November 1997

BY MIKE COFFEE

Seven leading vendors in the computer telephony integration (CTI) industry, including BICOM, Calibre Industries, Centigram Communications, Cole Technical Services, Commetrex, Pika Technologies, and QNX Software Systems, have joined forces and founded the MSP (media stream processor) Consortium. The goal of the Consortium is to develop a software specification which the members believe will help move the CTI industry towards shrink-wrap media-processing software.

The Consortium founders believe the industry can benefit from a valueadding interface between system resource hardware and media-processing software, just as the PC industry benefited from open architectures which have allowed different companies to develop a computer system’s hardware and software subsystems. Today, in CTI, the same company that develops the media-processing software also develops the underlying hardware.

NEW ARCHITECTURES CREATE NEW BUSINESSES
Every vendor in the industry works continuously to improve their product. But generally, these improvements do not result in changes in the business definitions of their competitors. It takes a fundamentally new product, which creates a new competitive space, to do so. Prior to Dialogic’s first voice board, other companies had developed multiline voice-processing hardware. But, by offering this system resource with its software interface to other companies, Dialogic created the value-adding computer-telephony industry. Suddenly, hundreds of companies appeared to take advantage of the opportunity Dialogic’s innovation offered.

When Natural MicroSystems and Rhetorex released DSP-based multiline voice boards in 1988, the technology was improved, but there was no change to the industry’s structure. That change occurred again with the 1990 announcement of MVIP by Natural MicroSystems and 6 other companies (many of them competitors). MVIP fundamentally altered the industry’s structure by creating a new value-adding interface which dramatically improved the industry’s value-creation efficiency.

The common factor of these critical milestones in the history of CTI are architectural innovations that improved the efficiency of the industry’s valueadding chain. The ambitious goal of the MSP Consortium is to bring about such an improvement in the efficiency of the CTI industry by developing the Media Stream Processor, a software specification which supports the separation of media-processing software (firmware) from the underlying hardware resources, creating new competitive arenas through the definition of a much-needed value-adding interface.

CTI AND THE PC INDUSTRY: VALUE-ADDING PARTNERS
The advantages of layers of value addition in the computertelephony industry are the same as those in the computer industry — it’s one of the chief reasons CTI has grown to its near-$10 billion size in just 13 years. With standardized interfaces between adjacent layers of value addition, companies can choose to compete at a particular valueadding level, as is done to a greater extent in the computer industry. You can produce PCs without having to develop network adapters. You can develop video controllers without having to develop operating systems. And when an OEM purchases these components to create an application-specific system, there’s a good chance they will all work together. What’s more, competition within a value-adding layer fosters rapid improvements in price, performance, and function. As a result, the market expands, producing returns to the participants which exceed those gained with closed-architecture strategies.

PHASE I — THE VOICE BOARD
In the history of value-adding com-putertelephony, there have been two main architectural innovations that rewrote the rules of the market, let’s call them architectural phases. Phase I began in 1984, when Dialogic created the industry with the introduction of the first multiline “voice board.” Prior to that, the entire telecommunications industry was vertically integrated and relatively inefficient. Dialogic’s innovation released a flood of entrepreneurism, which produced a multi-billiondollar industry in less than a decade.

Before Dialogic’s voice board, a voice mail product, for example, could only be brought to market if a company was willing to invest the development resources necessary to produce the entire vertically integrated system. But with Dialogic’s voice board, OEMs could add application value starting where Dialogic left off, as shown in the Phase I diagram (Figure 1). But until the second architectural phase (Phase II) the industry was essentially limited to one medium — voice.

PHASE II — THE PCM HIGHWAY
Phase II changed that when, in 1990, Natural MicroSystems and its partners introduced Multi-Vendor Integration Protocol (MVIP). As the name implies, MVIP allows an OEM to integrate sys-temlevel hardware resources (boards), that are independently developed and marketed, into a system that is more functionally integrated than was previously feasible. OEMs could then produce systems which supported fax, text-tospeech, and speech recognition — as well as voice — even though different companies produced those media-processing system-resource boards.

With MVIP, the OEM’s valueadding interface is the same as it was previously, but it became possible to develop a system which supported more than one medium, and switch the PCM stream between the dedicated-function resource boards (Figure 2). Moreover, MVIP increased the variety of network interfaces available to the OEM. MVIP allowed companies with competencies in ISDN interfacing, for example, to offer the CTI OEM an MVIPcompatible ISDN network interface. This one-func-tionper-board architecture of MVIP sys-tems is an effective way to marshal industry resources as long as integrated circuits, and DSPs in particular, do not significantly improve in performance.

ENTER TODAY’S DSPs
Today’s DSPs have the power to support either a single medium (such as voice, in densities well over 100 ports) or high-density multiple media. This brings into question the single-function-perboard Phase II architecture, since the increased processing power makes it uneconomical to limit hardware resources to one function. But developers of media-processing technology are faced with a very significant barrier to developing the products which will translate these advances in silicon into the benefit of the industry’s end users: There is no value-adding interface which allows them to develop mediaprocessing software product independently of the industry’s hardware vendors who are busy developing closedarchitecture media-specific products.

PHASE III — MEDIA INTEGRATION
Media-processing and hardware vendors must work together to produce integrated-media system resources because the vendors of voice boards, still the industry’s mainstay, do not have the core competencies to develop fax, text-to-speech, speech recognition, high-speed data, and video media-processing software. Clearly, OEMs need system architectures which allow them to flexibly bind a media stream to a media-processing resource. So, without an interface specification, media-processing and hardware partners are wasting millions of dollars a year porting different technologies to the board vendor’s closed-architecture DSPresource boards. From the OEM and enduser’s perspective this is value-shuffling, not value-creation. The same development resources, invested in improving the technology — rather than in moving it between proprietary boards — delivers more value to the end-user for a given industry investment.

