ApsaraVideo VOD:Data metric descriptions

Definition and interpretation:

Fast Startup Rate (TTFF ≤ 1s) is the percentage of actual plays where the first frame is successfully rendered within 1 second.

Formula:

Fast startup rate = (Number of plays with TTFF ≤ 1,000 ms / Actual plays) × 100%.

Scope and rules:

• This metric excludes cases where the user exits immediately or playback fails after the start() method is called, resulting in no first frame.

• For audio-only playback, the time to first frame is based on when the first audio frame is rendered.

Data source and unit:

This metric is expressed as a percentage.

Definition and interpretation:

Slow Startup Rate (TTFF ≥ 3s) is the percentage of actual plays where the time to render the first video frame is 3 seconds or more. As the counterpart to the fast startup rate, this metric reflects slow startup situations.

Formula:

Slow startup rate = (Number of plays with TTFF ≥ 3,000 ms / Actual plays) × 100%.

Scope and rules:

• This metric excludes cases where the user exits immediately or playback fails after the start() method is called, resulting in no first frame.

• For audio-only playback, the time to first frame is based on when the first audio frame is rendered.

Data source and unit:

This metric is expressed as a percentage.

Definition and interpretation:

Time to First Frame (TTFF) is the time from when the start method is called until the first frame is displayed. It is a key metric for measuring video startup speed and user experience.

Formula:

TTFF = AVG(First frame display timestamp - start call timestamp).

Scope and rules:

• This metric excludes cases where the user exits immediately or playback fails after the start() method is called, resulting in no first frame.

• For audio-only playback, the time to first frame is based on when the first audio frame is rendered.

Data source and unit:

This metric is calculated from the timestamps of events reported by the client-side SDK, in milliseconds (ms).

Definition and interpretation:

Average Seek Duration is the average video loading time after a user drags the progress bar. It reflects seeking smoothness.

Formula:

Average seek duration = Total loading time from dragging the progress bar / Number of seek operations.

Scope and rules:

• This metric includes the time for the video buffer to recover after a seek operation.

Data source and unit:

The average value is calculated in milliseconds (ms).

Definition and interpretation:

Play Failure Ratio is the percentage of playback events that are terminated due to an unrecoverable error. The failure rate reflects the overall reliability of playback. A lower value indicates higher stability. This metric is affected by the player SDK, user device stability, network conditions, and CDN stability.

Formula:

Play failure ratio = (Number of failed plays / Total plays) × 100%.

Scope and rules:

• This metric includes abnormal termination scenarios such as device errors, decoding failures, and stream interruptions.

• This metric excludes cases of brief stuttering or successful playback after an automatic retry.

Data source and unit:

This metric is based on the number of abnormal status reports from the player SDK. It is expressed as a percentage.

Definition and interpretation:

Non-Start Ratio is the percentage of playback attempts where the user exits before the first frame is displayed or the startup fails. This metric reflects the user experience during the initial playback phase.

Formula:

Non-start ratio = (Number of non-started plays / Total plays) × 100%.

Scope and rules:

• Startup can also fail for business reasons, such as abnormal playback parameters.

• This metric includes two scenarios: user-initiated abandonment and failure due to network abnormalities.

Data source and unit:

This metric is based on statistics of missing first-frame callback events plus exit events. It is expressed as a percentage.

Definition and interpretation:

Error Count per 100s is the average number of unrecoverable errors that terminate playback per 100 seconds of video watched. It measures the frequency of unrecoverable errors during playback.

Formula:

Error count per 100s = (Number of errors / Total watch time in seconds) × 100.

Scope and rules:

• This metric includes abnormal termination scenarios such as device errors, decoding failures, and stream interruptions.

• This metric excludes cases of brief stuttering or successful playback after an automatic retry.

Data source and unit:

The unit is 'count'. The number of unrecoverable errors per 100 seconds is calculated.

Definition and interpretation:

Rebuffer Count Rate is the percentage of playback sessions that pause due to network latency or an insufficient buffer. It does not include buffering caused by the user actively dragging the progress bar. This metric reflects video playback smoothness.

Formula:

Rebuffer count rate = (Number of plays with rebuffering / Actual plays) × 100%.

