This morning's tests, conducted all back-to-back included the definite all-time highest downstream speedof.me figure, at 7.77 Mbps
Time: 2017-01-21 05:56
down / up Mbps
7.72 / 0.94
7.77 / 0.99
7.74 / 1.06
7.67 / 1.00
7.72 / 0.96
This is with lines 3 and 4 set to a downstream target SNRM of 6dB now, and line 1 is on a 3dB target. Upstream target SNRM is 6dB for each.
Despite this the actual downstream SNRM figures are currently really low, lower than 3dB even for those on a 6dB target, and it seems possible that the SNRM drifts down fairly rapidly after initialisation time. Changing the target from 3dB up to 6dB doesn't really raise the medium-/long-term SNRM that much, it keeps it a bit above the really low 0.6-0.8dB region.
I have been using 3dB downstream target SNRM on all lines off and on for a year, occasionally going up to 6dB temporarily if there was suspected trouble. Until the autumn, I would say that the 3dB target was ok-ish, although it didn't provide much of a performance improvement. Over the last few months, during which there have been suspected and proven copper faults on the lines, stability has been less satisfactory than during the first half of the year, with a modem resynching about every six days very roughly.
Now as things are, I am not at all convinced that there is any performance benefit to be had from the use of 3dB for me. This could either be because
(i) too many uncorrected errors hurt TCP performance, or
(ii) the extra level of error correction overhead required is sapping throughput (seems less likely), or simply because
(iii) the downstream targets don't matter and the actual SNRM long-term ends up being about the same regardless of what the original target value was.