How do you find out how far you can push connection limits with DMT tools?

[Admacp]

How do you find out your maximum stable connection speed without trial and error spread over weeks or months? At the moment I have been pushing my connection rate up to 1920kbps using DMT tools (7 days ago) to give me an IP Profile of 1500. I have a very long line (66.7dB/31.5dB) and at the moment my connection appears to be 100% reliable over the course of this week with an snrm that slowly crept up from an average of 5.9dB last week to around 6.5dB earlier this evening. I have just pushed the connection rate up to 2016kbps to jump me up to the next IP Profile of 1750. This has dropped my snrm down to about 5.6dB and so far everything seems to be running smoothly (11pm at night and now up at the higher connection rate for about 2 hours).

DMT Tools is reporting that my RCO is 88% and the max. connection is 2304kbps (IP Profile of 2000). I have noticed that each time I tweak the connection upwards the RCO seems to follow it upwards as well. Does that mean that "potentially" it might be worth tweaking ever so slightly upwards as long as the snrm doesn't get too low? Has anyone managed to unravel the mysteries of the max. amount of error seconds, CRC errors etc. before your connection suffers dropouts? I've just checked my stats and in the last 15 minutes I've received 0 CRC's and 0 Error seconds. The previous 15 minute spell it was 2 CRC's and 2 error seconds. The previous 15 minute spell to that, it was 5 CRC's and 4 error seconds and the 15 minute spell before that was 8 CRC's and 5 error seconds. It looks like the line is settling down and the snrm has risen about 0.3dB since the increase took place (in fact it looks like I am receiving fewer CRC's per hour now than I previously was but possibly slightly more error seconds!)

What is the likely result of pushing things too far - I would assume that I would experience a few router dropouts but what I am worried about is the DLM raises the snr in one big jump and I'll have to start all over again and wait my 3 days for the IP Profile to rise again after re-tweaking.

One last question regarding DMT Tools. It shows in the middle graph aSNR dB per tone figure of High 37dB and a low of 9dB. Before I tweaked it tonight it showed a high of 37dB and a low of 11dB. Any idea what these figures are?

[Mick]

In a nutshell, the more you 'push it' the higher the error count your modem will see.  Unfortunately, the error limit at which the DLM will decide that your line is unstable and increase the target SNR by 3dB to stabilise it is not documented, or publicly available.  It is also very likely that the error rate at which the DLM increases the SNR is different between different DLM chipsets and modem combinations.

From my experience the absolute FECs, CRCs counts is not important - during thunderstorms I had FECs increasing to thousands - but the rate of their change is critical to the DLM kicking in.  The sensitivity of the DLM to line error rates is determined by the 'service setting' (extra stable, stable, standard) that BT have applied to your line the exchange.  Have a word with your ISP if you want to change this, but bear in mind that with your distance from the exchange you may alsready be at your  best setting anyway.

Another thing to pay attention to is REIN.  Your line stats look nice and steady until REIN occurs and then suddenly you lose weeks of 'good behaviour' and get a penalty from DLM.

[sevenlayermuddle]

There's a BT patent that seems to (can't guarantee that)  describe DLM, which can be seen here :

http://www.freepatentsonline.com/EP1953959.html

If you want to read it, you may want to register yourself with the patents office and then you can download a version with proper diagrams (it's a while since I did so, can't remember exactly what it involved).  Basically the diagrams descibe a method for counting up to ten  

The BT patent explains how lines that are 'unstable' may (basically) have their target margins increased by 3dB steps.  It doesn't define what it means by 'unstable'.

More problematic perhaps, it also describes that 'very unstable' lines may have the target increased by 6dB in one go.  It seems to define 'very unstable' as 10 resyncs ( for any reason) in a single hour.  That's clearly worth avoiding, so if your router keeps losing sync with a tweaked SNR,  then don't keep trying....

[jeffbb]

Hi
quote during thunderstorms I had FECs increasing to thousands

FECs(forward error correction) are nothing to worry about they are the errors that were corrected. CRC  on the otherhand are different these are not corrected and so have to get retransmitted ,so slowing down throughput ,if really bad then can cause resynch .

quote from Kitz:
At the receiving end a decoder then performs the same calculation on the data in the packet and then compares it to the transmitted CRC code.
The decoder produces the output data only if the codes match indicating that there are no errors. If an error occurs then the decoder signals an erasure (loss of data), and a higher level ARQ then re-requests the data.

http://www.kitz.co.uk/adsl/error_correction.htm


Regards Jeff

[Mick]

Hi
quote during thunderstorms I had FECs increasing to thousands

FECs(forward error correction) are nothing to worry about they are the errors that were corrected. CRC  on the otherhand are different these are not corrected and so have to get retransmitted ...

This makes sense in terms of my observations of a resync.  During the thunderstorms CRCs were relatively low (around 2 - 6 per 5 minute average) and the modem held the line.  Other times although the FECs are realtively low, CRCs increase and bang - a resync is ievitable.  Are you saying then that only changes in CRCs affect the DLMs algorithm that decides to force a resync?

[jeffbb]

Hi
quote : Are you saying then that only changes in CRCs affect the DLMs algorithm that decides to force a resync?

As far as FECs and CRCs  yes , there are other parameters that will upset the DLM.

more info at http://www.kitz.co.uk/adsl/maxdsl2.htm

Regards Jeff

[sevenlayermuddle]

Are you saying then that only changes in CRCs affect the DLMs algorithm that decides to force a resync?

From that BT patent paper...

The default line profile which is initially applied to all lines has a target margin of 6db and interleaving disabled (often referred to as being in fast mode). Changing these parameters is based on three performance metrics, errors (in particular, in this embodiment, errors caused by code-violations), re-trains (i.e. re-syncs) and failed initialisations.