Why the 2nd diff is so different from the others, using gettimeofday in a loop?

Question

This function is quite new for me, so I just writed a little program to get familiar with it. Here's a my program (just printing laps between several calls of gettimeofday).

#include <stdio.h>
#include <unistd.h>
#include <stdlib.h>
#include <sys/time.h>
#include <pthread.h>

typedef struct timeval t_timeval;

t_timeval    get_diff(t_timeval prev)
{
    t_timeval current_time;
    gettimeofday(&current_time, NULL);
    printf("diff : %ld seconds and %ld micro seconds\n",
        current_time.tv_sec - prev.tv_sec, current_time.tv_usec - prev.tv_usec);
    return (current_time);
}

int        get_time_laps()
{
    int            i = 0;
    t_timeval    prev;

    gettimeofday(&prev, NULL);
    while (i++ < 50)
        prev = get_diff(prev);
    return (0);
}

int        main()
{
    get_time_laps();
    return (0);
}

the result is :

diff : 0 seconds and 0 microseconds
diff : 0 seconds and 47 microseconds
diff : 0 seconds and 1 microseconds
diff : 0 seconds and 1 microseconds
diff : 0 seconds and 1 microseconds
diff : 0 seconds and 1 microseconds
diff : 0 seconds and 1 microseconds
diff : 0 seconds and 1 microseconds
diff : 0 seconds and 2 microseconds
diff : 0 seconds and 1 microseconds
diff : 0 seconds and 1 microseconds
diff : 0 seconds and 0 microseconds
[...]
diff : 0 seconds and 1 microseconds

So I m wondering why the 2nd diff is so different from the others.. FYI I made the test several times, every times I get this pattern.

It's impossible to say precisely without profiling the running code. One possibility is that the first iteration is paying for caching costs. After that the cache is hot for subsequent iterations. — kaylum, Jun 08 '21 at 06:49
Note that POSIX has marked `gettimeofday()` obsolete in favor of `clock_gettime()`. — Shawn, Jun 08 '21 at 07:30

George Glavan · Accepted Answer · 2021-06-08T10:34:41.850

The reason this happens is quite interesting. Since gettimeofday would use a syscall that is called very often, linux uses a thing call vdso, which is a so that is mapped automatically in your program in order to access the things related to the time information without the need of a syscall (since syscalls are costly).

The way this works is that when you call gettimeofday, your program will look for the symbol in vdso. and when it finds it, i will cache it in its memory for faster accesses. thats why the first call is slower. You can find a way better explanation here

Oftentimes the C library will do detection with the first call and then cache the result for subsequent calls.

And a nice way to benchmark the vdso vs the actual syscall here

Why the 2nd diff is so different from the others, using gettimeofday in a loop?

1 Answers1