Scope and rules:

• Number of plays with rebuffering: The number of plays where the rebuffer count is ≥ 1.

• This metric counts playback records that have experienced network waiting.

• This metric does not count buffering caused by seek operations.

Data source and unit:

This metric is based on the player's buffer event count. It is expressed as a percentage.

Definition and interpretation:

Rebuffer Duration per 100s is the average total network rebuffering duration for every 100 seconds of video played. It reflects the severity of rebuffering during the user's viewing process. It excludes rebuffering before the first frame loads and rebuffering caused by seek operations. This metric can be used to monitor network quality and playback smoothness. A lower value indicates a better experience.

Formula:

Rebuffer duration per 100s = (Total rebuffer duration in seconds / Total watch time in seconds) × 100.

Scope and rules:

• This metric only counts the rebuffering duration caused by network buffering during user viewing.

• This metric does not include the player's startup buffer time or the buffer time after the user drags the progress bar.

Data source and unit:

The unit is seconds (s). Both playback duration and rebuffer duration are accurately collected by the client.

Definition and interpretation:

Rebuffer Count per 100s is the average number of network rebuffering events that occur for every 100 seconds of video played. It measures the frequency of rebuffering events during playback.

Formula:

Rebuffer count per 100s = (Number of rebuffering events / Total watch time in seconds) × 100.

Scope and rules:

• This metric directly reflects the frequency of rebuffering.

• This metric does not include player startup buffering or buffering after the user drags the progress bar.

Data source and unit:

The unit is 'count'. The number of rebuffering events per 100 seconds is calculated.

Definition and interpretation:

Total plays refers to the total number of valid video playback sessions initiated by users on the client. A play is counted when the client first calls start after prepare is called. Repeated calls to start after a pause do not count as new plays.

Significance:

This metric reflects the activity of video playback and is a basic metric for measuring user usage frequency.

Scope and rules:

• This metric only counts the number of times the client confirms receipt of the initial start call.

• This metric does not count replaying the same stream after a pause.

Data source and unit:

This metric is counted by player SDK event triggers. It is a unitless count value.

Definition and interpretation:

Actual Plays is the number of plays where the first frame is successfully rendered. It excludes plays where start is called but a successful first-frame rendering callback is not received.

Significance:

This metric measures whether a video was successfully played. It is crucial for reflecting a smooth user viewing experience.

Scope and rules:

• This metric excludes the following scenarios:

  • The user calls play, but playback cannot start due to network or device issues.

  • The video cannot start playing due to errors in the video source's encoding or decoding, or video authentication issues.

Data source and unit:

This metric is counted based on the number of first-frame rendering callbacks from the player SDK.

Definition and interpretation:

Completion Count is the number of plays where the last frame (end frame) of the audio or video is successfully displayed. A loop playback is counted as one completion.

Formula:

Completion count = Total number of plays that successfully displayed the end frame (finish_cnt).

Significance:

Completion count measures the number of times users watch a video completely. It is an important metric for evaluating user stickiness and video appeal.

Scope and rules:

• All of the following scenarios are counted as completions:

  • A user seeks directly to the end of the video and displays the end frame.

  • A user seeks to any position and then continues to play normally to the end frame.

  • During loop playback, the user exits after the end frame is displayed.

• Loop playback is not counted repeatedly.

Data source and unit:

This metric is counted by the player SDK through the playback completion event. It is a unitless count value.

Definition and interpretation:

Completion Rate is the percentage of completed plays out of the total number of plays. It is an important measure of user viewing completion.

Formula:

Completion rate = (Completion count / Actual plays) × 100%.

Scope and rules:

• This metric uses the display of the end frame at the end of playback as the completion marker.

• Loop playback is counted only once.

Data source and unit:

This metric is expressed as a percentage.

Definition and interpretation:

5-Second Bounce Ratio is the percentage of actual plays where the user actively exits or stops playback within the first 5 seconds.

Formula:

5s bounce ratio = (Total number of plays with a duration of less than 5 seconds / Actual plays) × 100%.

Scope and rules:

• Even if a user seeks to a position beyond the 5-second mark, if the playback time is less than 5 seconds, it is still included in the statistics.

