[Asterisk-Dev] channel driver sample - Coding/Transcoding/Conferences

John Todd jtodd at loligo.com
Sun Jul 4 19:35:14 MST 2004

At 10:49 AM +1000 on 7/5/04, Adam Hart wrote:
>  Steve Underwood wrote:
>>People don't want a pile of xbox2s to run a telco. They want neat, 
>>compact, manageable systems. Packaging is important. One 
>>disadvantage of a DSP card separate from am E1 card is you can't 
>>get 2 PCI cards in a 1U rackmount.
>TI's chips aren't that impressive at their price, espically if I 
>require a seperate DSP card. A combined E1 card is different, an E1 
>card delivering a G.729 stream would rock.

To comment on this and a prior message about "dual-purpose" cards: 
look, I have a good deal with the Digium cards.  I cannot ever 
possibly knock the cost per port that I get right now for Digium's 
E1/T1 interfaces, and I suspect that the only thing I'd want to see 
is the DS3 card, but even that is marginally interesting as I move 
away from physical interconnects to the TDM world and more towards 
SIP interconnects.  I'm not an SS7-based carrier, so this isn't a 
problem for me.  Let Sonus et. al. carry that burden, if they ever 
get the ability to do crazy stuff like codecs other than G.711. 

However, I strongly disagree about having the physical interface 
(E1/T1/DS3/etc) and DSP on the same card.  I have requirements _right 
now_ for just transcoding between different codecs without any TDM 
interfaces, and having to pay for a physical interface that I'll 
never use seems to be pretty wasteful to me.  Digium cards are too 
cheap to worry about the cost for the interfaces; the DSP chipset and 
the TDM interconnect should be modules, not combined.  The comment 
about using up another slot is valid, but not that important either - 
I can get 1u boxes with 1 PCI slots, and also I'm not heartbroken 
about getting a 2u machine with three PCI slots in it (8 E1/T1 cards, 
plus a DSP.)  The trick is getting enough horsepower in the DSP card 
to make it worth the jump now instead of waiting 2 years until the 
CPU's catch up.  2 years is a long time; anyone who can come up with 
a reasonable card now has a chance to make some good money if they 
can get Asterisk to speak to their card "transparently."

I'm with Steve again - having a neat, compact system that "just 
works" is better than a pile of machines that to the job 20% cheaper 
but cost me serious time and effort to manage - if I had a million 
customers and a fifty million dollar budget, my tune would change, 
but I don't and NOR DOES ANYONE ELSE.  The Google model will not work 
yet with Asterisk unless you already have the sysadmin staff 
in-house, and even then it's questionable.  I'm not foolish enough to 
say that I want Cisco boxes everywhere in my network, but if I could 
somehow get two hundred AS5300's each at the same cost as a low-end 
PC with a Digium card in it, you know I'd leap at the chance.  (But 
I'd still use Asterisk in the "core".)


More information about the asterisk-dev mailing list