[Asterisk-Users] DSP Coding

[John Todd]

At 10:19 PM -0400 on 6/5/04, Andrew Kohlsmith wrote:
>  > Even the telco's breakout of a DS-3 takes more space than you think.
>>  How would you troubleshoot one DS0? (Very carefully I'd imagine)
>
>In software, naturally.  A physical DS0 needn't exist.
>
>>  (I can all ready imagine the Inermod/crosstalk, RFI of all those DSP's
>>  crammed into a confined space..brrr)
>
>Why would you do something that crazy?  You could put 8 high-end DSPs on a
>half-height PCI card and have each one handle a DS2's worth of channels (up
>to 96) and then have the 8th do general housekeeping of the entire DS3 and
>PCI interface.  Why would you use one DSP per channel?
>
>-A.

Andrew's points are correct.

There exist already cards that will do this, that are even PCI (CPCI, 
though.)  They tend to be crazy expensive despite relatively 
inexpensive parts, as has already been noted.  They also tend to be 
surrounded in marketing gobbledy-gook that makes it impossible to 
determine the true capabilities of the equipment without getting a 
'sales engineer' to cut through the BS and tell you what the card 
actually does.  And as a last nail in the coffin, typically these 
boards are part of larger "architectures" which are impossible to 
purchase in individually useful or programmable components.  ("OH! 
You want the SOFTWARE LICENSE, then, as well!  That's a separate 
contract and price sheet!")

We here in the Asterisk community sometimes fail to see the larger 
possibilities that surround us, and focus only on what the hobbyist 
or single IT person working alone can afford and understand.  The 
telephony hardware market is huge, and has an impressive array of 
vendors producing some really nice cards.  Alas, most of them are 
overpriced because of the niche nature of some of this gear - if you 
spend $300,000 developing the hardware, software, and certifications 
for a card then you can't charge $750 for it, even though that might 
be the cost of the chips and manufacturing.

We (the * community) have this single-minded focus because of the 
items I mention in paragraph 1.  If it's too difficult to understand, 
purchase, or if it's too much money to afford experimentation, we 
won't use it.  That's a shame, since I think there could be some 
really cool parallel-CPU stuff done with third party cards 
(encryption, transcoding, echo cancellation, faxing) if they became 
more available and approachable by the open-source community.  Look 
at the neat stuff that OpenBSD does with the PCI-based encryption 
cards.

I expect a DS-3's worth of physical and transcoding traffic can be 
pushed through a PCI bus machine and into Asterisk, if the 
appropriate amount of 'real' development was put towards the effort. 
('real' in this context equals a team of developers working full 
time, for money.)  I have some doubts if it could be marketed and 
sold in a cost-effective manner by anyone other than Digium at this 
point, though, so it's a moot point.

There have been discussions here on this list already on the 
availability of boards like SBEI's channelized DS-3 card (they've 
been a reasonably approachable vendor.)  All that we need is what 
Andrew describes (a few high-end DSP's on a card) and the software 
extensions to glue all of that into Asterisk.  Markets exist for such 
a combination(I know - I've been in three firms now that would have 
bought such a system) but the real revenues are out there in the land 
of slick salespeople and big trade show booths, which jack up the 
prices out of the range where anyone running Asterisk would be 
interested.   I think if that could be delivered for $5000 (not 
including the PC) then there would be some buyers.  Compare against 
buying a used (not new) Cisco DS-3 card for a 58xx or a Quintum or a 
Nuera with the same capacity.  I will say that the big problem with 
this whole discussion is that when you reach DS-3 levels, running PRI 
just isn't elegant (but certain it's possible.)  Implementing SS7 on 
Asterisk is a much larger issue, and more fraught with danger.  That 
being said, I can also get M-13 DS3-to-T1 muxes pretty cheap these 
days, so just the space savings of a DS3 into a single Asterisk box 
still makes it look appealing versus a slew of PRI's and associated 
card madness.

I don't expect any real comments to come out of this post, and I'm 
uncertain why I even made it.  The people reading this list (you know 
who you are - Hi, guys!) who have an interest in high-density 
Asterisk installations have not and will not ever post to this list 
directly.  There are dozens of companies in this situation (ssssssh! 
It's a secret that they run Asterisk!  What embarrassment that an 
RBOC was using <gasp> OPEN SOURCE!) and it's a shame that this type 
of platform will not be developed due to everyone's reluctance to 
practice what they preach with open source information.

Anyone want to fund an egg or a chicken?

JT