[ntp:questions] ntp-4.2.6p5 on Win 7 x64

Martin Burnicki martin.burnicki at meinberg.de
Tue Jul 22 08:44:33 UTC 2014


Rob wrote:
> Martin Burnicki <martin.burnicki at meinberg.de> wrote:
>> Except what I've mentioned before I have had rare cases where the
>> Windows timekeeping was generally broken due to some drivers.
>>
>> If I remember correctly then one case was a hard disk driver, and a some
>> latency checker program was used to show that the driver had blocked
>> IRQs for too long, so the timekeeping was strongly degraded.
>
> There have been early IDE disks, 20 years ago, that required the entire
> transfer of a block to occur without interruption.  The drivers for those
> disks disabled interrupts during the transfer.
> This happened on Linux as well.  You could set a flag not to do this,
> on your own risk, and you would start to see corrupted blocks.
>
> Of course, later the disk transfers happened under DMA and probably
> the braindead disk firmware was fixed as well.

Yes, I know those problems dated 20 years ago.

However, both cases I've mentioned occurred in 2012 under Windows Server 
2008 R2 and Windows 7 x64.

When searching for details I found this discussion here where another 
user had similar problems caused by a serial card driver:
http://lists.ntp.org/pipermail/questions/2012-January/031744.html

The cases I meant originally were with Windows Server 2008 R2, where a 
specific version of the Nvidia display driver caused such effects, and 
also with Windows 7 x64, where a user ran some DPC latency checkers 
which found out that the ataport.sys driver shipped with Windows caused 
latency peaks of up to 4 ms, occurring every 3s.

Those cases were not discussed in the NG but via email with Meinberg 
customers.

We may have to keep in mind, though, that at the time back in 2012 when 
the cases were discussed bug 2328 (time adjustments of less than 16 are 
ignored by Windows)
https://bugs.ntp.org/show_bug.cgi?id=2328

had not been reported, and thus no workaround was in ntp-dev at that 
time, so you can't tell if that would have improved the timekeeping in 
these cases.


Martin
-- 
Martin Burnicki

Meinberg Funkuhren
Bad Pyrmont
Germany



More information about the questions mailing list