[time-nuts] Win XP and NIST time
Dan Kemppainen
dan at irtelemetrics.com
Wed Mar 27 15:30:52 EDT 2013
On 3/27/2013 2:54 PM, time-nuts-request at febo.com wrote:
>> Then all bets are off if you have a CPU that runs at variable speed if you
>> >want the result to be actual time.
> I think that got fixed on newer CPU chips. I don't know when.
>
> Another interesting problem in that area is that the temperature changes with
> the CPU load. The crystal on most PCs actually makes a pretty good
> thermometer.
My modern processors (Phenom II x4's etc) don't have this issue with the
performance counters. I believe that has been sorted out for a while
now. (Thus my qualification of 'should' give you a time stamp from
system start). I wouldn't use the performance counters in any critical
code, just for testing said code loops while writing them.
As for affinity, I do lock my time critical loops in windows to a single
core before starting them. Windows is terrible at time critical data
streaming stuff when handing off threads from one core to another. You
can see the problems pop up as soon as the CPU load shifts from one core
to another. And, don't get me started about hyperthreading! :)
The bottom line is that windows is not good at doing anything in a
timely manner. (Thus the use of lots of buffering!) It was never was
meant to run time critical applications, and probably never will.
So, knowing what we do about windows, why worry about locking it to
anything? Even if the core and RTC were locked to a GPSDO, what will you
gain? Correct time of day within a few seconds is good enough for me!
Dan
More information about the time-nuts
mailing list