[Asterisk-Dev] High resolution timers using POSIX clocks instead of zaptel

[Steve Kann]

Tilghman Lesher wrote:

>On Friday 27 May 2005 05:33, David Woodhouse wrote:
>  
>
>>Using zaptel just for a time source is silly. We should use POSIX
>>timers instead. Here's a start, but I don't know how long it'll be
>>before I can get to finish it, so I thought I'd post it here in the
>>hope that someone else will pick it up.
>>    
>>
>
>As noted here:
>
>http://lists.digium.com/pipermail/asterisk-users/2002-April/002072.html
>
>software timers are lacking.
>  
>

I think that is just narrow thinking.

app_conference works all day and week long, without any hardware 
support.  It also does it without posix timers.  It just uses nanosleep 
and gettimeofday.

You can make things go X times per second, with great accuracy over a 
long period of time with just those tools.  What you can't always do is 
make sure that you don't have a period where you oversleep for a few ten 
of milliseconds.  But if you keep track of time, you can just do more 
work when you do wake up.


i.e.  in pseudo-code

do_something_n_times_per_second(func_t something, int n) {
    long now, last, begin = getcurrentmillis();

    last = begin;  

    for(;;) {
        usleep(n);
        time now = getcurrentmillis();  
        while(last < now) {
             something();
             last+=n;
        }
    }  

}


where getcurrentmillis() can be based on gettimeofday, of course.

This will call something() 1000/n times per second. It might not be 
spaced out so that it happens _exactly_ every n ms, but it will, over a 
particular time period, happen 1000/n times per second.

If this runs in a real-time process, and the something() takes less than 
n ms to run, it will happen very close to every  n ms.

Of course, most things don't need to run every ms, and this will work 
better when n>1.

-SteveK