visual c libraries: Sleep() and SleepEx() inaccuracy -- visual c libraries

Sleep() and SleepEx() inaccuracy

D. Uptagrafft
11/2/2004 12:25:02 PM

visual c libraries: The accuracy of the Sleep() and SleepEX() functions have changed with newer
processors/PCs. The code listed at the end of this message should execute
100 times for each delay value. I ran this code on several PCs. I get
different results depending on the PC hardware used for the test.

The following PCs give accurate results:

Gateway E3110-300 Pentium II 300 MHz AL440LX chip set
Gateway GP7-550 Pentium III 500 MHz Intel 440BX chip set

Result from the above PCs:
Sleep(10) executes 100 times in 1000 milliseconds; error = 0 milliseconds
Sleep(20) executes 100 times in 2000 milliseconds; error = 0 milliseconds
Sleep(30) executes 100 times in 3000 milliseconds; error = 0 milliseconds
Sleep(40) executes 100 times in 4000 milliseconds; error = 0 milliseconds
Sleep(50) executes 100 times in 5000 milliseconds; error = 0 milliseconds
Sleep(60) executes 100 times in 6000 milliseconds; error = 0 milliseconds
Sleep(70) executes 100 times in 7000 milliseconds; error = 0 milliseconds
Sleep(80) executes 100 times in 8000 milliseconds; error = 0 milliseconds
Sleep(90) executes 100 times in 9000 milliseconds; error = 0 milliseconds
Sleep(100) executes 100 times in 10000 milliseconds; error = 0 milliseconds
Sleep(110) executes 100 times in 11000 milliseconds; error = 0 milliseconds
Sleep(120) executes 100 times in 12000 milliseconds; error = 0 milliseconds
Sleep(130) executes 100 times in 13000 milliseconds; error = 0 milliseconds
Sleep(140) executes 100 times in 14000 milliseconds; error = 0 milliseconds
Sleep(150) executes 100 times in 15000 milliseconds; error = 0 milliseconds
Sleep(160) executes 100 times in 16000 milliseconds; error = 0 milliseconds
Sleep(170) executes 100 times in 17000 milliseconds; error = 0 milliseconds
Sleep(180) executes 100 times in 18000 milliseconds; error = 0 milliseconds
Sleep(190) executes 100 times in 19000 milliseconds; error = 0 milliseconds
Sleep(200) executes 100 times in 20000 milliseconds; error = 0 milliseconds

The following PCs give inaccurate results:

Compaq Deskpro EN P1.0 GHz Intel 815e chipset
HP Compaq d220 Celeron 2.0 GHz Intel 845GV chipset
Dell Dimension 2400 - Celeron Processor Intel 845GV chipset
HP d530 P4 2.4 GHz Intel 865G chipset

Result from the above PCs:
Sleep(10) executes 64 times in 1000 milliseconds; error = 360 milliseconds
Sleep(20) executes 64 times in 2000 milliseconds; error = 720 milliseconds
Sleep(30) executes 96 times in 3000 milliseconds; error = 120 milliseconds
Sleep(40) executes 86 times in 4000 milliseconds; error = 560 milliseconds
Sleep(50) executes 80 times in 5000 milliseconds; error = 1000 milliseconds
Sleep(60) executes 96 times in 6000 milliseconds; error = 240 milliseconds
Sleep(70) executes 90 times in 7000 milliseconds; error = 700 milliseconds
Sleep(80) executes 86 times in 8000 milliseconds; error = 1120 milliseconds
Sleep(90) executes 96 times in 9000 milliseconds; error = 360 milliseconds
Sleep(100) executes 92 times in 10000 milliseconds; error = 800 milliseconds
Sleep(110) executes 88 times in 11000 milliseconds; error = 1320 milliseconds
Sleep(120) executes 96 times in 12000 milliseconds; error = 480 milliseconds
Sleep(130) executes 93 times in 13000 milliseconds; error = 910 milliseconds
Sleep(140) executes 100 times in 14000 milliseconds; error = 0 milliseconds
Sleep(150) executes 96 times in 15000 milliseconds; error = 600 milliseconds
Sleep(160) executes 94 times in 16000 milliseconds; error = 960 milliseconds
Sleep(170) executes 99 times in 17000 milliseconds; error = 170 milliseconds
Sleep(180) executes 96 times in 18000 milliseconds; error = 720 milliseconds
Sleep(190) executes 94 times in 19000 milliseconds; error = 1140 milliseconds
Sleep(200) executes 99 times in 20000 milliseconds; error = 200 milliseconds

The inaccuracy seems to be related to the chipset. PCs with the 440 chipset
give accurate results. PCs with the 8xx chipset give inaccurate results.
The problem exists on Windows 2000 and Windows XP. The problem exists with
Visual C++ 6.0 and Visual C++.net 2003. Uncommenting the calls to the real
time clock (_ftime) verifies that the Sleep() time is inaccurate.

I use Sleep() to delay between readings from a device. I take readings for
several seconds. The inaccuracy of Sleep() means that I get a significantly
different number of readings from a device when I use a different PC.

Has anyone seen this problem and found a solution?

Example code:

#include <Windows.h>
#include <stdio.h>
#include <conio.h>
#include <sys/timeb.h>
#include <time.h>

void main(void)
{
int i,
iDelay,
iDelayx100;
char *timeline;
clock_t tyStartTick;
struct _timeb timebuffer;

for(iDelay=10; iDelay<=200; iDelay+=10)
{
iDelayx100 = iDelay * 100;
tyStartTick = clock();
i = 0;
while((clock() - tyStartTick) < iDelayx100)
{
/* _ftime(&timebuffer);
timeline = ctime(&timebuffer.time);
printf("%.19s.%-3.3hu\n",timeline,timebuffer.millitm);*/

SleepEx(iDelay,FALSE);
i++;
}

printf("\nSleep(%d) executes %d times in %d milliseconds;"
" error = %d
milliseconds",iDelay,i,iDelayx100,iDelayx100-(i*iDelay));
}

printf("\ndone\n");
_getch();

return;