Data source and unit:

This metric is expressed as a percentage.

Definition and interpretation:

Unique Users is the number of deduplicated users who have played videos using the player SDK.

Formula:

Unique users = count(distinct(uuid)).

Significance:

This metric measures the actual number of active devices covered by the player, reflecting the user base scale.

Scope and rules:

• Devices are deduplicated based on the unique identifier uuid. The uuid is generated on the client side based on an algorithm. Uninstalling and reinstalling may cause the uuid to change, leading to overcounting of users. For the ApsaraVideo Player SDK for web, clearing the browser cache or switching browsers will cause the uuid to change.

• Only users who have successfully started playback are counted.

Data source and unit:

This metric is based on SDK device ID statistics. It is a count value.

Definition and interpretation:

Average Play Duration per User is the average effective playback duration accumulated from startup to exit for a user. It reflects the user's viewing depth of video content. Higher quality video content will increase the average play duration. This metric can guide video content optimization.

Formula:

Average play duration per user = Total actual user playback duration / Number of unique users.

This is equivalent to: Average play duration per user = sum(watch_duration) / count(distinct case when watch_cnt > 0 then uuid end).

Scope and rules:

• Total actual user playback duration excludes the following times:

  • Pause time

  • Seek time generated by dragging the progress bar

  • Network rebuffering time

  • Error retry time

  • Other progress bar pause times

• Loop playback duration is included in the statistics.

• For variable-speed playback, the duration is based on the actual watch time. For example, for a 10s video played at 2x speed, the recorded playback duration is 5s.

Data source and unit:

This metric is based on player SDK event timing statistics. The minimum unit is milliseconds.

Definition and interpretation:

Average Plays per User is the average number of successful plays per user. It reflects user activity and usage frequency. This metric can differ significantly between long-video and short-video services.

Formula:

Average plays per user = Actual plays / Unique users.

Data source and unit:

This metric is calculated by dividing the actual plays count by the number of unique users. It is a unitless count.

Definition and interpretation:

Average Completion per User is the average number of times a user watches a video completely. It measures the completeness and interest of user video viewing.

Formula:

Average completion per user = Completion count / Unique users.

Data source and unit:

This metric is calculated by dividing the completion count by the number of unique users. The result is an average count statistic.

Definition and interpretation:

Average Play Duration is the effective watch time of a single play. It measures the viewing depth of a single playback session.

Formula:

Average play duration = Total actual user playback duration / Actual plays.

Scope and rules:

• Total actual user playback duration excludes the following times:

  • Pause time

  • Seek time generated by dragging the progress bar

  • Network rebuffering time

  • Error retry time

  • Other progress bar pause times

• Loop playback duration is included in the statistics.

• For variable-speed playback, the duration is based on the actual watch time. For example, for a 10s video played at 2x speed, the recorded playback duration is 5s.

Data source and unit:

The minimum unit is milliseconds (ms).

Definition and significance:

Average Video Duration is the average video length based on statistics from the video resource header information. It reflects the duration distribution of the played content.

Formula:

Average video duration = AVG(play_video_duration).

Data source and unit:

The video duration recorded in the video header is reported by the player SDK. The minimum unit is milliseconds.

Definition and interpretation:

Average Bitrate per Start is the video start bitrate reported by the client. It reflects the video quality level at which the user starts playback.

Formula:

Average bitrate per start = AVG(start_bitrate).

Scope and rules:

• This metric includes bitrate data from all completed and uncompleted video streams.

Data source and unit:

The minimum unit is bps.

Definition and interpretation:

Average Bitrate per Play is the average video bitrate reported by the client. It reflects the video quality level of the user's actual playback. For multi-bitrate videos, users can actively select different definition videos during playback. If the player is in auto mode, the player's Adaptive Bitrate (ABR) algorithm automatically selects the appropriate bitrate based on the current network conditions. Different ABR algorithm policies affect the final average playback bitrate.

Formula:

Average bitrate per play = AVG(play_bitrate).

Scope and rules:

• This metric includes bitrate data from all completed and uncompleted video streams.

Data source and unit:

The minimum unit is bps.