[ntp:questions] Windows-8 improved timekeeping

Charles Elliott elliott.ch at verizon.net
Fri Dec 28 16:47:23 UTC 2012


Win 8 is also better at switching tasks according to priority.  My Win 8
machine runs 8 Seti at Home (CPU intensive) CPU tasks at the lowest priority
(low) and 2 GPU tasks at the second lowest priority (below normal).  There
is no interference with other tasks or with the screen.  On versions before
Win 8, I had to move the priority of Dragon Naturally Speaking to above
normal on Win 7 and to high on XP to achieve decent recognition accuracy.
It appeared as though Dragon was CPU-starved for long periods since there
was meaningless gaps in the text on the screen; the words were spaced
correctly but whole verb or noun phrases were missing.  On Win 7, I finally
had to limit Seti at Home to 7 CPUs to see good recognition accuracy, whereas
on XP I did not have to do that.  Also on Win XP and 7 the screen would
become unresponsive for extended intervals.  There is none of that on Win 8.
My other machines run Server 2003, Win XP, and Win 7.  There is still slow
responsiveness on these machines when Seti at Home is running, although
Nvidia's most recent video driver update (310.70) has improved that
situation greatly.

There is a cost for that increased responsiveness though: According to
Performance Manager, Win 8 services about 69,000 interrupts a second, and
about 80,000 per second if NTPD_USE_INTERRP_DANGEROUS is set. 

I hypothesize that that high number interrupts serviced per second may imply
that Microsoft has reduced the scheduling interval from, I believe 10-15 ms,
to a shorter period.  (Goggling "time slice" most entries say 10-15 ms, but
that these figures vary between HALs, server and desktop O/S versions,
processor speeds, foreground and background processes, and single and
multiple processor environments; that Microsoft is very careful not to
disclose this number; and that a tool to measure the quantum is no longer
available from SysInternals.  One author wrote that he had a real time
application at his work site that used Windows successfully that
occasionally experienced 100 ms delay in being granted the CPU.  He wrote
code to detect this situation and was able to compensate for it.)  This
implies that NTPD is being given the CPU more quickly under Win 8 when it
needs it, instead of waiting.  Maybe, possibly.

> -----Original Message-----
> From: questions-bounces+elliott.ch=verizon.net at lists.ntp.org
> [mailto:questions-bounces+elliott.ch=verizon.net at lists.ntp.org] On
> Behalf Of David Taylor
> Sent: Friday, December 28, 2012 4:32 AM
> To: questions at lists.ntp.org
> Subject: [ntp:questions] Windows-8 improved timekeeping
> 
> Just for interest, I've provided a performance snapshot of a recent NTP
> running on Windows-8, with both Internet-only and LAN-based servers.
> 
> When running off just Internet (WAN) servers, Windows-8 shows an
> averaged jitter of less than a millisecond, perhaps 0.4 - 0.6
> millisecond after the initial transient has settled down.  Windows-7,
> by
> comparison, shows 3 - 9.5 milliseconds averaged jitter.  Likely the
> improvement is due to the use of the new precise-time instruction in
> Windows-8, which is used by recent development releases of NTP.
> 
> When running off the LAN, averaged jitter is around 40-80 microseconds
> in another Windows-8 system.
> 
>    http://www.satsignal.eu/ntp/Win-8+Internet.html
> 
> There still remains an issue with Windows-8 and the resolution of the
> timer adjustment, which allows fine adjustments and reports them back
> (IIRC), but internally only makes coarser adjustments.  Yes, these
> results are not as good as you might expect with FreeBSD or Linux, but
> it does seem that Windows-8 timekeeping is an improvement over Windows-
> 7.
> 
> I hope that's of some interest.
> --
> Cheers,
> David
> Web: http://www.satsignal.eu
> 
> _______________________________________________
> questions mailing list
> questions at lists.ntp.org
> http://lists.ntp.org/listinfo/questions



More information about the questions mailing list