Lots of other apps, meanwhile, endure no such constraints. "Name dialing" and dialogue-based speech recognition, for example, are handily done via centralized media processing on the IP side. Indeed, the fast-growing popularity of IP telephony points the way to a relatively near future in which gateway-style inline call-progress-and-core-audio-related signal processing is entirely decoupled from applications - where most "telephony" apps live entirely on the IP side of the universe, do nothing but (asynchronously) process and generate packet streams via RTP, and do no low- or network-level signal processing at all.
The result has been a sudden explosion of interest in host-based media processing - use of the present generation of really fast CPUs (and future generations of really, really fast CPUs) to do content-based signal processing tasks - either on a client/server basis (coarse asynchronicity) as in name-dialing, or on a much more granular basis, as required by audioconferencing.
The progress curve, here, is dramatic. Just five or six years ago, building a 64-line PC-based conference server required a significant investment in more or less task-specialized DSP-based hardware - bringing with it issues of cost, reliability, integratability and maintainability.
Now you can do the same app on a stock, high-speed Pentium 4 (or in fact, Celeron or PIII) server using Intel's Host Media Processing platform v.1.0, which presents (in programmer-friendly fashion) as a DM3 board, and works in a programming environment that encapsulates loads of functionality (up to and including procedures for conference management). The recent 1.1. release of HMP doubles capacity to 120 ports in a dual-processor server, and adds support for T.38 fax, speech, and G.XXX transcoding.
BLADE SERVERS
