maxpoll of 6 polls every 64 seconds.  The default maxpoll is 10, or 1024
> seconds, so that's a 16x[1] increased load on the servers.  Some/many people
> would consider that to be abusive use of a resource.
>
> If you are using your servers, you can do whatever you want.  If you are
> using your ISP's servers or a friends, then whatever they agree to is fine.
> The NIST servers are already heavily/over loaded.  I'm not sure if the pool
> could stand a 16x increase in load.

I agree that limiting maxpoll isn't the best choice if you can avoid it.

However, by default ntpd starts at minpoll 6 anyway, it determines 
automatically if the polling interval can be increased towards 10, or be 
decreased again towards 6.

You can see how the interval is decreased automatically in the loopstats 
for 4.2.6/Win XP, if the time discipline becomes too bad.

So public services should anyway account for clients sticking a the 
default lowest poll interval, namely 6.

And of course it is generally not a good idea to let each laptop get its 
time directly from the primary servers at NIST or PTB, but this is a 
different problem.

Martin