Obviously, there are no DS3 TDM cards that are currently compatible with Zap channels. (or are there?) Does anyone know of an inexpensive DS3 card that could perhaps be used with Asterisk if one were to try to port the Zap drivers to such a card? PCI, of course, would be the bus of choice. I think there are quite a few discouraging comments to be made on that question. Firstly, most companies that produce telecom hardware have silly overhead, and thus the price of their cards is astronomical. Secondly, most companies that produce telecom hardware are of the opinion that transcoding (compression) should be done via DSP's, which inflates the cost of the card significantly. Thirdly, most telecom hardware vendors would not consider allowing their drivers into the public domain if such development were to happen. I've talked to some parties (you know who you are) who have expressed some interest in building this type of interface, but a situation where I can actually put my hands on equipment is far better than speculative interest by those who have not even decided to go forward with design, no matter how interesting the end product sounds on the whiteboard. However, regardless of all these negatives, I'm interested in any vendors anyone can offer as a starting point. Please, don't pester me with comments like "Why do you need 28 PRI's?" or "You'll never use that much capacity." Assume that I actually DO have that volume of traffic, and assume there are several dozen other people on this list (lurkers and active people) who have the same requirements, and assume there are hundreds more people out there who have the requirement but haven't considered Asterisk because DS3 isn't an option. JT
I believe there are boxes that will take a DS-3 from the Telco and spit out T-1's to your telecom equipment. Not sure what they are called. John Todd wrote:> > Obviously, there are no DS3 TDM cards that are currently compatible with > Zap channels. (or are there?) > > Does anyone know of an inexpensive DS3 card that could perhaps be used > with Asterisk if one were to try to port the Zap drivers to such a > card? PCI, of course, would be the bus of choice. > > I think there are quite a few discouraging comments to be made on that > question. Firstly, most companies that produce telecom hardware have > silly overhead, and thus the price of their cards is astronomical. > Secondly, most companies that produce telecom hardware are of the > opinion that transcoding (compression) should be done via DSP's, which > inflates the cost of the card significantly. Thirdly, most telecom > hardware vendors would not consider allowing their drivers into the > public domain if such development were to happen. I've talked to some > parties (you know who you are) who have expressed some interest in > building this type of interface, but a situation where I can actually > put my hands on equipment is far better than speculative interest by > those who have not even decided to go forward with design, no matter how > interesting the end product sounds on the whiteboard. > > However, regardless of all these negatives, I'm interested in any > vendors anyone can offer as a starting point. > > Please, don't pester me with comments like "Why do you need 28 PRI's?" > or "You'll never use that much capacity." Assume that I actually DO > have that volume of traffic, and assume there are several dozen other > people on this list (lurkers and active people) who have the same > requirements, and assume there are hundreds more people out there who > have the requirement but haven't considered Asterisk because DS3 isn't > an option. > > JT > _______________________________________________ > Asterisk-Users mailing list > Asterisk-Users@lists.digium.com > http://lists.digium.com/mailman/listinfo/asterisk-users
On Thu, 2003-12-04 at 14:06, John Todd wrote:> Obviously, there are no DS3 TDM cards that are currently compatible > with Zap channels. (or are there?) > > Does anyone know of an inexpensive DS3 card that could perhaps be > used with Asterisk if one were to try to port the Zap drivers to such > a card? PCI, of course, would be the bus of choice.Your first problem will be bus speeds. A single DS3 is 44.736Mbps. each way. So if you double this and get the 89.472Mbps, you are going to be coming close to the real limits of the 32/33mhz PCI bus without having done any work on the data you are shuffling. So while it could be done here, I'd start worrying about stability. Sure you could switch up to faster PCI buses like the 64/66mhz bus, but then you will start limiting what systems you can use. Then again, if you are putting that many channels through a single machine, there wasn't many choices for the hardware to begin with. My question I guess would come down to why bring a DS3 into a PC when you could get a multiplexer that took your DS3 and split it down to T1s so you could use already developed hardware. You could then build in some redundancy and if a machine goes down and takes a few T1s down with it, you just route around it till you fix it. Figure you could go 7 1U super micros with a TE410P in each one will get you your 28 T1s. Then either make one machine do the work of traffic cop and connect calls between T1s or point them down the line to other machines that can then terminate the call. Since the traffic cop machine wouldn't need to actually service calls but for a short period during routing, if it were to fail it would just drop any calls it was in the middle of routing and wouldn't route new calls. In this case you could have a hot spare waiting in the wings to do a on fail dial here type route and it could service the new calls. In this case you have 9u of space used and no machine can take down more than 4 T1s worth of calls. Ohh, you need to add 1u of space fore the multiplexer. So 10u of space for a DS3 using currently available technology and software with a bit of failover support. -- Steven Critchfield <critch@basesys.com>
