Network Computingwww.networkcomputing.com

To confirm how well your design and equipment choices will play out on the real network, load testing is essential. Systems like Agilent Technologies' FASTest and LAN Analyzer, Shomiti Surveyor, Sitara QoSWORKS, and Spirent Communications' SmartTCP or SmartQoS can be employed to generate traffic and/or measure jitter and delays in selected network segments.

We recently had an opportunity to examine a valuable new tool from Agilent that provides actual end-to-end voice quality measurement on a network. In our Real World Live Labs testing setup at Networld + Interop, we connected a pair of Komodo EF200 Ethernet adapter beta units back-to-back over a simple 10/100 IP switch. On each of the EF200's analog telephone input ports, we made an RJ-11 connection to the Port A / Port B test points of an Agilent Telegra R Voice Quality Tester (VQT), which is packaged in a rugged Dolch Flexpac PC carrying case.

Cisco plans to make the Komodo adapters, which are still under development, available through its service provider line of business, although it declined to specify a ship date. These adapters are 15 x 14 x 4 cm desktop units designed for resale by Internet service providers to connect two legacy telephones or fax units to a 10Base-T IP network. The Komodo/Cisco units can be set to use G.711 or G.723.1 voice encoding/decoding (G.729A will be added later), and payload sizes of 2,4,8, and 12 frames per packet.

We were interested in learning the differences in voice quality and end-to-end delay that could be observed with different vocoder algorithms and packet size settings. A Telegra R VQT can measure key voice quality parameters of clarity, echo, and delay, and uses Perceptual Speech Quality Measurement (PSQM), Perceptual Analysis Measurement Systems (PAMS), and real speech samples.

The Telegra R VQT measures end-to-end or round trip delay by performing an impulse response measurement and cross-correlating the received signal with the transmitted signal. This method is much less susceptible to noise, loss, and attenuation, than acoustic pinging. Perceptual speech quality is measured directly, using techniques and algorithms specified by the ITU recommendation P.861. (See the output displays in Fig. 1 and Fig. 2.) Under PSQM, lower scores are better. Because many practitioners are familiar with the traditional Mean Opinion Scoring technique, the Telegra R VQT uses a correlation function to report an average PSQM and an equivalent MOS after each test.

A cautionary note: These tests were aimed at showing the value of using a consistent, automatic, and repeatable testing device rather than relying on MOS or other subjective listening tests to evaluate implementations of voice over the WAN. However, these Telegra R VQT test runs reveal some interesting things that may help explain the Komodo/Cisco adapter's performance.

Figure 3 documents the adapter's performance when both transmitting and receiving units are set to use G.723.1 encoding and decoding, an algorithm designed to compress voice to about 5 or 6 Kbps. Given that the ping time through our simple 10/100 Base-T switch was less than 30 ms, we were surprised to observe total end-to-end delay times, which include encoder look-ahead, compression, network transit, and decompression, as high as 170 ms to 180 ms. PSQM ratings were high and MOS low--probably a function of the overall delay time. Although the adapter's default configuration is to load two audio sample frames per packet, we noted that performance improved slightly at 4 frames per packet, then fell off dramatically as bigger packets contributed to higher latency.

Figure 3 - Komodo/Cisco Adapter using G.723.1

Figure 4 - Komodo/Cisco Adapter using G.711A