Impact of DLM and DSM for operators

Anyone seen this before?

Presentation to ICCCN 2011 - Covering DSM levels 1 (DLM), 2 and 3 (Vectoring).

Pages 13 and 14 shows the effect of introducing TRA (the kind of DLM used in WBC ADSL2+) in "Europe", with the improvements in stability and speed. Other pages have more results.

It is interesting evidence about why DLM works from an operator's perspective over many thousands of lines, even if many individuals don't like what it has done to their single line.

not got time to read it now but I believe you are reffering to power management so DLM changes power masks to try and manage crosstalk, this I suggested in another thread.

Is a bit different tho to DLM been over the top applying interleaving after one off or rare problems.

As I understand it, DSM refers to all attempts to adjust DSL lines to make them "better". The "whom" that it is better for differs though.

DSM Level 1 is where they make static adjustments to a single line at a time, based on readings from that line. The 3 current DLM mechanisms seen in 20CN, 21CN and FTTC all fit squarely within this definition.

In most technical articles, DLM is only ever equated with this first level of DSM.

DSM Level 2 is where they make static adjustments to a single line at a time, based on readings from many lines. Power management and changing power masks would fit into this definition better, because the reduction in power as a whole, or reduction in power at different frequencies benefits different lines from the one being adjusted.

I'm not sure if UPBO is considered a DSM-level-2 technique, or if it is just set once-and-for-all.

DSM Level 3 is about making dynamic adjustments to many lines, based on readings from many lines. Vectoring fits here.

As I read it, these levels are not mutually exclusive. Lines that are being vectored will also require DLM running on them. However, as it seems that vectoring allows UPBO to be turned off, and gives rise to faster upstream speeds, perhaps it means less power management is needed.

Telecoms engineers like reliability, and accountants don't like truck rolls / engineer visits so the supply side is all about stability and reliability. Most end users want reliability and a minority want to get the last kbit/s out of their line.

Seems the industry is catering for the wishes of the majority in general.

As a telecoms engineer (software variety) I can tell you how right you are, but in a subtly understated way. You also need to add robustness to the reliability, while never forgetting the sheer vast scale.

The concept of "always on", preferably without fail, but continuing to work even if it *does* fail, does lead the mind in strange directions.

The exchange software I worked on would all require you to keep calls active, even if the software restarted. The hardware would be duplicated in case of faults. A huge amount of testing went into making sure it didn't restart. But a huge amount of design effort went into coping when it did.

A fact which I was rather glad of when I accidentally took out one of the networks in the South-East one night

Imagine Microsoft having to get their software to never BSOD, but when it does BSOD anyway... then keep working while it recovers.

In reply to a post by yarwell:

Telecoms engineers like reliability, and accountants don't like truck rolls / engineer visits so the supply side is all about stability and reliability. Most end users want reliability and a minority want to get the last kbit/s out of their line.

Seems the industry is catering for the wishes of the majority in general.

Whilst I agree with you, its a bit harsh to be calling it the last kbit as crosstalk can amount to 10s of mbits/sec.

In reply to a post by WWWombat:

As I understand it, DSM refers to all attempts to adjust DSL lines to make them "better". The "whom" that it is better for differs though.

DSM Level 1 is where they make static adjustments to a single line at a time, based on readings from that line. The 3 current DLM mechanisms seen in 20CN, 21CN and FTTC all fit squarely within this definition.

In most technical articles, DLM is only ever equated with this first level of DSM.

DSM Level 2 is where they make static adjustments to a single line at a time, based on readings from many lines. Power management and changing power masks would fit into this definition better, because the reduction in power as a whole, or reduction in power at different frequencies benefits different lines from the one being adjusted.

I'm not sure if UPBO is considered a DSM-level-2 technique, or if it is just set once-and-for-all.

DSM Level 3 is about making dynamic adjustments to many lines, based on readings from many lines. Vectoring fits here.

As I read it, these levels are not mutually exclusive. Lines that are being vectored will also require DLM running on them. However, as it seems that vectoring allows UPBO to be turned off, and gives rise to faster upstream speeds, perhaps it means less power management is needed.

yeah vectoring should reduce the need for DSM. Although I expect it will still be used in tandom with vectoring.

In reply to a post by WWWombat:

As a telecoms engineer (software variety) I can tell you how right you are, but in a subtly understated way. You also need to add robustness to the reliability, while never forgetting the sheer vast scale.

The concept of "always on", preferably without fail, but continuing to work even if it *does* fail, does lead the mind in strange directions.

The exchange software I worked on would all require you to keep calls active, even if the software restarted. The hardware would be duplicated in case of faults. A huge amount of testing went into making sure it didn't restart. But a huge amount of design effort went into coping when it did.

A fact which I was rather glad of when I accidentally took out one of the networks in the South-East one night

Imagine Microsoft having to get their software to never BSOD, but when it does BSOD anyway... then keep working while it recovers.

Microsoft have sort of managed to achieve that already.

eg. in vista they changed it so sound was processed via software instead of hardware which in turn brought about less system crashes due to sound processing. Creative didnt like it but it reduced BSOD's and other fatal scenarios. We now see with windows 7 and newer also display drivers are less likely to cause a BSOD, you may have seen a message like "the graphics card has recovered from a display driver failure" or similiar, now its possible for the graphics driver to crash and auto recover without any crash or loss of data.

Thinking of the same concept on brodband is the line been resynced a similiar thing? generally I think no on dynamic ip's as that kills any established connections, however on static ip's (or if ppp has a forced long timeout on dynamic to maintain same ip) it can simply freeze throughput momentarily and then everything carries on, isp's like BT could quite possibly reduce complaints on unstable lines simply by supplying static ip address's which could mask disconnections. SRA I would consider the same concept as that allows adjustments and recoveries without a outage.

Edited by Chrysalis (Fri 19-Apr-13 06:31:58)