What does the accuracy of timing functions depend on? The computer hardware? Simultaneously running programs?

[mavavilj]

What does the accuracy of timing functions depend on? The computer hardware? Simultaneously running programs? Something else?

I'm running https://github.com/ElectricRCAircraftGuy/eRCaGuy_hello_world/blob/master/c/timinglib_timestamps_basic.c on an Elitebook 8470P and I see:

~/Documents/codings/c/gettime> ./hello_world
Hello World.

millis() = 74693423 ms;    micros() = 74693423313 us;    nanos() = 74693423313140 ns
millis() = 74693423 ms;    micros() = 74693423341 us;    nanos() = 74693423341053 ns
millis() = 74693423 ms;    micros() = 74693423346 us;    nanos() = 74693423346025 ns
millis() = 74693423 ms;    micros() = 74693423353 us;    nanos() = 74693423353034 ns
millis() = 74693423 ms;    micros() = 74693423357 us;    nanos() = 74693423357136 ns
millis() = 74693423 ms;    micros() = 74693423361 us;    nanos() = 74693423361226 ns
millis() = 74693423 ms;    micros() = 74693423365 us;    nanos() = 74693423365412 ns
millis() = 74693423 ms;    micros() = 74693423369 us;    nanos() = 74693423369428 ns
millis() = 74693423 ms;    micros() = 74693423378 us;    nanos() = 74693423378081 ns
millis() = 74693423 ms;    micros() = 74693423384 us;    nanos() = 74693423384539 ns
millis() = 74693423 ms;    micros() = 74693423390 us;    nanos() = 74693423390749 ns
millis() = 74693423 ms;    micros() = 74693423396 us;    nanos() = 74693423396853 ns
millis() = 74693423 ms;    micros() = 74693423402 us;    nanos() = 74693423402182 ns
millis() = 74693423 ms;    micros() = 74693423408 us;    nanos() = 74693423408212 ns
millis() = 74693423 ms;    micros() = 74693423413 us;    nanos() = 74693423413594 ns
millis() = 74693423 ms;    micros() = 74693423419 us;    nanos() = 74693423419628 ns
millis() = 74693423 ms;    micros() = 74693423426 us;    nanos() = 74693423426006 ns
millis() = 74693423 ms;    micros() = 74693423432 us;    nanos() = 74693423432297 ns
millis() = 74693423 ms;    micros() = 74693423438 us;    nanos() = 74693423438391 ns
millis() = 74693423 ms;    micros() = 74693423444 us;    nanos() = 74693423444672 ns

which seems to be 1) less accurate and 2) more jittery than the sample output.

[mavavilj]

But I also guess these are still negligible errors, since the header talks about e.g. sleep being at best at 55 \mu s accuracy:

https://github.com/ElectricRCAircraftGuy/eRCaGuy_hello_world/blob/master/c/timinglib.h#L114

[ElectricRCAircraftGuy]

What does the accuracy of timing functions depend on? The computer hardware? Simultaneously running programs?

Yes, both of those--probably mostly your hardware. Also, the Linux scheduler chosen to run this thread makes a huge difference. One more factor that makes a really big difference is the clock you choose to obtain timestamps. I specify CLOCK_MONOTONIC here: https://github.com/ElectricRCAircraftGuy/eRCaGuy_hello_world/blob/master/c/timinglib.c#L33

You can see the various clock options here: https://man7.org/linux/man-pages/man3/clock_gettime.3.html

For a rough estimate of the resolution for your computer with the concurrent events running and with the default SCHED_OTHER Linux scheduler, run get_estimated_resolution() repeatedly and print out the output: https://github.com/ElectricRCAircraftGuy/eRCaGuy_hello_world/blob/master/c/timinglib.h#L91-L95:

/// Obtain an estimate, in nanoseconds, of the resolution of the underlying clock used to obtain all
/// of the timestamps above. This estimate is determined empirically through testing.
/// - Testing on my x86-64 Linux Ubuntu system shows my estimated timestamp resolution to be
///   ~15~30 ns. Run your own tests via "timinglib_get_resolution.c".
uint64_t get_estimated_resolution();

To turn on the Linux SCHED_RR soft real-time round robin scheduler in order to get better resolution (and an improved ~4 us nanosleep resolution on my system, rather than ~55 us nanosleep resolution), run use_realtime_scheduler() once at the start of your program: https://github.com/ElectricRCAircraftGuy/eRCaGuy_hello_world/blob/master/c/timinglib.h#L183-L189

// Call this function to turn ON the Linux soft realtime scheduler for the
// calling thread, in order to get minimum sleep intervals and resolution of
// ~4 us with worst-case results being 100~400 us (0.1~0.4 ms), rather than the
// default minimum sleep time of ~55 us with worst-case results being ~8000 us
// (8 ms)! Calling this function allows for much better sleep resolutions and
// timing accuracies than the default `SCHED_OTHER` scheduler can provide!
void use_realtime_scheduler();

[ElectricRCAircraftGuy]

@mavavilj , how did you find this code by the way?

[mavavilj]

https://stackoverflow.com/a/67731965