I'll have to check my calculations, and possibly account for SES too, but the day before DLM intervened, I think I had 2200 ES's. A week earlier I'd had 2100 but no intervention.

The calculation is a work-in-progress, though, so could easily be wrong.

My line was running for a long, long time averaging around 500-800 ES per day. That didn't trigger any DLM.

well averaging 1500 crc a day I was kept on fast path for ages but I Cannot rememebr what ES I was averaging.

This line I have 13 ES and only 19 CRC in about 3 hours uptime.

My CRC count tends to average around 10x the ES value. I guess my noise comes in distinct bursts.

In reply to a post by AndyHCZ:

2880 errored/serious errored seconds a day on the Openreach Speed (BT Wholesale 'Standard' in their lingo) DLM profile. But it will flag your line if more than 288 in a day.

Interesting. Since swapping from a Fritz!Box 7390 to a 7490, my line has sync'd for over 7 days. Attainable rate is up 20Mb and I've had 965 ES in that time. Hopefully interleaving will be removed. If it happens will it be automatic, or should I reboot the modem?

Cheers, Simon.

It will be automatic. DLM works overnight, between 2am and 5am.

There are 3 levels of downstream interleaving and 2 levels of upstream interleaving. I believe everything happens one stage at a time.

If your line has less than 288 severe errored/errored seconds in a 24 hour period, your line will be flagged green and this should eventually trigger a profile upgrade - it's a bit more complicated when this happens though because DLM has a caution counter system to avoid moving someone up and down like a yo-yo.

When DLM removed intervention on my line, it was precisely 8 weeks after intervening. I need to look at more of my ES stats to figure out the low threshold...

Do you know whether, when DLM reduces intervention, it is only looking at an ES threshold, or is it also making some value decision about the level of FEC's being encountered (or the plausible, but never visible, FECS counter)?

Yeah, that's the idea.

In terms of removing interleaving/banding, DLM has a barrier designed to stop it making changes too quickly that will cause instability. When your line is set up for the first time or DLM is reset, then the first 48 hours DLM will not intervene.

At the start, your barrier is set to 0. If DLM intervenes, you barrier will be set to 8 - which means that it will take 9 days for DLM to remove interleaving/change profile. If you have too many errors on day 5, then DLM shouldn't intervene but your barrier will increased to 16. It can keep being increased all the way up to 64 (so you would have to wait 65 days for it to removing interleaving/increase profile).

If DLM intervenes and makes a positive change, then the barrier is reduced by 2 for each good day thereafter. It's reset back to what it was if you have too many errors.

So if your barrier is 64, you have to wait 65 days for DLM to make an improvement. If it makes an improvement, then you next have to wait 21 days (barrier decreasing from 64 by 2 each day until the days since the last profile change matches with the barrier for the next improvement).

Kind of complex!

well this line is a lot better on ES

in 20 hours

155 CRC and 102 ES - before my fault on 6db snrm I averaged around 1500 CRC a day.

US has just 11 CRC and 11 ES

Great post. Thanks for taking the time to write it.

Paul

Aha. I think I understand the way the barrier thing works to restrict the de-intervention, but now I wonder how it affects further intervention...

If we start with a scenario where ES goes over 2880 for the first time, then DLM intervenes: this adds FEC and interleaving, and sets the barrier to 8.

If, during those 8 days, the ES value goes above 2880 again, what happens: I'd have originally though that DLM would intervene further, increasing FEC/interleaving. Are you saying that the barrier is increased instead? Or perhaps as well? Or are there different ES-thresholds for the barrier increase and for the extra DLM intervention?

And adding to the complexity...

You mention a case where "DLM intervenes and makes a positive change," and that the barrier is set back to what it was if there are subsequently too many errors: Is this adjustment to the barrier instead of DLM re-intervening with a negative change, or do both changes occur?