[Asterisk-Dev] Hard real-time?

[Steve Kann]

Ricardo Poppi wrote:

> Ricardo Poppi wrote:
>
>
>>> Hi all! There are some linux kernel projecs that provides hardware 
>>> real-time kernel for real time applications. I´ve been looking 
>>> around about RTLinux and RTAI. Those hard real-time kernel 
>>> implementations provides a hardware abstraction layer where the 
>>> regular linux kernel is executed as a low priority task. But - now 
>>> comes the candy - all the real time applications are executed by 
>>> this HAL as high priority so it should guarantee the real time 
>>> responses needed.
>>>
>>> But - now comes the issue - the "real-time application" needs to be 
>>> written using those "new" cappabilities. Take a short look at this 
>>> sentence about RTLinux programing impact: "...From the application 
>>> programmers point of view, the queues look very much like standard 
>>> UNIX character devices, accessed via POSIX read/write/open/ioctl 
>>> system calls. Shared memory is currently accessed via the POSIX mmap 
>>> calls..."
>>>
>>> Do anybody knows about a hard real-time asterisk dev 
>>> project/prototype/beta going on?
>>
>>  
>>
>
> I think that getting proper soft-real-time support to work more sanely 
> would be a much simpler first step (more "bang for the buck", 
> so-to-speak).
>
> I have patches which implement the ideas in a mantis bug about this,  
> and I'll be putting them up soon.  Basically, they'll allow you to, 
> safely, run asterisk as real-time under linux.
>
> -SteveK
>
>
>
> Nice SteveK! Actually I´m just looking around about QLinux and Ingo´s 
> patch for kernel.
>
> The main issue about it - I did not tested yet, probably I will do 
> until weekend time - is that
> QLinux, for example, just runs in 2.4.x kernel and I got affraid that 
> this environment could lack
> some important resource - or bug fix - for asterisk and SER to run and 
> Ingo´s patch - as they say on lists - are still experimental.
>
> Do you have any experience about those soft real-time ways?

There's some nice kernel patches out there;  Ingo's RLIMIT_RT_CPU would 
solve the problem easily -- it's actually similar to the API that's 
provided on Darwin.

I needed something to work sooner, so I have a simple wrapper program, 
called "asterisk-rt", whose usage is like this:

./asterisk-rt
usage: asterisk-rt <-u uid> program

And, you can run it like "asterisk-rt -u asterisk-user /path/to/asterisk 
<options>"

It then will start the asterisk process with real-time priority, and 
also fork two other processes:  A normal-priority "canary", and a 
higher-real-time priority "watchdog".  These processes are connected via 
a pipe.  The canary says "hello" on the pipe every second.  If the 
watchdog doesn't hear from the canary in 5 seconds or so, then it will 
whack the asterisk process with a "non-real-time priority stick", 
assuming that it's gone into a spin-loop, and made your system unusable.

Also, there's a patch to asterisk, which causes it to always drop 
real-time priority when it does things like fork AGIs etc, since it's 
really not necessary to have some perl script loading a million perl 
modules with real-time priority :)

-SteveK