April 2003

Measuring Voice Quality in VoIP Networks: Going Beyond "Can You Hear Me Now?"

BY LAURA HOLLY

Enterprises everywhere are looking carefully at VoIP these days. And no wonder: the economic advantages of sending voice and data over the same low-cost, IP network can be attractive.

Of course, adopting VoIP means different things to different companies. Some find that bridging their existing PBX systems via IP trunking is a cost-effective way to leverage their IP network and preserve their investment in traditional circuit-switched infrastructure. Others are extending IP telephony throughout the enterprise with software-based call servers and IP phones on the desktop. Additionally, many enterprises are also taking advantage of IP-based audio conferencing and call centers. Smaller companies, eager to offload voice infrastructure maintenance, are turning to provider-hosted solutions, such as IP Centrex or unified messaging services.

But whereas some IT professionals are leaping in with both feet, others are hanging back. And while some are not yet convinced of the VoIP business case, the vast majority is hesitant due to well-founded concerns about voice quality -- especially for calls that will be routed over a WAN or the Internet. Voice is a mission-critical application and many enterprises are simply unwilling to make the leap to VoIP until they can be assured that users will enjoy the same voice quality and call reliability they currently have with their legacy PBX/PSTN infrastructure for each and every call.

IT�S THE NETWORK

While a number of factors can degrade voice clarity, such as poor-quality endpoint equipment or repeated voice signal compression and decompression, by far the number one culprit is the underlying IP network itself. Packet loss, either due to packets dropped by the network or by an endpoint jitter buffer because of excessive network jitter and latency, can significantly degrade the quality of voice transmission. Although voice clarity is an essential indicator of quality, other factors must also be considered. For example, irritating delays caused by excessive network latency can result in parties talking over one another or adopting a walkie-talkie style of conversation. In addition, the call connection experience also impacts users� overall perceptions of call quality. If users don�t receive a prompt dial tone when taking a phone off-hook, or if their calls take a long time to connect, fail to connect, or disconnect prematurely, users will be dissatisfied -- even if the voice quality is stellar.

Taken together or separately, any of these problems can make carrying on a normal conversation an exercise in frustration. To make matters worse, these performance factors are not static targets -- the fact that call quality was acceptable yesterday or even ten minutes ago does not ensure good quality on the next call, due to the dynamic nature of IP networks and their ever-changing usage patterns.

MEASURING QUALITY

Given these network realities, how can enterprises ensure both clear voice quality and reliable call connections? While television ads imply that call quality can be assured simply by roaming about with a phone and asking, �Can you hear me now?,� the reality is quite different. To meet users� high-quality standards, they must have an effective approach to measuring VoIP network and application performance. This approach must be:

� Pervasive -- Measurement must encompass all IP network infrastructure and services engaged in delivering the VoIP application.

� Continuous -- The VoIP network must be measured continuously, in real-time, and throughout the life cycle of the VoIP application.

� Comprehensive -- Network measurement must include both active performance verification and passive performance monitoring.

Let�s review each of these critical network measurement characteristics in more detail.

Pervasive Measurement

The only meaningful measurement approach is one that accurately measures the entire IP infrastructure engaged in delivering the VoIP service. Anything less will result in incomplete data and a misleading analysis of service performance. All of the components that support the voice call must be exercised, including, of course, network, DNS and DHCP service, gateway, gatekeeper, proxy server, or other call and application server performance.

Performing effective measurement requires an enterprise-wide approach, with instrumentation at all service origination or delivery points in the network, such as each branch office or data center. Enterprises deploying VoIP to the desktop should also extend their measurements within locations by measuring call performance directly to endpoint or IP phones at the site.

Each measurement point is instrumented with a verifier, which, depending on the application and deployment location, may either be a purpose-built hardware appliance or a lightweight software agent designed specifically to perform VoIP testing. Verifiers measure VoIP transactions and communicate raw transaction performance data to a centralized administration portal that provides data analysis, aggregation, and real-time reporting. This pervasive approach is the only way to precisely pinpoint the causes of poor voice and call quality.

