Will H.323-Based Video Swamp My Network?
The packaged DVC (desktop videoconferencing) systems we tested, Intel Corp.'s Business Videoconferencing System (BizVideo) and PictureTel Corp.'s LiveLAN, behaved well. However, network managers often cringe at proposals to add videoconferencing traffic to a network that may already be carrying substantial TCP/IP data loads.

We found that the sophisticated audio and video compression algorithms used by these H.323 systems, driven by on-board hardware (an 8x8 chip in LiveLAN) or software (using Intel 200-MHz or greater processors with MMX for BizVideo) can maintain video transmission rates up to 15 fps (frames per second) while using less than 400 Kbps per two-party conversation over a LAN. And if you're willing to accept poorer video quality, these DVCs can be throttled back to 128 Kbps. Audio quality is a separate issue. Gargly, broken up or delayed sound can become a real problem if there are too many bridges or routers introducing variable latency between the end points. The interaction between connected H.323 and H.320 videoconferencing systems is complicated. Video performance is strongly affected by the nature and speed of the processors at each end, but also by the choices of audio compression algorithms in use. Video performance, measured by delivered frame rates, depends most strongly on the power and efficiency of the encoding processor. Intel engineers predict that a Pentium CPU in the 166-to-200-MHz range will use nearly 100 percent of its processing power to encode full-CIF (Common Intermediate Format) video at rates no greater than 5 fps. Add the MMX instruction set, and the output should increase to about 10 fps. You'll need a Pentium II (which incorporates MMX) running at 266 to 300 MHz to achieve video performance that approaches 15 fps.

One inference to be drawn from our tests is that implementation of the new H.323 standards suite is still in its infancy. Developers have a lot to learn about configuring and fine-tuning their systems. If you're considering deployment of a LAN/WAN-based videoconferencing system, then you shouldn't hesitate to invest--at least in some pilot projects--since only basic hardware is required (the hearts of these systems are in software or firmware). And you'll be able to update and improve most systems as vendors release updates.

How We Tested H.323 Videoconference Gateways
To baseline our testing, we first measured the performance of two typical H.323 DVC (desktop videoconferencing) systems: Intel Corp.'s Business Videoconferencing System and PictureTel Corp.'s LiveLAN were used as client systems. The diagram below depicts the simple environment we created. To reduce latency, we did not use bridges or routers.

One reassuring conclusion from our baseline testing is that these H.323 DVCs are not LAN-bandwidth hogs. They can be set to provide remarkably good performance while consuming less than 400 Kbps per two-party connection. Overall, a Business Videoconferencing System transmitting with H.323 directly to a LiveLAN, or vice versa, produces good-quality pictures at 13 to 15 fps (frames per second).

Because Business Videoconferencing System also can place BRI ISDN calls, we connected it directly to our target PictureTel Concorde 4500 using H.320 protocols and observed transmission rates of 10 fps. Finally, we created transmissions from the H.323 DVC units via the LAN, through the two gateway units via ISDN, to the H.320 Concorde and observed performance from 10 to 14 fps. Overall performance ceilings were determined by the end point H.323 or H.320 systems and subtle bandwidth compromises made by their audio and video codecs.