Turn on suggestions
Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.
Showing results for
10 day training period - what's the exact process?
Topic Options
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page
- Plusnet Community
- :
- Forum
- :
- Help with my Plusnet services
- :
- Broadband
- :
- Re: 10 day training period - what's the exact proc...
10 day training period - what's the exact process?
07-08-2012 9:00 AM
- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Report to Moderator
Following a physical line fault being fixed by BT, I now have my 21CN ADSL2+ service back in training mode. I know what this is at a high level but can't find any info online about precisely what happens during this 10 day period.
I've been able to see that during the first 36hrs I had an upstream sync fixed at 888kbps as upstream interleaving was set at 8. As of yesterday and today (approx 60hrs since the reset) the upstream has pushed up to 1175kbps with an interleaving depth of 4. Downstream is still using an interleave depth of 64 and is sync'd at 11381kbps.
I'm happy to let the training continue and am not some impatient speed freak looking to crank it up before it's ready.... But I'd be really interested to understand the process the DSLAM and DLM software is going through. Does it gradually decrease the interleave depth over time until it reaches FastPath (no interleaving)? And if so, what period does it test each incremental point? Does it depend purely on time or the amount of data sent over the connection? (or purely a maximum number of unrecoverable errors). What sequence of tweaks are tried? (ie interleaving, SNR margin, trellis, etc.)
The Internet appears to be full of people talking about the 10 day training period but no-one talks about it with any transparency. Just "leave it alone and it'll try to find the best configuration". Does anyone from PlusNet have some technical detail they're able to share?
Thanks,
Rob
I've been able to see that during the first 36hrs I had an upstream sync fixed at 888kbps as upstream interleaving was set at 8. As of yesterday and today (approx 60hrs since the reset) the upstream has pushed up to 1175kbps with an interleaving depth of 4. Downstream is still using an interleave depth of 64 and is sync'd at 11381kbps.
I'm happy to let the training continue and am not some impatient speed freak looking to crank it up before it's ready.... But I'd be really interested to understand the process the DSLAM and DLM software is going through. Does it gradually decrease the interleave depth over time until it reaches FastPath (no interleaving)? And if so, what period does it test each incremental point? Does it depend purely on time or the amount of data sent over the connection? (or purely a maximum number of unrecoverable errors). What sequence of tweaks are tried? (ie interleaving, SNR margin, trellis, etc.)
The Internet appears to be full of people talking about the 10 day training period but no-one talks about it with any transparency. Just "leave it alone and it'll try to find the best configuration". Does anyone from PlusNet have some technical detail they're able to share?
Thanks,
Rob
Message 1 of 9
(4,543 Views)
8 REPLIES 8
Re: 10 day training period - what's the exact process?
07-08-2012 3:59 PM
- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Report to Moderator
Hi Bobsta,
Information regarding the 10 day training period can be found <a href="http://www.kitz.co.uk/adsl/maxdsl2.htm">here</a>. I've also flagged this up to get our content updated so information is available via our website for this.
If you've got any further questions, just ask
Chris
Information regarding the 10 day training period can be found <a href="http://www.kitz.co.uk/adsl/maxdsl2.htm">here</a>. I've also flagged this up to get our content updated so information is available via our website for this.
If you've got any further questions, just ask
Chris
Message 2 of 9
(2,320 Views)
Re: 10 day training period - what's the exact process?
07-08-2012 6:39 PM
- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Report to Moderator
I think this is a very good question, not well answered on the Kitz web site linked by Chirspurvey.
At the start of my training period I was getting a sync rate just above 9500kbps at an SNR of 6dB. Now it's day 14 and for the last 40 hours I have been synced at 9187kbps at an SNR of around 4dB. Meanwhile my upload rate has increase from 888kbps to 1155 kbps. So it's almost as if I have had ~300kbps shaved off my download rate and given to my upload rate!
At the start of my training period I was getting a sync rate just above 9500kbps at an SNR of 6dB. Now it's day 14 and for the last 40 hours I have been synced at 9187kbps at an SNR of around 4dB. Meanwhile my upload rate has increase from 888kbps to 1155 kbps. So it's almost as if I have had ~300kbps shaved off my download rate and given to my upload rate!
Message 3 of 9
(2,320 Views)
Re: 10 day training period - what's the exact process?
07-08-2012 10:32 PM
- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Report to Moderator
Thanks Chris but I have to agree with ReedRichards.
I've browsed the Kitz info before and whilst it's some of the best info on ADSL on the Internet, it still doesn't really tell me how training works on a 21CN setup. The article you linked to talks about training on a 20CN setup, leading to setting MSR, etc... which AFAIK just doesn't apply on 21CN.
I've browsed the Kitz info before and whilst it's some of the best info on ADSL on the Internet, it still doesn't really tell me how training works on a 21CN setup. The article you linked to talks about training on a 20CN setup, leading to setting MSR, etc... which AFAIK just doesn't apply on 21CN.
Message 4 of 9
(2,320 Views)
Re: 10 day training period - what's the exact process?
08-08-2012 10:51 AM
- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Report to Moderator
Since moving to ADSL2+ the DLM did quite a good job of selecting a sustainable DL and UL rate and SNRM (around 9700kbps/1000kbps/6dB respectively) with no interleaving, but the SNRM would slowly decrease after each resynch until after about a week of continuous connection it would drop to around 1dB during the evening peak and start to produce huge numbers of errors, and sometimes drop the connection.
