Troubleshoot Zoom Sessions Using Shapley Feature Ranking

You can use Shapley feature ranking to troubleshoot Zoom sessions. Shapley feature ranking is a technique that determines the average marginal contribution of each feature to the user experience. It takes into account all possible permutations of features to generate a Shapley value that indicates a feature’s contribution.

Juniper Mist™ uses the Shapley feature ranking technique to rank the impact of each feature on Zoom calls. Here, features indicate network characteristics such as received signal strength indicator (RSSI), latency, jitter, and so on. The Shapley feature ranking is computed for every minute of bad audio or video that a user experiences. This ranking helps you to quickly identify the WAN, wireless, and client issues that impact Zoom calls. You can read more about Shapley feature ranking computation for user experience here.

You can see information about Shapley feature ranking from the Client Insights page, which lists details of all the meetings that a user joined over a certain period of time. See View Client Insights for information about how you can access and view the Client Insights page. The Client Insights page provides:

• Meeting Insights—This section provides time series graphs depicting RSSI, latency, packet loss, and jitter.

  

• Meeting Details

  

This section provides a Shapley feature ranking graph for each meeting which reports bad user experiences. In this example, the last two rows include the caret symbol (^), which indicates that the meeting has a Shapley feature ranking graph associated with it.

Click ^ to view the Shapley feature ranking graph. Here is an example that shows feature ranking for audio latency. The X axis indicates latency in milliseconds (ms). The Y axis shows the average latency that the wireless, client, and WAN features contribute. The graph provides:

• Observed average latency for all Zoom calls associated with a site.

• Aggregate latency attributed to each feature category (wireless, client, and WAN).

This graphical representation enables you to quickly identify the network segment that needs attention.

If you observe several minutes of bad user experience in a Zoom meeting, you can select the date and time of that meeting to view the associated feature ranking computation. You select the date and time from the drop-down list above the graph.

Here are two examples that show how you can use Shapley feature ranking to troubleshoot user experience issues.

Example 1:

In this example, you can see that the average latency for the site is 256.8 ms. The dominant feature category is WAN, which increases the latency by 187.7 ms.

You've now identified the feature category that contributes the most to the increase in latency. You can drill down further—click the drop-down list in the WAN section—to see the specific WAN feature that causes the maximum latency. In the following expanded view of the graph, you can see that among the WAN features, Site WAN download capacity contributes to the longest period of latency.

Example 2:

In this example, you can see that the highest contributor to latency is the client. Within this feature category, the dominant feature is VPN. This scenario occurs when a client is connected through a full tunnel VPN instead of local egress.