nanosleep granularity

[Massimo Brandini]

Hi all,

some collegues of mine are having troubles in getting a granularity finer than 10 ms when using the system call nanosleep, they need to be able to make a process (eventually a thread) wait for a very precise amount of time.
Is there some system parameter to change for allowing intervals smaller than 10 ms? (without affecting the other system components ;-)

Many thanks in advance

[rick jones]

i'm reasonably certain that 10ms is the limit at the point in time.

[A. Clay Stephenson]

Actually this is hopeless. Even if you were able to change the miminum timing resolution there is still no guarantee that nanosleep will
suspend the thread for a precise amount of time. A process (thread) with higher priority,
signals, and other scheduling activities could clobber any attempt at precise delays. You would have better luck running at real-time priority but that is only done under very special circumstances. Remember, this is UNIX; it's really not a good fit for applications relying upon very precise timing and scheduling. Your programmers really need to rethink the problem and come up with a better synchronization scheme.

Regards, Clay

[rick jones]

that precise timing may not have been for synchronization. it could have been for pacing of data or some such. indeed, if it was for synchronization, the programmers need to be re-educated.

i suppose that some sort of loop with gettimeofday() and sched_yield() could allow a thread to perform other actions with greater timing precision, but the overhead would be (likely) considerable

[Klaus Crusius]

The restriction of 10ms comes from the internal clock resolution seen by the kernel. See the constant CLK_TCK = 100 (ticks per second) in /usr/include/sys/time.h.

Klaus