Continuous Measurement

Measuring the network intermittently will not tell you much about VoIP quality. That�s because IP networks -- unlike more static and predictable circuit-switched networks -- are by their very nature dynamic. With asymmetric routing, dynamic routing modifications, ever-changing traffic loads, and bandwidth availability, network performance can vary from one moment to the next. Continuous measurement is, therefore, essential.

When we say �continuous� we mean both around the clock and throughout the VoIP application lifecycle, including pre-deployment testing right through the deployment and production phases. Measuring application and network performance before deployment is critical to avoid unpleasant surprises when the application is launched. Many VoIP quality problems -- and unhappy users -- could have been avoided had they been identified pre-launch. Incremental testing before and during deployment also makes the task of validating the network more manageable, accelerating time-to-launch, and avoiding costly and time-consuming analysis and redesign after deployment.

The need for monitoring doesn�t end once the VoIP application is successfully rolled out. In fact, continuous monitoring is most important once the service goes live and is subjected to real world, highly changeable traffic loads and network conditions. Around-the-clock testing and monitoring of network performance enables administrators to pinpoint trouble areas rapidly to minimize or avoid impact on service quality. Continuous testing also provides the data to perform meaningful trend analyses that enable better capacity planning and network utilization decisions.

Comprehensive Measurement

To provide that meaningful data, network measurement must be comprehensive, employing both passive monitoring and active testing.

Active testing should simulate actual end-user usage patterns. This means executing the same high-level transactions, such as a call, using the same types of protocols, like SIP, H.323, MGCP, or SCCP, and involving the same call components, such as gateways or proxy servers, that users do. This testing should be flexible enough to allow precise control over every aspect of configuration, scheduling, and measurement. And it should provide the flexibility to test all aspects of the service from call initiation, media transmission, and delivery through call completion as a single transaction or by isolating specific portions. Most importantly, because active testing can be done independently of users, administrators can proactively uncover potential voice quality problems before users do.

However, it is also important to have the option to passively monitor the quality of a particular call to identify user, location, or time-specific complaints. For example, for a poorly performing on-net call between a user in the New York-based headquarters and a colleague in the Paris sales office, the network administrator must be able to identify whether the impairments can be attributed to performance issues in the WAN provided by the service provider, the LAN in New York, or the LAN in Paris in order to fix the problem. By measuring a call�s performance, passive testing provides insight on the user experience -- as it happens.

A PER-CALL QUALITY METRIC

While measurement of packet loss and latency is required to diagnose performance problems, the bottom line for network administrators is whether they�re providing toll-quality service to their users. How do you achieve this single measure of user satisfaction or a MOS (Mean Opinion Score) without relying on the subjective opinions of human testers? Today�s most sophisticated measurement suites deliver an objective measure of voice quality based on the ITU-T G.107 (E-Model) standard.

PUTTING THE QUALITY IN QoS

As cost pressures continue to mount on enterprises, and legacy equipment continues to age, more and more organizations will embrace the advantages of VoIP. As they do so, however, enterprise IT managers must recognize the importance of adopting a VoIP measurement strategy that is pervasive, continuous, and comprehensive. Not doing so leaves mission-critical voice applications at risk for crippling service degradations and outages -- and carries with it its own steep price tag.

Planned and implemented correctly, VoIP network measurement can:

� Provide a precise picture of call quality for each and every call, using the same criteria that determines user satisfaction;

� Pinpoint trouble areas, in time to prevent service interruption;

� Enable enterprises to evaluate service provider responsibility for performance issues; and

� Provide the information needed to plan network and service enhancements effectively.

Using a sophisticated and flexible measurement approach gives enterprises what they really need: insight into the myriad network factors that impact voice quality, and finally giving them control over the user experience.

Can you hear me now?

Laura Holly is market development manager at Chelmsford, Massachusetts-based Brix Networks, a provider of service assurance and performance management solutions for advanced IP services, including VoIP, VPNs, Web-hosted applications, and streaming media. For more information visit the company online at www.brixnet.com.

[ Return To The April 2003 Table Of Contents ]