Re: What could be the best for Timing
From: Joseph M. Newcomer (newcomer_at_flounder.com)
Date: 05/12/04
- Next message: Joseph M. Newcomer: "Re: What could be the best for Timing"
- Previous message: Joseph M. Newcomer: "Re: waiting for a spawned process to end"
- In reply to: Robert Scott: "Re: What could be the best for Timing"
- Next in thread: Ben: "Re: What could be the best for Timing"
- Messages sorted by: [ date ] [ thread ]
Date: Wed, 12 May 2004 12:56:56 -0400
Depends on what they mean by "timing", which is part of why I asked about wall-clock time.
Windows is not a real-time system, and it is an illusion to pretend that it is. The
question was a bit vague, because it asked about "timing", which usually means
"performance measurement". If the questioner wanted to be unambiguous, the question might
have said "time-stamping events with an accurate timestamp representing their arrival
time" or "meeting hard realtime deadlines". These are quite different questions. I
interpreted the question as a performance measurement question.
Also, venturecom (www.vci.com) has a true realtime operating system that runs Windows as a
low-priority realtime task. My information is out-of-date, but they used to claim that
they could dispatch, from interrupt to running in your DLL, in 200 microseconds. I suspect
on modern processors it would be substantially faster, only because I believe my data was
gathered in the pre-1GHz processor era. I know nothing of their product other than what
I've picked up at places like WinHEC and in reading their Web site (they have licensed the
Windows source from Microsoft, if I recall correctly). Also, they charge a per-unit
royalty for distribution, but that's not too much out of line with what many other
realtime OS vendors charge.
joe
On Wed, 12 May 2004 02:14:57 GMT, no-one@dont-mail-me.com (Robert Scott) wrote:
>On Tue, 11 May 2004 19:28:56 -0400, Joseph M. Newcomer
><newcomer@flounder.com> wrote:
>
>>QueryPerformanceCounter. Of course, with timing as sloppy as 100ms, pretty much anything
>>would work.
>
>Not necessarily. QueryPerformanceCounter does provide sub-microsecond
>resolution, but attaching that reading to a specific event with an
>uncertainty of less than 100 msec. is not always possible. Your
>application program can be interrupted for more than 100 msec.
>immediately before or immediately after you fetch that precise
>reading, so you don't always get the precision that you think you are
>getting. If good precision 99.9% of the time is good enough, then go
>ahead and use just about anything. But if you need precision 100% of
>the time, then you will have to write a device driver, or use
>something other than Windows.
>
>
>-Robert Scott
> Ypsilanti, Michigan
>(Reply through this forum, not by direct e-mail to me, as automatic reply address is fake.)
Joseph M. Newcomer [MVP]
email: newcomer@flounder.com
Web: http://www.flounder.com
MVP Tips: http://www.flounder.com/mvp_tips.htm
- Next message: Joseph M. Newcomer: "Re: What could be the best for Timing"
- Previous message: Joseph M. Newcomer: "Re: waiting for a spawned process to end"
- In reply to: Robert Scott: "Re: What could be the best for Timing"
- Next in thread: Ben: "Re: What could be the best for Timing"
- Messages sorted by: [ date ] [ thread ]
Relevant Pages
|
