Reseñas PISIS 2020

Posgrado en Ingeniería de Sistemas

Semblance

M. Sc. Logan Hanks

Twitter

Mr. Logan Hanks got a Master of Science in Computer Science from the Virginia Polytechnic Institute and State University. He has over ten years of experience in the software industry, having been employed in companies like Google and Reddit, among others. Currently, he is a staff software engineer for Twitter.

Introduction

The talk aims to know how the protocols employed for online video streaming works and the trade-offs of using them; particularly explains the protocol HLS, how it works, and its impact on interactive streams.

Summary

A video file is usually considered a container since it contains multiple information such as audio, video, or even text bundled into one or more programs. MSc. Logan discusses a very popular container format called MPEG Transport Stream, which was designed to transmit video over unreliable connections, for instance, broadcasting over the air for television or over the internet. Each video itself is composed of a series of frames with a lot of information packed in, thus making the video files extremely large and unable to render easily without compressing into smaller files. In encoding video files, large chunks of information are discarded because the final image does not need to simulate a replica very close to the original, whose difference will be imperceptible for humans. MSc. Logan presented three types of compression methods used, chroma subsampling, infra-frame coding, and iter-frame coding. In 2009, Apple produced HTTP live streaming (HLS), which is the most widely used method for video distribution. This tool divides the video into a series of small segments, all coordinated via a master playlist that recognizes and optimizes all sources capable of displaying, transmitting, and decoding the video files. The master playlist also lists all the available and variant streams and describes the bit rate, resolution, and encoding. For each variant stream, there is another playlist which is the listing of the segments, and for a live stream, this playlist updates all the time. The major downside is latency, which refers to the time between recording a frame and playing it back, an acceptable rate is 10 seconds, for an interactive video the best is less than one second. Segmentation delay is a problem almost inevitable related to latency since it takes time to download all until the full segment is recorded before one can even start to download it and play it back. Also, the playback device probably wants to have a buffer of segments that reads ahead, so that if there is any interruption in the network, it still has data that can be played back while it switches to a lower bit rate stream without interrupting the playback. Record segments before they get recorded is a strategy to deal with latency but it is more complex to deploy and maintain.

WebRTC is a very different approach more suitable for more interactive models; it can capture media streams from the device as well as sending and receiving media streams over the network. WebRTC uses UDP, which is a protocol for NAT transversal schemes that do not require all packets sent in order, and it can even skip some by using predicting frames. The initial structure of WebRTC was a mesh in which all clients are connected to the same network and exchange, download, and update video from each making the process burdensome.

Finally, he talks about Mixer, a structure that uses a controller unit that receives and broadcasts video from all the participants but requires expensive hardware to handle it.

Conclusion

In recent years, technology in the video format has evolved. Many technical inconveniences have been broken; compression and segmentation techniques have been the key to allowing higher speed in the transmission of videos on the internet. Currently, technology has sought to reduce latency in sending and receiving videos in real-time.