[Bayonne-devel] bayonne2 dialogic drivers, hmp, etc

[David Sugar]

I am actually setting up to do work on bayonne2 dialogic drivers. I have
not looked at HMP yet, however. Ideally, some of what happens in
ccaudio2, particularly audio processing and codec routines, should also
be better optimized for target processors that have specialized
instructions, but this has not happened yet.

Julien Chavanton wrote:

> I have made a few modification to globalcall drivers to handle
> TASKFAIL more efficiently
>
> I always reset the port when there is a TASKFAIL because anyway when
> there is a TASKFAIL the call is lost and most of the time the TASKFAIL
> is caused by short call
>
> Or abandon calls.
>
> This is producing positive results.
>
> The only difficulty was to keep bayonne state sync with Globalcall
> state because RESETLINEDEV is not the normal state transition.
>
> -------------------------------------------------------------
>
> I keep an eye on Bayonne 2 evolution; it would be interesting for me
> to help with the Globalcall Drivers in Bayonne 2.
>
> Something positive about Globalcall is that Intel is moving to HMP
> host media processing “Voip library optimized for Xeon CPU”.
>
> And this may be the best performance Voip solution available due to
> the integration in the Xeon CPU internal instructions.
>
> If they succeed Globalcall should be almost compatible with HMP, and
> bayonne could be HMP enabled without to much work.
>
> However HMP is commercial I think it is 10$ a port and it may be too
> much for a software solution.
>
> Julien
>
> ------------------------------------------------------------------------
>
> *From:*
> address@hidden
> [mailto:address@hidden
> *On Behalf Of *Julien Chavanton
> *Sent:* December 19, 2005 10:09 AM
> *To:* address@hidden
> *Cc:* address@hidden
> *Subject:* [Bayonne-devel] Bayonne Globalcall TASKFAIL and
> TRUNK_CALL_FAILURE
>
> Hi David,
>
> I am looking to modify TRUNK_CALL_FAILURE in bayonne globalcall since
> this is currently handled as a disconnection.
>
> It was not that bad but most of the time this is not enough.
>
> And we loose port, only when there is a TASKFAIL that is really caused
> by the Dialogic drivers.
>
> The main source of TASKFAIL is abandon calls and short calls the
> Dialogic drivers are not capable of handling this automatically and
> this result in TASKFAIL
>
> The only problem I think is that the script may be left attached and
> the next call will fail to attach.
>
> Here is what I have in mind:
>
> globalcall/driver.cpp
>
> case GCEV_TASKFAIL
>
> event.id = TRUNK_CALL_FAILURE;
>
> trunk->postEvent(&event);
>
> gc_ResetLineDev(linedev, EV_ASYNC);
>
> break;
>
> globalcall/trunk.cpp
>
> case TRUNK_CALL_FAILURE:
>
> // We need to prepare Bayonne for recovery of next call
>
> Trunk::detach();
>
> if(join)
>
> Part();
>
> if(tgi.pid)
>
> ::kill(tgi.pid, SIGHUP);
>
> if(thread)
>
> thread->stop();
>
> Trunk::flags.dsp = DSP_MODE_INACTIVE;
>
> More information about TASKFAIL if any interest:
>
> How to Handle GCEV_TASKFAIL Events (However I have made some test and
> I am not sure about this article complete accuracy)
>
> http://resource.intel.com/telecom/support/tnotes/tnbyos/2000/tn061.htm
>
> You can find information about abandon calls here:
>
> http://resource.intel.com/telecom/support/releases/winnt/SR511FP1/onldoc/htmlfiles/gcprgdw/gc-sta15.htm
>

dyfet.vcf
Description: Vcard