Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[sevenlayermuddle]

I've just noticed my target SNRM has been reduced to 6dB.   

I'd been half-expecting it, as I'm forming a theory about what DLM 'looks for' before reducing the target margin.   That theory is, and all of my reductions would be consistent with it, is that what DLM looks for is SNR sustainability in relation to the target, in addition to - and maybe much more that - error rates. 

As I recorded last year, I believe I twice 'fooled' DLM into a reduction by contriving an artificially high initial SNRM (>> target), so that my actual SNR rarely dropped below target.  For the past few weeks, although there had been no change in my connection speed (I've held sync since about Christmas) I'd noticed a significant improvement in my actual SNRM, such that it's daily wanderings were ranging from about 8.5 to 13, but rarely dropping much below the 9dB target.   I've no idea why my SNR's been better, but  that's not really relevant.  I resisted the urge to reconnect at a better speed, but decided instead that I'd wait and see if DLM reduced the margin.  My error rates have good but nothing special, typically up to maybe one or two hundred crcs or more on some days. 

An interesting thing this time around was that my line didn't automatically reconnect when the margin got reduced.  I simply became so convinced the target may have been dropped that I manually restarted and lo, I was right.

Regardless of whether my theory has any merit, right now I'm a happy bunny,  connected at 5024 with a 56dB attenuation. 

I'll probably reconnect again early evening when the noise levels are a bit more typical, but I still expect a decent speed.

[roseway]

That's very interesting, and I'll have a deeper think about it.

[sevenlayermuddle]

As a footnote to all that, I've noticed a lot of reconnections since that DLM reduction.

Some reconnects are entirely reasonable.  If it connects at 6dB  mid-morning then, unsurprisngly. the margin erodes at night and a reconnect may be unavoidable.  And under these circumstances, of course, the new speed is poorer.

However, it is also reconnecting for no reason that is obvious, sometimes when the margin is comfortably above 6dB, hence I get a higher speed.

These symptoms could of course be attributed to interference bursts, but it's a bit of a co-incidence that it only started about the time the margin got reduced, whereas, prior to that, it had stayed connected for 2.5 months solid at about 3500, yet it's connected at similar speeds a few times and yet still reconnected again later (usually faster).

I'm wondering if it could have launched into new ten-day training period.  Any thoughts?  Would that explain the symptoms?  Could it have been triggered automatically, as I never asked for it?

I ask just out of curiousity, it's not interfering with my enjoyment of life with a 6dB margin   

[roseway]

I wouldn't have thought that DLM would automatically relaunch the 10-day training period, although the behaviour does look a bit odd. Has your connection been switched from interleaved to fast-path perhaps?

Incidentally, my own connection has been behaving a bit oddly recently as well. After Christmas I enjoyed over two months of total stability with very low error rates when using the 2700HGV router, but my target margin stayed stubbornly at 9 dB. Then the line went crackly again, and for a while the connection was dropping every few minutes (more than 10 times in an hour) and yet the target margin still stayed at 9 dB. It almost looks as though I've got a stuck target noise margin.

[sevenlayermuddle]

Has your connection been switched from interleaved to fast-path perhaps?

It certainly wasn't switched immediately after the target reduction, as I checked.  But it's an interesting question that I'll investigate when I'm home (in the office at the mo').

The only 'logic' for DLM relaunching a ten-day train would be that my FSR and FTR would have been set long ago, when I had crap wiring and a lousy router, and typically connected at about 1 to 1.5 Mbps.  I can just about imagine, when a line with these parameters improves to the extent that it connects, and appears stable at, 4 to 5Mbps, some automatic reaction could occur from the management software, maybe even inviting human intervention to review it?

[roseway]

I can just about imagine, when a line with these parameters improves to the extent that it connects, and appears stable at, 4 to 5Mbps, some automatic reaction could occur from the management software, maybe even inviting human intervention to review it?

I don't wish to sound too cynical, but that scenario sounds a bit over-generous for BT.

[sevenlayermuddle]

I don't wish to sound too cynical, but that scenario sounds a bit over-generous for BT.

[sevenlayermuddle]

Incidentally, my own connection has been behaving a bit oddly recently as well. After Christmas I enjoyed over two months of total stability with very low error rates when using the 2700HGV router, but my target margin stayed stubbornly at 9 dB. Then the line went crackly again, and for a while the connection was dropping every few minutes (more than 10 times in an hour) and yet the target margin still stayed at 9 dB. It almost looks as though I've got a stuck target noise margin.

Sorry Eric, I didn't notice that paragraph earlier, didn't mean to ignore it.  Very interesting indeed.

I wonder if DLM has changed, somehow?  BT's SIN 472, at http://www.sinet.bt.com, dated January this year, mentions DLM, though it still doesn't shed much light on DLM internals.   But it does suggest there's a ' new algorithm to improve response times'.   That SIN is for WBC rather than IPStream, but I suppose it's conceivable that  any changes to DLM required for WBC might have rubbed off on IPStream?

The SIN also discusses DLM stability options, which have been mentioned on the forums before, but again - the document is meant to be WBC-specific and it would be rash to assume it had any relevance to IPStream. 

As regards ADSL2,  it also mentions DLM control of Impulse Noise Protection (for ADSL2), which also overlaps on recent topics.