Books are failing me!

Newbie & Debugging Questions
#21

Blame the OS / hardware. No mainstream OS can provide nano second precision timing, so adding support for it in Java was basically a useless exercise. The best you can get is microsecond precision from it. But perhaps in the future such things will become possible, who knows.

#22

With a bit of C++ (and a JNI wrapper), we can do accurate milli/micro sleeps:


#include <time.h>
#include <errno.h>

void millisleep(const int ms)
{
    struct timespec sleeptime;
    struct timespec remaining;
    int result;               
    remaining.tv_sec  = (ms/1000);
    remaining.tv_nsec = (ms%1000)*1000000;
    do {                                 
        sleeptime.tv_sec  = remaining.tv_sec;
        sleeptime.tv_nsec = remaining.tv_nsec;
        result = nanosleep(&sleeptime,&remaining);
    } while(0>result && EINTR==errno);
}

void microsleep(const int us)
{
    struct timespec sleeptime;
    struct timespec remaining;
    int result;               
    remaining.tv_sec  = (us/1000000);
    remaining.tv_nsec = (us%1000000)*1000;
    do {                                 
        sleeptime.tv_sec  = remaining.tv_sec;
        sleeptime.tv_nsec = remaining.tv_nsec;
        result = nanosleep(&sleeptime,&remaining);
    } while(0>result && EINTR==errno);
}
#23

I am obviously ignorant and naive, but where would you need nanosecond accuracy anyway? I already find it hard to believe that a few microseconds here or there matter. It’s an honest question, in what scenarios do you need accurate microseconds or better?

Andreas

#24

For almost everything there is no use. There are, however, some real-time versions of Java where this may be useful. A 2 GHz processor is 0.5 nanoseconds / cycle so the unit is not meaning-less.

#25

If there are musical parts that are off by a couple milliseconds, they definitely start to feel “out of the pocket”. At a single millisecond or less, I’d think I’d have trouble identifying parts that lead/lag. Maybe a world-class professional percussionist could sense offsets at the micro-second level, but I think even so, it would be a matter of hundreds of microseconds, not a matter of just a few.

But even with micro-second timing, it’s moot if the GC interruptions take milliseconds to complete, yes? I mean, there are issues relevent to timing besides the degree of granularity of the “getCurrentTime” function.

Concerns at the nano-level can be left to the folks at CERN.