dialogic was RE: [Asterisk-Users] "Glare" condition - How well does asteriskhandle?

[Steve Underwood]

Scott Stingel wrote:

>Hi Steve-
>
>Just briefly:
>
>I was mentioning the old days to illustrate what an even low clock rate DSP
>can do.  More recently (2000-2001), using D/600's we were able to drive a
>large number of channels (8-12 E1's) for IVR.
>  
>
Ah, the D/600 - damned big heat sink, lots of heat, and lots of cooling 
troubles. I remember them well :-) Actually, we are still using quite a few.

>All I'm trying to do is to illustrate both the beauty and the limitations of
>taking the processor horsepower off the line interface card and doing
>everything in the central processor.  The Digium boards are MUCH less
>expensive than the Dialogic boards (about 15% the cost, if that, per
>channel), but are not a plug-in replacement.  I've hit a real-world limit,
>in my less-common environment, of about 4 E1's per chassis.  I believe this
>limitation is not so much in the bit-rate i/o, but the PRI call setup
>overhead.  I have communicated with a number of other asterisk developers
>who have experienced this limitation, again in a high-volume IVR
>environment.   I have demonstrated it to Mark as well...
>
>To get back to the original subject a bit, Dialogic developed an elegant API
>called Global Call, which maybe we can use, or at least learn something
>from.
>
>I'll let you have the last word if you like, Steve.... <s>
>
>Cheers
>Scott
>  
>
If the PRI setup is loading things that much there must be something 
wrong. That should be a very lightweight activity. It sounds like that 
bears investigation. Actually, with host based DSP short PRI calls 
should be a lighter load than long ones. The trunks spend rather more 
time not sending any audio, than with fewer longer calls.

Global Call is a mix of elegance and botchups. I actually have private 
code that implements something I call UniCall for the Digium cards, and 
a chan_unicall channel driver that works with it. This tries to be a lot 
like Globall Call, but without the botchups. I generally agree that an 
abstraction layer like that is a good thing.

I'm not trying to get the last word. I am trying to illuminate and 
improve things. If there are serious limitations they need kicking hard, 
not accepting. :-) There is no sound reason why things should be limited 
in the way you find. If you are doing call switching the Dialogic cards 
will win hands down on throughput. The audio never hits the CPU, and the 
limit is based on the number of call setup/cleardowns rather than active 
channels.
For simple IVR it is certainly not the heavy DSP load on the host CPU 
that is the limiting factor, and the Digium approach should do much 
better than you are finding it actually does. Heck, the Dialogic cards 
still don't bus master. The bus mastering on a TE405P/TE410P should give 
that a substantial win.

Regards,
Steve