I have tried a number of things. I’ve increased the downstream SNRM on each line from 3dB to 6dB and the line 1 upstream to 9dB, others 6dB. Not that that will last long!
That’s a just in case, as the error rates were not that high before. I’ve also lowered the AA downstream tx rates to 95% of flat-out (were at 98%). Trying to fix the latency but I don’t know why that would work if indeed it does turn out to be a successful measure.
On the CQM graph shown above, is there a
max-rate upstream tx at 11:00 ? (red-brown trace, labelled ‘Rx’) I’m wondering if that’s why everything went bad, because of simultaneous downstream and upstream, or because of downstream and simultaneous upstream @100%. If so, then I’m not testing it properly, not with a sufficient load.
CQM graph from a bidirectional test that aims to give a much more testing scenario the packet and a (more than) realistic load. The testmy.net upload 100 MB test was run while simultaneously performing several 100MB downloads on another machine. This is a worse test than the original Zoom case because with Zoom the download is not at 100%.
It does show the packet loss on line 1 (only) just like the Zoom scenario. So I have successfully reproduced it! I’m assuming that line 1 exhibits the problem because it has the slowest downstream sync speed so is the limiting factor for TCP in this test. But I can’t see how that’s relevant for Zoom, seeing as we presume there’s no TCP and it isn’t flat out downstream, so it won’t know anything about line 1 being any performance-limiting link.
Speeds now, after the changes are as follows. SNRM 6 dB / 6 dB, apart from line 1 which is now 9 dB upstream. When line 1 was at 6dB upstream, its sync rate was an amazing 200k higher ! Live sync rates:
#1: downstream 2378 kbps, upstream 566 kbps
#2: downstream 2503 kbps, upstream 525 kbps
#4: downstream 2719 kbps, upstream 563 kbps
[Moderator edited to merge four successive posts into one.]