May 2019 Summaries

Mux has introduced Audience Adaptive Encoding (AAE), an evolution from their work on fast Per-Title Encoding. AAE optimizes encoding ladders based on audience consumption patterns and bandwidth, rather than relying on theoretical assumptions made by engineers. This approach ensures that video quality is maximized for each bitrate, while also catering to the specific viewing habits of users. Mux Data platform enables this optimization by providing insights into how users consume video content. The methodology involves using machine learning algorithms to generate convex hulls and select top renditions, followed by estimating bitrate-resolution pairs based on user information. Although Mux has temporarily paused development on audience-adaptive and per-title encoding due to resource allocation for other projects, they remain committed to leveraging machine learning for video encoding solutions in the future.

Transport Layer Security (TLS) is crucial for secure internet transactions. Mux developed an open-source certificate-expiry-monitor tool that uses Kubernetes API to discover servers using TLS certificates and emits Prometheus metrics with expiration times for installed certificates on each server. This helps in alerting when certificates are not being renewed automatically, preventing service unavailability due to expired certificates. The certificate-expiry-monitor tool was built in Go and can be configured with key options such as polling interval, Kubernetes namespace, labels, and domains to monitor. It generates Prometheus metrics for each pod + domain combination, indicating the time-to-expiry, time-since-issued, and certificate status. The tool has been integrated into Mux's infrastructure with a Grafana dashboard for monitoring and Prometheus alerting rules for warning when certificates are nearing expiry or have expired.

Last month, the author attended LiveVideoStackCon 2019 in Shanghai to give a keynote session on designing a great video API. The event focused on new innovations and teaching video streaming architecture. Key highlights from the talks included machine learning applications, codec wars, commercial encoders for AV1, and discussions of alternate TCP stacks. China is becoming a leader in machine learning with numerous applications being demonstrated. Chinese companies are still pushing ahead with their own video codecs like AVS2, while commercial encoders for AV1 are starting to appear. Alternate TCP stacks are becoming common as companies experiment with improved throughput and reduced latency.

The company has released the first stable version of its Go SDK, named Mux Go. This lightweight wrapper simplifies the integration of video experiences powered by Mux into Go applications. It is built from OpenAPI definitions and enables easy content ingestion into a Mux account. Future SDKs in other languages are planned, with requests for specific languages welcome.

Mux simplifies adding video to applications or websites with a single API call. The media pipeline involves several stages of data analysis and transformation for each frame of audio or video. Packaging is an important part of this process, where the packager inserts timing and signaling information into the output of audio and video encoders. This results in a file or stream that can be consumed by a player, such as an mp4 file or HLS stream. TS (Transport Stream) format is commonly used for media streaming but has some drawbacks, including fixed packet size and additional features not necessary for internet-based streaming. Mux's transcoder uses techniques to minimize packaging overhead, resulting in more efficient video encoding and improved user experience.