How does DLM work?

[#1]

I have received much help already with understanding the REIN on my line, I have supplied stats and done tests in test sockets, changed filters etc. I try to follow the recommended way to disconnect/reconnect so as not to trigger the DLM though that is not always possible. Here, I am trying to ask a fairly specific question to help me carry that long process forward another step - not to drag you all over the same ground again that has been done to death elsewhere.
Firstly, have I got the initials correct. By DLM I mean the thing that monitors my line and sets the target noise margin. (Data Line Managment by chance?)
I tried discussing how DLM worked with CS the other day and thought I had got the idea but I'm not so sure now. Lets stick to 21CN and ADSL2 for the moment as I understand that has a different DLM from ADSL1 - a more sophisticated and quicker acting one.
I was told was that the DLM monitored my line constantly and would increase the target SNR margin if it saw too many interruptions and would decrease the target margin if it saw a stable line and would do both quickly - CS implied that the target margin would be reduced given as little as 3 or 4 hours of stable operation and increased immediately given only one qualifying interruption. They then said that the target changing did not immediately change my line performance as there had to be a disconnection and reconnection to give the modems a chance to put the new target margin into operation. An increasing margin is likely to be triggered by a disconnection so typically happens immediately. A decreasing margin has to wait for a disconnection to happen,  however, if a disconnection did not happen within 3 days, then the exchange equipment would trigger a disconnection automatically to allow that new margin to be implemented.
Three questions:-
1. Have I been sold a load of waffle or is CS essentially correct. - to the point where I can use the information as it stands without further qualification?
2. What constitutes stable operation? If disconnections are unstable then are no disconnections stable or is it cleverer than that - I would have thought it needed to be. Could it be that the DLM monitors the actual operating SNR in some way and looks for a period of stable high SNR before reducing the margin.
3. Where does the training period fit into this - if the above is true then training hardly sounds necessary - or is it the training period I have described- in which case what happens when training is complete.
And the punchline - why if CS are correct and I have a period of excessively high stable SNR and then disconnect/reconnect, do I not get a better speed than before :'(.

[#2]

DLM stands for Dynamic Line management. - agreed lets stick with 21CN
It is essentially a piece of software that monitors your line and issues instructions to the DSLAM that controls your line to do something as and where necessary.
It is quite clever and clearly stores a history of your line.
If it sees lots of disconnections it will indeed raise your target margin in steps (3,6,9,12,15 are typical targets)
How quickly it will do this depends on the number of disconnections/per fixed period and your line's history.
I have heard figures of 10 disconnections in an hour will trigger an immediate rise - but don't take this figure as definitive.
I've found that it  takes longer than a few hours to drop the margin back down again after an "event" - such as a fault.
I'd say a few days continuous stable operation is more likely.
It will monitor your SNR during that period and decide whether the SNR is stable enough to enable it to run at a lower target.
We are not just talking about true disconnections here, any from of glitches etc which cause the SNR to dip sharply , but do not cause a disconnection mean that the line is not suitable for a lowered target.  The system also has blip logic in it such that it will ignore isolated one off glitches - on the basis they are one off events from an unknown source.
I'm not sure whether it takes into account the level the SNR drops to overnight in its calculations as to whether it can lower the target margin. there are pros and cons of doing either.  The target margin is really to accomodate glitches on the line not daily variations in SNR which are dealt with by lowering the sync for a given target SNR margin.
Training is of less importance on 21CN (due to its continuous line monitoring) than on 20CN where the line tended to be 'fixed' at the end of training.
The main reason for training is to establish your sync fault level which is 80% of your lowest sync speed occuring during the 10 days.  Any sync in the future below your fault level is deemed to indeed be a fault and not general day to day variation - I think it set too low but that's another issue.
Your true throughput depends on other things and not just the sync speed and the target SNR margin which will determine the sync speed.  
Errors on the line meaning data has to be re-transmitted will result in the throughput will decrease even if the unit remains in sync and the SNR margin seems normal.
Oh and when the DSLAM changes your target SNR margin it WILL issue a re-sync command automatically .  You need do nothing.

