- Регистрация
- 1 Мар 2015
- Сообщения
- 6,843
- Баллы
- 155
I've been working with many sysadmins over the years and one question comes up at least twice a year: "I quickly need to create some dummy CPU load on this machine, what cpu stress tool should I install?"
If our need is very basic (i.e. we just want to see 100% CPU load on one or multiple cores), maybe we should consider building our own.
The One-Liner
All we need is to put this line of C code in a file, build it with gcc -o stressme stressme.c (or on Windows cl stressme.c) and run it with ./stressme (or stressme.exe).
int main() {while (1) {}}
And while the program runs, we'll see 100% CPU load on one core. For multiple cores, we could start the program multiple times.
Multi-Threaded
Or we could use threads, here's a variant that uses 4 POSIX threads:
#include <pthread.h>
#include <unistd.h>
#define NUM_THREADS 4
void *loop(void *arg) {
while (1) {}
}
int main() {
pthread_t threads[NUM_THREADS];
for (int i = 0; i < NUM_THREADS; i++)
pthread_create(&threads, 0, loop, 0);
pause();
}
(To build it, add the -pthread flag: gcc -o multistress multistress.c -pthread)
Why does this work?
We're running an infinite loop. When we look at the code of the C one-liner in assembly, it becomes clear the CPU is busy doing only one thing: Executing a jmp instruction that "jumps to itself", as fast as possible.
global _start
_start:
jmp _start
If we were on an older operating system with a cooperative multitasking scheduler, such an infinite loop would probably make our system unresponsive. On today's preemptive multitasking systems, infinite loops cause the program to consume all available processor time, but can still be terminated.
If our need is very basic (i.e. we just want to see 100% CPU load on one or multiple cores), maybe we should consider building our own.
The One-Liner
All we need is to put this line of C code in a file, build it with gcc -o stressme stressme.c (or on Windows cl stressme.c) and run it with ./stressme (or stressme.exe).
int main() {while (1) {}}
And while the program runs, we'll see 100% CPU load on one core. For multiple cores, we could start the program multiple times.
Multi-Threaded
Or we could use threads, here's a variant that uses 4 POSIX threads:
#include <pthread.h>
#include <unistd.h>
#define NUM_THREADS 4
void *loop(void *arg) {
while (1) {}
}
int main() {
pthread_t threads[NUM_THREADS];
for (int i = 0; i < NUM_THREADS; i++)
pthread_create(&threads, 0, loop, 0);
pause();
}
(To build it, add the -pthread flag: gcc -o multistress multistress.c -pthread)
Why does this work?
We're running an infinite loop. When we look at the code of the C one-liner in assembly, it becomes clear the CPU is busy doing only one thing: Executing a jmp instruction that "jumps to itself", as fast as possible.
global _start
_start:
jmp _start
If we were on an older operating system with a cooperative multitasking scheduler, such an infinite loop would probably make our system unresponsive. On today's preemptive multitasking systems, infinite loops cause the program to consume all available processor time, but can still be terminated.