New tool to analyze, improve live streaming services earns best paper

The study produced a new tool to analyze and correct performance issues in major streaming software and services.
Strimmer
A streamer broadcasts a game playthrough.

In the past 10 years, online live streaming video has exploded in prominence. Services like YouTube, Amazon’s Twitch, and built-in streaming on social media sites now fill a sizable portion of the cord-cutting void. Delivering this high volume of live video with a good user experience is a massive technical undertaking, and one that’s limited by highly varied connection speeds and technology on the user end.

A paper that sets out to strengthen design practices in these systems earned the Best Paper Award at the 2021 ACM Multimedia Systems Conference (MMSys’21). Titled “Livelyzer: Analyzing the First-Mile Ingest Performance of Live Video Streaming” by PhD student Xiao Zhu, Prof. Z. Morley Mao, and AT&T Labs, the study produced a new tool to analyze and correct performance issues in major streaming software and services.

Because internet streaming services often rely on things like WiFi and cellular data at some point in the broadcasting pipeline (either on the streamer’s end or the viewers’), they typically face bandwidth constraints and capacity limitations that vary over time. This variability and restrictive technology make the task of high-quality video delivery challenging.

There are two points in the live broadcasting process where these technical limitations can manifest – the so-called upstream path between the broadcaster and the video server, and the downstream distribution path between the server and the viewers. The bulk of research on improving the streaming video experience focuses on the latter path, while the path from the broadcaster has received little attention.

“Today there is little understanding of the state of the art in the design of this critical path,” the authors write.

This is why they developed Livelyzer, a general-purpose active measurement tool and black-box testing framework focused specifically on analyzing the broadcasting component of popular live streaming software and services. The researchers used their tool to study seven different combinations of popular broadcasting apps and network conditions, including third-party, browser-based, and mobile-based broadcasting apps streaming to commercial services including Facebook, YouTube, and Twitch. 

Among their key findings, they cover:

  • The different apps use very different encoding configurations, and many use a very inefficient one.
  • The different apps behave differently when network conditions change, and most of them sacrifice video quality inefficiently as a side effect of adapting.
  • Use of a more adaptation strategy uncovered by Livelyzer can improve performance, as well as using a different encoding codec.

Beyond these findings, the researchers say that the tool will make controlled, repeatable experiments on broadcasting technology much more feasible going forward. In the future they intend to use Livelyzer to assess other types of network conditions like 5G, and explore the impacts of different streaming protocols.

Explore:
Honors and Awards; Research News; Zhuoqing Morley Mao