![]() home |
Data (fixed work) |
Last Modified: 24 May 2001
|
Below is the test data taken relative to CPU overhead using a
fixed work method.
Test Paramters:
two nested for loops - partial outer cycle and 717,221,000
nine total timing runs with first thrown away (resulting
run times approximate CPU overhead II testing for 150,300 ns between interrupts)
Results:
average cost to user per interrupt: 11,120
nanoseconds
sample standard deviation:
333 nanoseconds
|
INTERRUPTS* ns |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
* "TIME BETWEEN" is based on 501
nanoseconds per Myrinet card RTC unit. So it is (RTC units) * 501.
Number in "[ ]" is the RTC units.
&& Nine runs of seven seconds each. First run is
not counted in average.
+ Based on (RTC units * 501)*(1-(percent
user time)/100)