Andrew Kohlsmith (lists) wrote: > On April 7, 2008 02:01:08 am Alex Balashov wrote: > >> A Lucent TNT Max outfitted with _plethoric_ VFCs might work okay. Apex >> too, perhaps. Haven't tried to see how much it can handle when TDM->RTP >> translation is required. >> > > I'm curious; are the cpu/tdm/dsp requirements for 672 g729 rtp streams that > much higher than 672 v92 data streams? I have done work for a dialup ISP > that has probably 20 of the damn things running for quite some time now with > zero issues, and I can't imagine that the RTP requirements are higher than > v92's. > A DS3 *could* be handled, as long as the compute load of transcoding and other heavy tasks is not too great. However, it cannot be satisfactorily handled with the zaptel way of doing things. Zaptel creates a huge storm of small reads and writes, with very little timing tolerance, which congests the machine. Reading and writing 672 modem streams is far more flexible and efficient. The timing is less critical, when things fall behind a little the reads and writes get bigger, automatically increasing their efficiency.
If the driver does a bundle of reads and writes for multiple channels in one go, the throughput on the telephony card side could potentially be quite high, with reasonably tight timing. Then, performance would come down to how many little RTP packet reads and writes you can do. The network stacks really need a mode for bundling there, too. I/O for streaming is still living within the constraints of an I/O model designed to suit bulk, loosely timed, data. The model is a poor fit. Steve _______________________________________________ -- Bandwidth and Colocation Provided by http://www.api-digital.com -- asterisk-users mailing list To UNSUBSCRIBE or update options visit: http://lists.digium.com/mailman/listinfo/asterisk-users