53 lines
1.2 KiB
Text
53 lines
1.2 KiB
Text
|
Programs run were
|
||
|
|
- busy: always lock, then check
|
||
|
|
- nested_if: check, the lock, then check again
|
||
|
|
- triple_if: check, check again, then lock, then check
|
||
|
|
- condvar: check then sleep on a single condvar, broadcast to sleepers
|
||
|
|
- two_condvars: check odd_cv or even_cv, sleep, signal opposite CV
|
||
|
|
|
||
|
|
Run on laptop with 4 cores (phaedrus)
|
||
|
|
|
||
|
|
5000 increments, output only when updating
|
||
|
|
|
||
|
|
Simulate work in critical region with usleep(1): 1 microsecond sleep
|
||
|
|
|
||
|
|
time busy:
|
||
|
|
real 0m2.113s
|
||
|
|
user 0m1.032s
|
||
|
|
sys 0m0.596s
|
||
|
|
-----
|
||
|
|
|
||
|
|
time nested_if:
|
||
|
|
real 0m2.982s
|
||
|
|
user 0m3.239s
|
||
|
|
sys 0m0.330s
|
||
|
|
|
||
|
|
-----
|
||
|
|
time triple_if:
|
||
|
|
real 0m2.514s
|
||
|
|
user 0m2.494s
|
||
|
|
sys 0m0.378s
|
||
|
|
|
||
|
|
-----
|
||
|
|
time condvar:
|
||
|
|
real 0m1.635s
|
||
|
|
user 0m0.261s
|
||
|
|
sys 0m0.820s
|
||
|
|
|
||
|
|
-----
|
||
|
|
time two_condvars:
|
||
|
|
real 0m1.581s
|
||
|
|
user 0m0.126s
|
||
|
|
sys 0m0.405s
|
||
|
|
|
||
|
|
The condition variable(s) leads to much more efficient system CPU
|
||
|
|
utilization as evidenced by the smaller user times. Using two
|
||
|
|
condition variables means only waking up a single sleeper that should
|
||
|
|
wake up.
|
||
|
|
|
||
|
|
Between the busy, nested_if, and triple_if, more checks of the
|
||
|
|
variable happen with the nested_if/triple_if leading to higher CPU
|
||
|
|
utilization for checks which is not useful work. The busy variant
|
||
|
|
causes threads to sleep more often lowering user time.
|
||
|
|
|