The purpose of the MSP specification, then, is to offer independent vendors a means of avoiding this constant porting effort by finally providing the industry with a truly open, resource-rich architecture appropriate for all media. MSP will be an open, published architecture, intended to be the enabling specification for any vendor to develop and market MSP-compliant hardware and/or mediaprocessing software products.

Today, voice-board vendors are taking advantage of the rapid advances in DSP power by developing products which support 64- and 128-voice streams. Although there are applications which can take advantage of such density, there are many more that could take better advantage of a system which used those same DSP MIPS to apply any required processing (fax, data, voice, TTS, ASR, video) to the media stream. The voice-board vendors are not closed to the idea of leveraging these MIPS to supply additional media. Dialogic has announced DM-3, a system which will support cooperative efforts between Dialogic and thirdparty media-processing developers. Brooktrout has announced its BOSTON architecture, which supports integrated media. But these architectures still don’t match the powerful paradigm of the PC industry. The MSP specification makes it possible to do exactly that. MSP fully supports multivendor media-diversity by managing the resource demands of media-processing software (Figure 3), just as the latest client/server software systems from NMS and Dialogic manage system resources from the host level.

THE MSP: INCREASED INDUSTRY EFFICIENCY
The goal of the MSP Consortium is to make it economical to leverage the power of today’s DSPs to put all of a system’s necessary mediaprocessing on one hardware resource, taking advantage of the incredible power of today’s DSPs. Just as we don’t use one PC for spreadsheets, one for text editing, and another for software development, we would no longer need one board for voice, another for fax, and yet another for speech recognition. The specification will make media-processing consolidation easier to accomplish.

It will provide a defined environment for the software facilities needed to allow independent vendors’ media-processing products to operate cooperatively on the same hardware resource. With MSP, instead of integrating fixed-function boards with an industrystandard PCM highway, such as MVIP, the OEM can integrate fax, voice, data, text-tospeech, speech recognition, and video media-processing products on a single, powerful, multi-vendor DSP resource.

The MSP specification could fundamentally change the structure of the value-adding CTI market: Today, the development of media-processing products usually involves developing not only the media-processing software, but the hardware platform as well. But MSP creates an additional vendorindependent value-adding segment by separating the media-processing software (firmware) from the DSP-resource hardware. With a standard definition of the media-processing system-resource platform, companies can choose to develop just MSPcompliant hardware resources; others can develop MSP-compliant media-processing products. Still others may choose to integrate MSP into different hostlevel software environments.

The big winners will be the industry’s OEMs and end users. And if the end user wins, everyone wins. The ultimate mission of the MSP Consortium is to create the age of off-the-shelf, plug-and-play, shrinkwrapped media-processing software.

ECTF SOFTWARE MODEL
The ECTF (Enterprise Computer Telephony Forum) has become the computer telephony industry’s de jure standards body, but it has not defined a standard at the value-adding layer of the MSP. The ECTF model supports a clientserver architecture through its incorporation of client APIs (defined by S.100), a protocol for communicating with server entities (S.200), a Service Provider Interface (S.300), and a PCM highway (H.100).

Were the MSP to be placed into the ECTF context, it would exist below the S.300 SPI. MSP defines the software environment for mediaprocessing resources which do the boilerroom work of CTI. The MSP specification does not specify protocols for communicating with a host entity; it does not specify a hardware architecture, PCM highway, DSP, or embedded OS. But it does define an environment for multiline media-processing software.

WHAT, THEN, DOES IT SPECIFY?
The keystones of the MSP are its Resource Manager and Task Processor Manager. These entities manage the widely varying resource requirements of different media. A voicecompression task may require fewer than 5 MIPS, while a V.17 (14.4-kbps) fax modem may require 13 MIPS. If each media-processing system resource is to be independently supplied by different vendors, there must be a board-level resource manager which processes requests for MIPS, RAM, and PCM streams. Of particular importance is the allocation of what the specification calls Task Processor (DSP) MIPS. The TP Resource Manager is responsible for managing the board’s DSPs as dispatchable processor objects.

The MSP specification also defines how to "package" a DSP processor object so that the stream-processing task can be dispatched in a DSP-independent manner. To build an MSP-compliant mediaprocessing software product, a developer presents DSP-specific mediaprocessing software to a utility, which takes the code and puts it into a task-processor object module, which can be handled by the TP Manager. DSP-specific packaging utilities will be developed by any interested vendor and made available to all vendors via the MSP Consortium Web site (www.msp.org). This means that, theoretically, an MSP-compliant media-processing product will execute on an MSP platform provided it is "packaged" for the DSP used by the platform.

The Consortium is currently working on API definitions for stream management, resource management, task processor management, CTI services, and communications services.

Mike Coffee, of Commetrex, Inc., is the designated MSP Evangelist. Commetrex has successfully ported its MultiFax media-processing software to numerous hardware platforms, including Natural MicroSystems’ VBX and AG boards, BICOM’s LS-E, and CCS’s FirstLine, in addition to traditional office fax products. The MSP Consortium founders are soliciting the assistance of any interested company. For more information, please visit the MSP Consortium Web site at www.msp.org and contact Mike Coffee at [email protected] or 770-449-7775 x310.