I don't want to criticize your idea, but you do have to consider certain points. Starting from (as has already been mentioned) the bandwidth of DS3 is far too much to reasonably shove down the PCI bus without data loss / excessive overheads. Thus a sensible approach would be one where the card performs the switching, (H100/H110 or otherwise), leaving the Asterisk unit to maybe handle signalling and call control only. You could go one further, and if you require 'voice' resource, to switch that onto the PCI bus as well for processing. The way I see this, the best implementation plan would actually be to take a standard DS3 card with a H110/H100 bus, and then look for a third party card which could switch timeslots on the H110/H100 bus to the PCI bus. This composite approach would allow a zero latency switching path, but still include the flexibility of Asterisk. However, considering the traffic volumes that you are talking about, is it really true to say that the traditional telco cards are astronomically priced, given the amount of revenue that can be generated per month on a DS3? Linus ----- Original Message ----- From: "John Todd" <jtodd@loligo.com> To: <asterisk-users@lists.digium.com> Sent: Thursday, December 04, 2003 8:06 PM Subject: [Asterisk-Users] Port density: DS3 cards?> > Obviously, there are no DS3 TDM cards that are currently compatible > with Zap channels. (or are there?) > > Does anyone know of an inexpensive DS3 card that could perhaps be > used with Asterisk if one were to try to port the Zap drivers to such > a card? PCI, of course, would be the bus of choice. > > I think there are quite a few discouraging comments to be made on > that question. Firstly, most companies that produce telecom hardware > have silly overhead, and thus the price of their cards is > astronomical. Secondly, most companies that produce telecom hardware > are of the opinion that transcoding (compression) should be done via > DSP's, which inflates the cost of the card significantly. Thirdly, > most telecom hardware vendors would not consider allowing their > drivers into the public domain if such development were to happen. > I've talked to some parties (you know who you are) who have expressed > some interest in building this type of interface, but a situation > where I can actually put my hands on equipment is far better than > speculative interest by those who have not even decided to go forward > with design, no matter how interesting the end product sounds on the > whiteboard. > > However, regardless of all these negatives, I'm interested in any > vendors anyone can offer as a starting point. > > Please, don't pester me with comments like "Why do you need 28 > PRI's?" or "You'll never use that much capacity." Assume that I > actually DO have that volume of traffic, and assume there are several > dozen other people on this list (lurkers and active people) who have > the same requirements, and assume there are hundreds more people out > there who have the requirement but haven't considered Asterisk > because DS3 isn't an option. > > JT > _______________________________________________ > Asterisk-Users mailing list > Asterisk-Users@lists.digium.com > http://lists.digium.com/mailman/listinfo/asterisk-users >
At 02:34 PM 12/4/2003, you wrote:>However, considering the traffic volumes that you are talking about, is it >really true to say that the traditional telco cards are astronomically >priced, given the amount of revenue that can be generated per month on a >DS3?Eight quad-span T-1 cards from Digium: $8,970 Three reasonable-quality asterisk servers: $1,000 One T-1/DS-3 MUX: $5000 Total system cost: $14,970 That actually sounds quite reasonable to me. However, if I were doing this myself I would look hard at getting a MAX TNT with VoIP capability off eBay. The price would be equivalent or less, the interface would be more complicated, but all the DSP would be done by the MAX. --Ernest
Correcting an idiot-math error (24/4 != 8 and 1000*3 != 1000) ... At 02:34 PM 12/4/2003, you wrote:>However, considering the traffic volumes that you are talking about, is it >really true to say that the traditional telco cards are astronomically >priced, given the amount of revenue that can be generated per month on a >DS3?Six quad-span T-1 cards from Digium: $8,970 Three reasonable-quality asterisk servers: $3,000 One T-1/DS-3 MUX: $5000 Total system cost: $16,970 That actually sounds quite reasonable to me. However, if I were doing this myself I would look hard at getting a MAX TNT with VoIP capability off eBay. The price would be equivalent or less, the interface would be more complicated, but all the DSP would be done by the MAX. --Ernest
