Agora Tutorial丨A typical case, teach you how to use crystal ball "data insight"

At the beginning of July, the official version of the "data insight" function of Agora Crystal Ball was launched. "Data Insight" can display two kinds of data, one is usage and the other is quality.

The greatest significance of the "Usage Overview" of "Data Insights" is to help you look back on the overall trend of audio and video minutes in the past period of time, and provide data reference for the review and adjustment of business operation strategies. These data charts are relatively easy to understand.

However, users who saw the “Quality Overview” in “Data Insights” for the first time may feel only the multiple data icons rushing to their faces. Go back to the source step by step. So we will use a test case to step by step from discovering the problem, investigating clues, and discovering the basis of the problem, so that everyone understands how to use it.

*Note: The following figure shows the test Demo data

By clicking "Quality Overview" in the left menu bar, we can view the quality trend based on the time range on the "Data Insights" page.

First of all, the page can be divided into three modules "User Experience", "Join Channel", and "Metric Analysis". "User Experience" and "Join Channel" show five quality indicators that mainly affect the real-time interactive experience: video freeze rate, audio freeze rate, network delay rate, login into Euro power and 5s login success rate.

"Indicator Analysis" shows the data distribution of the above five quality indicators in the dimensions of region, operating system, network type, device type, SDK version, and channel scale.

We chose to view the data from June 26th to July 4th. The "User Experience" module will display the trend of three quality indicators: video freeze rate, audio freeze rate, and network delay rate during this period. Moreover, it will automatically filter out the data of the worst day and display it on the top of the graph. As shown in the figure below, at a glance, the worst days of the three indicators coincided with the day of July 1. What happened on this day?

We can put the mouse on the data curve on July 1st. Click on the data point and you will see two options in the pop-up bubble: "View hours" and "View distribution".

In order to further investigate what went wrong on this day, we clicked "View Hour" to view the quality data of the day from a finer granularity. As shown in the figure below, we see that the worst experience is 21:00.

Next, we click on the 21:00 data point, and then enter the "channel data sampling". Of course, only data points that meet the "sampling rule" will display "channel data sampling". For detailed rules, please search for "sampling rule" .

After clicking "channel data sampling", sampling details will pop up on the right. A scatter plot of the "minutes-video freeze rate" data at that point in time will be listed here. Every point is a channel. The closer the data point is to the upper right corner, the higher the stall rate of this channel and the longer the call time, which means that the experience of this channel is relatively poor.

We can see from the figure below that the data points near the upper right corner are all the same channel.

At this time, if we click on any channel number, the number of users at the time of changing the channel (that is, the range of impact of the freeze), and the total duration of the video (that is, the duration of the impact of the freeze) will be displayed. At the same time, there will be a "user data sampling".

What happened on this channel?

Then, click "User Data Sampling", and the bottom of the window will show who are the users in this channel at that time. As you can see from the figure below, although the users experiencing quality experience problems are different, the opposite users are all the same (the red box in the figure). Explain that the experience problem may be related to the peer user.

When we click on the "call survey" on the right, we will jump to the crystal ball's "call survey" function to inquire about the quality data of the channel at that moment in detail.

After entering the call investigation, we can see the device status of the sender and receiver, video sending resolution, video sending frame rate, video frame rate and freeze, video uplink and network packet loss, video downlink and network packet loss.

From the data point of view, there is no abnormality in the network status of the sender, but the CPU is abnormal from 18:00 to 21:00 (the red data value indicates that the CPU usage is too high). Looking at the resolution of the video transmission, when the CPU usage is high, the resolution of the transmission is also reduced. However, the Wi-Fi signal quality is blue, which means that the network environment is good.

Therefore, it can be preliminarily judged that the insufficient performance of the device at the sending end has caused the video freeze during this period.

↓↓↓ Trailer

The multiple quality dimensions of "call survey" can be used to investigate the root cause of call problems. We will later explain the use of "call investigation" around two typical cases. Stay tuned.