[#3]

@x47c  Thanks for a really good answer.
I am trying to decide whether to go onto a banded profile (I think that's the terminology- - the DLM gets turned off and I live with a fixed highish SNR target margin)
At the moment, my REIN can give me a change in SNR of 8 db or more within minutes several times each day so I could select to be on a fixed margin of 9db and would suffer reduced speed (from ideal) all the time but would probably maintain a working internet during the times when the SNR is high. Alternatively,  I could stick with the DLM doing its thing in which case it will apply a much too high margin (15db at the moment) some of the time and a too low margin at other times when the connection will collapse if the REIN occurs. But when the source of the REIN blows up (soon please), I will get a much better speed than if I am stuck on a profile.
You can see that the answer depends quite a lot on whether  "a couple of hours" or "a couple of days" for the recovery time from a high SNR is correct.
Any advice from those with experience of banding would be welcome.

[#4]

REIN used to cause DLM to apply a banded profile to my line (much to my disusgust).  A good while back there were trials for ISP set 'fixed' DLM profiles which was absolutely brilliant for intermittent REIN and much preferable to banding but this was dropped in preference to other DLM changes, luckilly I was able to stay on the trial type profile until I changed to FTTC which finally solved my REIN issue instantly.
My REIN issue was shown particularly well by logging cumulative CRC errors in routerstats:
Graph flat is no (significant) errors occurring, graph at maximum slope is massive regular errors,the sudden drop was a router reboot (back to 0 errors).
My noise margin during a REIN event (different time to the above one so they don't match):

[#5]

@walker23  Can you give a bit more information about the difference between banding and the fixed DLM profile you mention. CS seemed to offer to turn off the DLM which sounds more like your fixed to me.
Latest episode in my maddening story of country folk and our broadband.
Router has been running happily for several days without a disconnection, a generally excellent level of SNR (14db) though there was one lower period (9db) yesterday. So I confidently expected to see my margin drop and speed go up. This morning the automatic resync happened (I guess) and lo and behold, the SNR Margin went up to 17db and my speed went down to 2268kbps. Totally the wrong way. (This on a line that BT say is capable of 17Mbps and on which I have seen 6Mbps)
So I tried a more modern router- did the change over carefully and the noise margin dropped to 7dB and the speed stayed exactly where it was at 2268kbps.
There is something here that I do not understand. Noise margin down should be sync speed up and vice versa. But it does not seem to work that way at all. I seem to get noise margin up, speed down, noise margin down, speed down
The more modern router is poor and I cannot much in the way of statistics off it - shortly, I will change back to the older router (DG834 with an AR7 chipset) and see if I can get the CRC count to plot as suggested above.
Any thoughts on why the speed/noise margin does not work out as it is supposed to.

[#6]

This is the only thread I can find about it at the moment (plenty there to read  😞 http://community.plus.net/forum/index.php/topic,82898.0.html
If you're lucky enough to be put onto this kind of 'ISP managed profile' rather than DLM (that's what I'd interpret as 'DLM turned off') then you should see a big improvement though I did think it had been abandoned (or was very difficult to apply).  The main plus of the trial system was the ability to resync whenever required without suffering raised target noise margin or banding (profile still drops when a low sync occurs though).
Good luck

[#7]

I think the reason sync speed stayed at 2268kbps is most likely that the line has been banded on the 1152-2272kbps profile because the DLM is not happy. Initially noise margin increased because the sync speed was held to a lower value by the banding.
When the router was changed the same sync speed was achieved, but the much reduced noise margin could be due to high noise on the line (power supply interference problems?) or the router is not in the best of health.

[#8]

Quote from: spraxyt

....... line has been banded on the 1152-2272kbps profile because the DLM is not happy.....

That implies that the banding is based on speed rather than SNR Margin. I have been assuming that banding meant that the DLM was limited to a range of SNR Margins that it was not allowed out of rather than a range of speeds. Can you confirm that I have understood you correctly and perhaps point me to the source of your figures?
Secondly, does the DLM impose the banding and what prompts it to remove it. Given that my line has run at 6Mbps not that long ago and should run faster than that - this is appalling.

Quote from: spraxyt

When the router was changed the same sync speed was achieved, ......... or the router is not in the best of health.

I am not impressed by the router I switched to although it was new 3 or 4 weeks ago. It might just be a software bug where it reports 17 without the tens digit ie 7. That would explain the effect neatly but depends on your view on the people who write software!
The interference shows up whichever router I use - with different PSU bricks and I have had everything else turned off at one time or another and still seen the interference so I am pretty convinced it comes in from outside.