[Mcast-wifi] Video quality with GroupCast with Retries (GCR): some results

[PABLO SERRANO YAÑEZ-MINGOT <[e-mail]>]

Hi all,

We've been experimenting for some time with the delivery of multicast
video, comparing the performance of the legacy 802.11 behaviour and
the novel GCR mechanisms, namely, UR and BA (in fact, the draft
[Perkins2018] cites a work we did some years ago, Oliva2013)

Instead of objective metrics such as jitter or packet loss, we've
estimated the "Quality of Experience" (QoE) using Netflix's VMAF tool
[VMAF]. While the scope of the study is of course limited to our
scenario (academic environment, 802.11g, 25 receivers), some take-away
results that could be of interest to the draft are (full paper
[Gringoli2018]):

* A loss of 5% (as mentioned in Section 3.1.1 of the draft) is way too
high for a watchable video experience. According to our results, even
with losses around 1% or less the QoE could be pretty bad.

* Unsolicited Retries (mentioned in Section 4.7) does improve a bit
performance, but fails to provide an adequate QoE. For video delivery
to multiple receivers, better to stick to BA.

* The use of the default EDCA differentiation slightly improves
performance, but if the video quality was already bad, it won't result
in a watchable experience.

Rgds, Pbl

References:

[Perkins2018] https://datatracker.ietf.org/doc/draft-ietf-mboned-ieee802-mcast-problems/

[Gringoli2018] http://doi.org/10.1109/TMC.2018.2876000

[VMAF] https://github.com/Netflix/vmaf