March 2017 Summaries

Rebuffering is a significant issue that negatively impacts viewer experience during video playback. It occurs when there isn't enough bandwidth for the video to load, leading to interruptions in playback as the video has to "catch up." In the era of Adaptive Streaming, rebuffering can be attributed to various factors such as Internet Service Providers (ISPs), Content Delivery Networks (CDNs), encoding processes, and players. Mux provides four different rebuffering metrics: Rebuffering Percentage, Rebuffering Frequency, Rebuffering Duration, and Rebuffering Count. These metrics help publishers identify trends in video delivery problems and investigate possible causes of rebuffering issues. By analyzing these metrics, publishers can determine whether the issue lies with ISPs, CDNs, encoding processes, or players, and take appropriate measures to improve viewer experience. Rebuffering is a crucial factor that affects user engagement, as studies show that up to 40% of viewers abandon a video after just one rebuffering event. To tackle this issue effectively, publishers need access to relevant data and should work towards eliminating rebuffering for better viewer experience.

The text discusses the factors that contribute to good video quality and how it can be measured. Video quality is determined by three main components: bitrate, codec, and source quality (including resolution). Bitrate refers to the optimal encoding bitrate for any given video, while codec pertains to H.264 or other emerging technologies like VP9, H.265, and AV1. Source quality involves the original file's encoding and potential issues such as artifacts, interlacing problems, color issues, or over-exposure. Content quality is a broader concept that includes video quality, audio quality, and subtitle/closed caption quality. Audio quality is crucial to good video content, as viewers can tolerate poor visuals if the audio remains clear. Poor internet connection and conditions of actual wiring are common problems for delivering video online. Adaptive streaming technology helps provide a comparable viewing experience with less buffering by offering several different videos at varying resolutions and bitrates in parallel. Upscaling and downscaling can impact video quality, where upscaling refers to stretching a video beyond its original size, while downscaling involves compressing it into a smaller size. Mux measures the percentage of upscaling and downscaling that occurs during video playback. By understanding these metrics, content creators can optimize their video delivery pipeline for specific devices and screen resolutions to ensure high-quality video experiences.

Mux will be participating in the Flink Forward SF conference on April 11, 2017 at Hotel Kabuki in San Francisco. Scott Kidder from Mux will deliver a talk about their real-time anomaly detection system powered by Apache Flink. They are open to discussing stream processing and invite attendees to reach out via Twitter or email ([email protected]) for more information or to schedule a meeting.

Video Startup Time is a crucial metric in video analytics that measures the time it takes for a video to start playing after a viewer's intent to watch. It can be influenced by factors such as page load times, pre-roll ads, and pre-loading of videos. Mux offers solutions like splitting metrics into Video Startup Time and Aggregate Startup Time, providing detailed tracking, and offering a comprehensive overview of the viewer experience. Accurate measurement of this metric is essential for understanding audience engagement and optimizing video performance.

Stream processing systems like Apache Flink can easily span a large number of hosts, but managing stateful distributed applications presents challenges in performing administrative tasks without causing downtime. Mux solved this problem by introducing a control stream to their Flink application. This allows for monitoring and inspecting the inner workings of the streaming application, which is crucial for maintaining confidence in its correctness and preventing regression defects. By using RabbitMQ as a new stream-source for control messages, Mux was able to easily add a control stream that has performed well and enabled offline analysis and validation of machine-learning models.