Re: [Boost-users] boost time resolution

From: Anthony Williams (anthony_w.geo_at_[hidden])
Date: 2008-05-08 05:45:48

"Patrick Loney" <Patrick.Loney_at_[hidden]> writes:

> I'm using the following code to trace the time different sections of
> code take to execute (where btime = boost::posix_time):
>
> btime::to_iso_extended_string( btime::microsec_clock::universal_time() )
>
> I am finding that a lot of things are happening simultaneously (or
> rather in the same micro second) and then some random operation (perhaps
> the declaration of a variable, perhaps a DB query) will take 0.015 (ish)
> seconds.
>
> The software is running on windows XP service pack 2.
>
> Is there a known resolution issue using the microsec_clock on windows
> xp?

Windows reports times to a precision of 100ns, but typically the system clock
has a resolution of about 10-15ms, so it is common to see 10-15ms jumps in the
time.

Windows Multimedia timers can give you a time period of 1ms if your hardware
supports it. The only other alternative is QueryPerformanceCounter, which has
a system-specific frequency (which you can get from
QueryPerformanceFrequency). Some system device drivers have a bug which means
QueryPerformanceCounter values vary between CPUs, and are therefore unreliable
for really high precision timing unless you set the processor affinity for the
thread.

Anthony
-- 
Anthony Williams            | Just Software Solutions Ltd
Custom Software Development | http://www.justsoftwaresolutions.co.uk
Registered in England, Company Number 5478976.
Registered Office: 15 Carrallack Mews, St Just, Cornwall, TR19 7UL