LinuxQuestions.org - Creating a milisecond timeout timer for threads

- Programming (https://www.linuxquestions.org/questions/programming-9/)

- - Creating a milisecond timeout timer for threads (https://www.linuxquestions.org/questions/programming-9/creating-a-milisecond-timeout-timer-for-threads-213423/)

Creating a milisecond timeout timer for threads

Hi!
I have a few threads (pthread) that are created at the same time and run simultaneously. Each of them should last for a given amount time and then finish. For example, if thread#1 should last for 5 seconds and thread#2 for 3 seconds, thread#2 should finish 2s before thread#1, but the total amount of time (for both) has to be around 5s.
If I use the time() function (time.h), it works but with a resolution of 1s. In order to have a resolution of 1 millisecond, I was using the clock() function but it works completely wrong depending of the number of threads.
I would appreciate ANY comments about a accurate millisecond timer to create a timeout (or virtual interruption) for threads.
Thanks.

You can use select() to make it work like a timer with that resolution.

nanosleep() is good but if you give too much time to sleep lets say 2 sec, it may not work because of overflow, so you can use usleep() that takes input in microseconds or if you want to use nanosleep, use it in a loop like if you want to sleep for i milisecond, give value for 1/100 milisecond and then put that statement in a loop.

// sleep for 1 milisecond

for (i=0;i<100;i++)
//nanosleep for 1/100 miliseconds

timeval gtod_now_time;//Do a gettimeofday on this when the prog starts

You may also use such a function to query (roughly) millisecond accurate time
progress in a thread:

unsigned long _my_mmtime(void){
unsigned long result=0;
// long vtime=0;
#ifdef _WIN32
result=timeGetTime();
return result;
#endif
#ifdef LINUX_APP
timeval gtod_now_time;
gettimeofday(&gtod_now_time,0);
result=(gtod_now_time.tv_usec-gtod_start_time.tv_usec)/1000;
result+=(gtod_now_time.tv_sec-gtod_start_time.tv_sec)*1000;
return result;
#endif
return 666;//Not a recognized os
}

Do a nanosleep/usleep in your thread and query the exact time which has evolved after each successing sleep.

BTW If you're thinking of a timing inside a thread which is *guaranteed* to be as
accurate as 1 millisecond, I believe it will not work, as Linux (nor Windows) is
a realtime os, although at least Linux behaves nicely under that respect. But you don't have too much influence on thread scheduling.

Quote:

Originally posted by cppkid
nanosleep() is good but if you give too much time to sleep lets say 2 sec, it may not work because of overflow, so you can use usleep() that takes input in microseconds or if you want to use nanosleep, use it in a loop like if you want to sleep for i milisecond, give value for 1/100 milisecond and then put that statement in a loop.

// sleep for 1 milisecond

for (i=0;i<100;i++)
//nanosleep for 1/100 miliseconds

No, you don't need to do that.
Nanosleep() takes a struct as a parameter which has a seperate filed for seconds, just like pselect().

From "man nanosleep":

Code:

int nanosleep(const struct timespec *req, struct timespec *rem);



struct timespec

{

    time_t  tv_sec;        /* seconds */

    long    tv_nsec;      /* nanoseconds */

};

So, just fill in the seconds and nanoseconds in the right member of the struct, and you won't have that problem.

Also, you cannot use usleep() together with sleep() or alarm(). And nanosleep can also be used to continue the "sleep" when interrupted. All-in-all, nanosleep() is the best to use. (or select(), or pselect() for that matter, though they're more complex to use).