Is the voice in voice over IP good enough to bet your business on?

That's the make-or-break question for many network executives as they consider the promise of integrated voice/data. And it's one of several key questions we'll try to answer in this four-part series designed to help companies plotting convergence strategies.

Most users and industry watchers say it's relatively easy to achieve toll quality in the lab, but it can be more difficult in a production network. That helps explain why some companies see a future with VoIP everywhere, while others are hesitant to use it on critical links, or for customer-facing applications, such as in call centers or help desks.

"Traditional businesses like ours are . . . on the conservative side. We're not willing to jeopardize our brand name," says Jeff Fountaine, a network analyst with Armstrong World Industries, a Lancaster, Pa., maker of industrial and home flooring and ceiling products. Among his concerns would be a phone order from going awry because of poor sound quality over an IP link.

Is this a do-it-yourself project?
Part 2 of this series
Users hoping SIP's the answer
Part 3 of this series
Answers to your VoIP questions
Part 4 of this series

Still, the potential cost savings are strong enough that Armstrong is willing to give VoIP a shot for certain applications. The company is planning to test an internal campus-to-campus IP telephony deployment before delving deeper into the technology.

Quality measurement

In a recent Network World Survey of 250 IT executives, the top perceived drawback for network convergence was the lack of quality-of-service (QoS) assurance on corporate networks. Almost half of those surveyed said that the quality of IP voice was a drawback.

With voice quality being such a sticking point, the trick is to come up with a good way to ascertain whether your network can support toll-quality VoIP. Some experts say there are hard tests and metrics for proving an IP telephony system, while others say the process is more art than science.

Factors that might diminish the quality of an IP phone conversation rarely lie in the actual VoIP gear anymore, says Mike Hommer, manager of lab testing at Miercom, an independent IT testing and consulting firm and a member of the Network World Global Testing Alliance.

"When we started testing VoIP products in 1997, 80% of the metrics we looked at were related to the performance and voice quality. Now that's down to about 10%," he says. "The quality issues - as far as IP voice equipment being able to efficiently encode and decode voice - have become less of a concern."

The issue now is on the network, Hommer says.

"Some people may have no idea how good or bad their network is for supporting real-time protocols like voice," he says.

Network latency is the No. 1 killer of real-time packetized voice traffic, Hommer says. The result of latency is jitter, which can cause an IP voice conversation to break up. Most IP voice products have jitter buffering technology or other technology that smoothes out and reorders voice packets before turning them into audio, but sometimes excessive network latency cannot be overcome.

Packet analyzers and port mirroring applications can be used to measure IP traffic volume and patterns to determine the sources of latency on a LAN or WAN. There are also voice-specific tests and methods that can be used for a finer level of testing.

One time-tested method for determining voice quality is the Mean Opinion Score (MOS), a test accepted by the International Telecommunications Union whereby a sample of 40 or more people from different ethnic or language backgrounds are given an audio sample that's several seconds long to listen to, and each person rates the quality of the audio on a scale from 1 to 5. By ITU standards, 5 is a "perfect" MOS score, while 4 is considered "toll-quality" or a high enough standard for delivering land-line service by public switched telephone network carriers.

"The good side of [MOS] is that you're using a human ear to determine voice quality," Hommer says. "The bad thing is that it doesn't take into account things like bidirectional speech quality," where a segment of speech might sound clear, but network congestion or inefficient routing might create pauses in a conversation, he adds.

Another metric for gauging VoIP quality is the Perceptual Speech Quality Measurement (PSQM), which is a computer algorithm used in testing tools from vendors such as Agilent, Empirix, Finisar and NetIQ for checking out IP voice equipment and network performance. Hommer says these tools are useful but don't always produce accurate results. Some of the computerized metrics, he says, are so stringent, "that the algorithm gives it a bad score, but from a human ear standpoint, you can't tell the difference."