Bufferbloat and the pandemic was: V6 still not supported

Thu Mar 24 14:27:48 UTC 2022

>  Given the tremendous growth of video conferencing  which strains the
    upstream, I wonder how many calls ISP's are getting because the
    "internet is slow" which is attributable to bufferbloat. Is there really
    anything that ISP can do if they don't supply the ÇPE? What percentage
    of providers do supply the CPE in the form of cable and dsl modems, etc,
    that they could solve the problem with a swap out?

In my experience this is not really a problem of lack of bandwidth (not to say that this is not important) but of queue behavior (the issue is at root one of 'working latency'). So you can solve this for example with AQM. But solving it on the CPE only moves the bottleneck to the LAN/WLAN (in which case use distributed APs, optimally with Ethernet backhaul to the CPE). Next on the horizon is dual queue - which is in discussion at the IETF (TSVWG).

Check out the paper at https://www.netforecast.com/wp-content/uploads/NFR5137-Videoconferencing_Internet_Requirements.pdf and see Figure 8. This suggests the network (WLAN to server) has a budget of 130-280 ms of delay (latency), depending on the video conferencing app. See also my paper about AQM deployment at https://arxiv.org/abs/2107.13968 and the recent BITAG paper on the subject at https://www.bitag.org/documents/BITAG_latency_explained.pdf.

Also, if you have Mac OS check out the cool new "responsiveness" tool from Apple: https://mailarchive.ietf.org/arch/msg/113attendees/gfvFljIMgsmCTUUPs9TMeBA2wFU/

Jason