Looking at my latest graph, and what you're saying, I'm surprised DLM intervened. I had some very minor packet loss, but then ran for ~8 hours without any lost packets. I've had no where near as many lost packets as your graphs.
Don't use my graphs as a way of showing the *minimum* effect needed for DLM to intervene. Those effects were really pretty bad.
Here is the latest graph, would you have expected DLM to intervene here?
You can't tell enough from the graphs to decide whether DLM should intervene: The ping graphs can only show packet loss when it gets as high as 1% (1 in 100), while DLM is probably monitoring for failure rates more like 1 in 5,000 (for DLM to increase intervention) or 1 in 100,000 (to decrease intervention).
The ping monitor only sends 1 small packet every second, while your connection is capable of passing hundreds of CRC-checked blocks every second. DLM monitors all of those blocks, even when they don't contain end-user data - so it knows the state of your line hundreds of times better than can be seen by a once-per-second ping packet.
Here are some example numbers...
Plusnet has a page describing the line error-checking procedures that go on; follow that link, and looks for a section on "Indicative line quality".
NB 1: The principle probably applies to all kinds of broadband, but it isn't clear that those exact numbers apply to DLM for FTTC. Openreach DLM is different from BT-Wholesale's.
NB 2: ISPs can order different DLM thresholds to be applied, such as "standard", "stable" and "super-stable". A "super-stable" line will allow fewer errors than a "standard" one. The figures in the link probably apply to one of these settings, but not to all three.
You can see the thinking on that page: an MTBE of less than 10 seconds would trigger DLM to intervene. As there are 86,400 seconds in a day, an ES value higher than 8,640 would be enough to trigger that. At worst, that ES level could be caused by as little as 8,640 CRC errors in 8,640 separate seconds.
My modem statistics suggest that it transfers CRC blocks at a rate of 604 per second, which amounts to 52 million per day. In the worst case, if 8,640 of these were faulty, DLM would intervene - which makes for an error rate of 1 in 6000.
Thankfully, while I do get around 8000 CRC's per day, I only get around 700 ES's per day. Errors do seem to come concentrated.
Looking back at the Plusnet figures, once DLM has intervened, the MTBE has to increase to over 420 seconds in order for DLM to de-intervene. That would be an ES value lower than 205 in a day. In the worst case, that would require an error rate of around 1 in 250,000 or better.
The ping graphs only give a hint that something might be wrong.
Getting hold of the full line statistics are needed to start to get to the bottom of things, such as from an unlocked HG612. And we, joe public, are still only guessing at what really goes on.
Pages in this thread: 
Print Thread
