I finally got a chance to look at this line properly on MDWS, after my original "at a glance" look.
Expectations
Broadly, the line shows an attainable of 73/22. Most other signs back this up:
- An attenuation of around 15dB, it is a shorter line than my old 350m one (16dB) that could "just" get 80/22 attainable.
- An Hlog that looks fine for the attenuation
- A QLN that shows more noise than my old 350m line, explaining why attainable is a little lower than 80/20.
Those properties look good ... noting like @kitz that Hlog and QLN graphs are merely snapshots of the resync, rather than live.
Symptoms
The upstream speed looks to have been banded, with the banding speed reduced at intervals: 8.5Mbps, 7.2Mbps, 6Mbps, 5Mbps.
Upstream FEC and CRC rates have been high, but have gradually reduced as each level of banding has been introduced. In this case, banding is successfully reducing the symptoms.
The upstream SNRM seems to spike downwards nastily.
Deeper Details
When looking at the SNRM for individual bands, I originally noticed that U1 seemed to be worst, with the biggest drops from 15dB to zero. It turns out - from looking at the bitloading - that this is almost certainly caused by bitswapping removing and adding one or two bits in the U1 band, causing it to toggle between "wholly unused" and "having one or two bits used". The SNRM for U1 becomes a red-herring, shifting the question over to: why is U1 so devoid of bits, such that a couple of bitswaps can take everything away?
One aspect to remember is that "upstream power backoff" has a tendency to reduce usage of U1, putting more bits into U2.
In this case, I think the banding (ie an artifical speed cap) has reduced the need for bits so much that, combined with upstream power-backoff, the modem no longer needs to allocate bits in U1, leaving things mostly to U2.
Today (Aug 24th), things don't look so bad. But I notice that the "bitswap" graph is show no bitswaps upstream. Probably not a coincidence, but hard to say how much is cause vs effect.
Ignoring U1, then...
Spikes are also present in the SNRM for U2 - and these ones seem to match the spikes in the combined SNRM graph (little surprise, once you realise U1 doesn't contribute much).
The spikes suggest that something is wrong with the physical line, but they also suggest that once banding has had an effect and U1 has gone AWOL, the spikes are very intermittent, and of short duration.
If the spikes were of a longer duration, cause a resync and stuck around, then we might hope to see signs of the degradation in an Hlog or QLN graph. Being such short duration means we're unlikely to see anything that helps identify the cause from the statistics pages. We'll just keep seeing symptoms - but with banding in place, we'll only see a reduced set of symptoms.
As @ns says, the next step is for an engineer ... the choice will be down to the voice or broadband one. Unless, of course, today's improvement is the result of an unreported fix.