I would be seriously wary of putting a DS3's worth of voice traffic on a TNT. I don't believe they are rated to handle that much voice. The APX1000 would be a much better platform, but I don't know if you can find one used. Stephen> -----Original Message----- > From: Ernest W. Lessenger [mailto:ernest@oacys.com] > Sent: Thursday, December 04, 2003 4:51 PM > To: asterisk-users@lists.digium.com > Subject: Re: [Asterisk-Users] Port density: DS3 cards? > > At 02:34 PM 12/4/2003, you wrote: > >However, considering the traffic volumes that you are talking about,is> it > >really true to say that the traditional telco cards areastronomically> >priced, given the amount of revenue that can be generated per monthon a> >DS3? > > Eight quad-span T-1 cards from Digium: $8,970 > Three reasonable-quality asterisk servers: $1,000 > One T-1/DS-3 MUX: $5000 > > Total system cost: $14,970 > > That actually sounds quite reasonable to me. However, if I were doingthis> myself I would look hard at getting a MAX TNT with VoIP capability off > eBay. The price would be equivalent or less, the interface would bemore> complicated, but all the DSP would be done by the MAX. > > --Ernest > > > _______________________________________________ > Asterisk-Users mailing list > Asterisk-Users@lists.digium.com > http://lists.digium.com/mailman/listinfo/asterisk-users
John Todd wrote:> > Obviously, there are no DS3 TDM cards that are currently compatible > with Zap channels. (or are there?) > > Does anyone know of an inexpensive DS3 card that could perhaps be used > with Asterisk if one were to try to port the Zap drivers to such a > card? PCI, of course, would be the bus of choice. > > I think there are quite a few discouraging comments to be made on that > question. Firstly, most companies that produce telecom hardware have > silly overhead, and thus the price of their cards is astronomical. > Secondly, most companies that produce telecom hardware are of the > opinion that transcoding (compression) should be done via DSP's, which > inflates the cost of the card significantly. Thirdly, most telecom > hardware vendors would not consider allowing their drivers into the > public domain if such development were to happen. I've talked to some > parties (you know who you are) who have expressed some interest in > building this type of interface, but a situation where I can actually > put my hands on equipment is far better than speculative interest by > those who have not even decided to go forward with design, no matter > how interesting the end product sounds on the whiteboard. > > However, regardless of all these negatives, I'm interested in any > vendors anyone can offer as a starting point. > > Please, don't pester me with comments like "Why do you need 28 PRI's?" > or "You'll never use that much capacity." Assume that I actually DO > have that volume of traffic, and assume there are several dozen other > people on this list (lurkers and active people) who have the same > requirements, and assume there are hundreds more people out there who > have the requirement but haven't considered Asterisk because DS3 isn't > an option.I think this is a worthwhile thing to investigate. Does anyone here have experience with higher order cards under Linux? Which ones work well, and have solid drivers? Although the driver would (probably) need to be heavily modified if it is currently a data, rather than telephony, oriented driver, a good existing driver should save a lot of work. A DS3 is well within a PCI channel's capacity (the sum of the two directions is less than a 100mb Ethernet, although the DS3 is continuous), but it is quite a lot of data. A suitable card would need an efficient interface if this is to work. For the Zaptel environment that would mean that it can burst data in 1ms (8 sample chunks), and would need to bus master the data into memory in a form that doesn't require masses of manipulation by software - e.g. reshuffling out of sequence data. Doing that for so many channels might create interesting latency challenges :-) Still, if you don't try, you can't start to address these issues, and work out a solution. It shouldn't be impractical to make a DS3 to TDMoE solution, provided the DS3 hardware is right, and the PC has no quirky throughput issues. This subject often comes up on the IRC channel. There seem to be a number of people interested in higher order links, but it really needs some positive action somewhere to kick off a real project. Regards, Steve
> > >I have no reason to disbelieve this report, but I will offer some >minor scepticism at this reply. A well-equipped PC can currently >handle 8 T1 channels, and it seems that only the IRQ issue is causing >more channels to not be viable in the current TE410P environment. It >would seem reasonable to think that a very well equipped PC (4-way, >8-way?) would be able to handle the "processing power" requirements >of a DS3, whatever was meant by that statement. Of course, there may >be other underlying issues specific to ImageStream that make this >impossible; I don't know. > >JT >Try to look this building block that should allow you to do T3 to TDMoE at wirespeed. The chip can move data between T3 and Ethernet without touching the PCI bus but you can still keep full control via the PCI bus if you want to. http://products.zarlink.com/product_profiles/ZL50111.htm
On Thursday 04 December 2003 14:06, John Todd wrote:> Obviously, there are no DS3 TDM cards that are currently compatible > with Zap channels. (or are there?)This isn't so much a technological limit as much as a detail of implementation, but the current Zaptel drivers have a limit of 252 channels per machine. This is due to each channel number getting assigned a device minor number, which has a limit of 256 (4 are reserved by zaptel for special purposes). Any attempt to write a driver for a DS3 card should take this into account. I don't know if expanding beyond 252 would be an easy change to the Zaptel architecture or if this would require a major change to the way the drivers are written; but it is a limitation that needs to be considered. -Tilghman