How Cisco Keeps Quality High For Voice Calls, Part 2: Metrics
In Part 1 of this post, I described how Cisco IT addresses the first key question—about reporting on voice service availability. In this Part 2, we’ll cover the second question: How does the call sound to all of the connected parties?
Cisco IT Metrics for Measuring Call Quality
Although it seems counter-intuitive, the best source of information about voice quality may not be the people who were on the call. Of course, user trouble tickets about problems such as static and echo can be important indicators of bigger issues in a voice system. But we often find that users don’t report voice quality issues, so additional tools are needed.
For routine operations, Cisco IT uses more detailed and objective tools to measure call quality, including:
- The Mean Opinion Score (MOS) calculated by the Cisco UCM for each call, based on data reported by the connected Cisco IP phone(s) and / or voice gateways.
- Data collected by the Cisco Unified Operations Manager tools from sensors we have installed in front of key aggregation points, such as Cisco Unity voicemail systems and Cisco WebEx conferencing systems. These sensors collect the audio stream in a call for analysis, then generate call quality data.
- Network statistics on jitter, packet loss, and delay, all of which can negatively impact the call experience.
We track against industry standard quality level expectation based on the codec in use at each site. For example, e.g., a G.729 codec allows more compression and lower bandwidth utilization for voice traffic, so we expect a lower definition of call quality than a higher fidelity codec such as G711.
But with Cisco’s 540+ buildings worldwide, local issues still come up, even if overall voice quality is at extremely high levels at Cisco. For example, when a region or local office has lower voice quality than we expect, we look for a local source such as a WAN circuit that is producing packet loss or that has an out-of-date QoS configuration.
In these two posts, I’ve given a broad view of how Cisco IT monitors quality levels for voice calls. What kinds of techniques have you found to be effective?