Buffer Sizing and Video QoE Measurements at Netflix
In this talk we will present some highlights from recent research results on how the sizing of router buffers affects Netflix video traffic. Our results were publi …
|Talk Title||Buffer Sizing and Video QoE Measurements at Netflix|
|Speakers||Bruce Spang (Stanford University)|
|Location||San Francisco, CA|
|Date||Feb 10 2020 - Feb 12 2020|
In this talk we will present some highlights from recent research results on how the sizing of router buffers affects Netflix video traffic. Our results were published at the recent Workshop on Buffer Sizing at Stanford (link to paper: http://buffer-workshop.stanford.edu/papers/paper12.pdf).
In certain locations, Netflix streams video over TCP New Reno from racks of servers that are directly connected to large routers, which in turn directly peer with commercial ISPs. We varied the size of the router buffers during periods of persistent congestion, and logged metrics such as the number of rebuffering events, video quality, and video play delay. We observed buffers that are too small and too large, both of which worsen video QoE.
Our main takeaways are:1. The effects of router buffers on TCP New Reno matches our intuition: packet loss increases and RTT decreases as buffers shrink.2. Video performance has a sweet spot in terms of buffer size: buffers can be both too small and too large, and in both cases increase the number of rebuffers and decrease the video quality.3. We learned some surprising things about the buffer architecture and scheduling policy in our VOQ-based chassis routers, and will discuss how this scheduling complicates–and potentially masks–our clear understanding of buffer sizing.
Bruce Spang: Bruce is a PhD student at Stanford, advised by Nick McKeown. His research is on a combination of internet networking and theoretical computer science