HTTP adaptive streaming and access router management: The users' and network's perspectives

The share of video streaming in the Internet traffic is expected to reach 80% by 2019. To deliver these video services in possibly strained settings and to meet with the users requirement of anywhere and multi-screen experience, HTTP Adaptive Streaming (HAS) is gaining momentum. However, TCP-transport of HAS flows is known to entail a number of issues both on Quality of Experience (QoE) and network metrics. Thanks to testbed experiments, we examine the video streaming performance in a home access scenario by exposing the interplay between the video rate adaptation logic (Buffer-Based-BBA, or Rate-Based-RBA) and the buffer sizes and policies (Droptail, Active Queue Managements - AQMs). A main finding is that AQMs move the bulk of TCP losses from the initial burst and ACK-trailing phases, problematic to HAS flows, to the more favorable ACK-clocking phase. We thereby show that implementing AQM CoDel at the access router is a simpler means, working for any family of client's adaptation, to improve QoE and network metrics than striving to smooth out the HAS bursts by pacing at the client or middleboxes, as previous approaches do.