After going away for a few days and switching everything off, on rebooting I found that downstream interleaving had been applied for no apparent reason, and the target SNRM had dropped to 3dB with a downstream sync of over 12000kbps. This proved to be unsustainable in the evenings with huge numbers of errors being generated, so for the last few months I've run it with the SNRM tweaked from 3-5 via Routerstats and am perfectly happy with a stable connection rate of 10764/1184 at 5.8dB as I type this.
I think the 10 day training period is a handy (but probably sensible) way that ISP’s make sure they don’t have to waste their time dealing with non-existent “faults” – as for what happens subsequently (my interleaving “event” and 3dB target SNRM happened at least 4 months after the switchover) I can’t help feeling that for me at least the DLM is a bit over-ambitious, not to mention unpredictable!
After going away for a few days and switching everything off, on rebooting I found that downstream interleaving had been applied for no apparent reason, and the target SNRM had dropped to 3dB with a downstream sync of over 12000kbps. This proved to be unsustainable in the evenings with huge numbers of errors being generated, so for the last few months I've run it with the SNRM tweaked from 3-5 via Routerstats and am perfectly happy with a stable connection rate of 10764/1184 at 5.8dB as I type this.
I think the 10 day training period is a handy (but probably sensible) way that ISP’s make sure they don’t have to waste their time dealing with non-existent “faults” – as for what happens subsequently (my interleaving “event” and 3dB target SNRM happened at least 4 months after the switchover) I can’t help feeling that for me at least the DLM is a bit over-ambitious, not to mention unpredictable!
Message 5 of 9
(2,320 Views)
Re: 10 day training period - what's the exact process?
08-08-2012 2:10 PM
- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Report to Moderator
During the first 9 days of my ADSL2+ training period the router re-synced about 5 times and each time the sync rate went down. I don't know what prompted this because there were long periods of apparent stability between each re-sync. Perhaps it was the upstream parameters that required adjustment? Since day 9 the downstream sync rate has been going back up but I have no idea what SNR is now being targeted; my average seems to be about 4 dB.
Message 6 of 9
(2,320 Views)
Re: 10 day training period - what's the exact process?
08-08-2012 10:12 PM
- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Report to Moderator
I think the honest answer is that PN won't know.
If a line was to be re-trained all PN/BT probably do is log on to DLM Element Manager and put the line into training mode.
How the DSALM carries this out, the parameters it looks at and the algorithms it uses to come to a decision as to the best settings for the line will only be known to the DSLAM manufacturers. In other words it is commercially confidential info.
No one (outside the manufacturers) seems to really understands how depth of the interleaving is decided upon.
I did do some comparisons and it seems you can have the same depth - but with different framing parameters.
Perhaps interleaving depth is a bit like the router attenuation figure quoted in that it is a single figure which everyone 'uses' wheras in reality this figure is merely indicative of a lot more complex stuff going on in the background.
I suspect you also need a degree level education in data transmission and compression techniques to have the capability to understand the concepts - I don't!
From what I have seen I'd suspect in the case of the upstream interleaving question having reduced the depth from 8 to 4 and removed the interleaving cap of 888 and let the sync 'run free' at whatever it can at 6 db upstream target margin ------ it will leave it there.
I know you lot on here are obsessed with fast path and I know why - however I tend to the opinion that all lines should be interleaved unless there is a very good reason why not.
Some lines/routers do seem to have a problem with uncapped upstreams. When uncapped the upstream bits/tone are full loaded rather than only being part loaded when capped. Those fully loaded higher frequency parts of the upstream frequency spectrum seem to bleed through into the adjacent downstream lower tones, thereby reducing the bit loading capacity of the lower downstream tones and hence sync speed.
I personally suspect it the the less than perfect routers rather than the lines!
A bit like pre DAB some radios are better at segregation/discrimination of adjacent radio stations than others are/were. Aviation grade radios have very strict specifications to only receive/transmit very precisely on frequency and not to affect adjacent ones - and they cost a bomb as a result. and we/the ISP's don't want to pay for expensive routers.
On training..........
It certainly does not depend on the amount of data being downloaded. ADSL runs at I think 4000 frames per second whether or not you are downloading stuff. If you are not, then the frames contain dummy pseudo-random data. This can best be seen leaving on your computer overnight on idle and running full routerstats with telnet access and a suitable router. look at the errors graph and you will see most probably loads of CRC errors occuring overnight unless your line is exceptionally quiet/good. How so? you say as I'm not transmitting any data up or down so there should be no errors. Well the router/DSLAM are still communicating at the 4000 frames per second regardless - they are just sending dummy data and registering the CRC's on that instead.
The SNR margin target may be set to (for example) 3, however it seems on my line anyway that the sync aims to actually initialise at around 3.3 to 3.5. A great deal goes on in terms of analysing the line in those first few seconds before it comes to a decision on the correct sync to use. Once it has sync'd of course the actual SNR margin may go up or down depending on the amount of "noise" appearing on the line when compared to the amount at the instant of sync.
Training on 21CN seems less relevant as the line is managed more actively than 20CN which seemed to set the line parameters during the 10 days and then stick to them. 21CN seems more willing to change things should conditions change.
If a line was to be re-trained all PN/BT probably do is log on to DLM Element Manager and put the line into training mode.
How the DSALM carries this out, the parameters it looks at and the algorithms it uses to come to a decision as to the best settings for the line will only be known to the DSLAM manufacturers. In other words it is commercially confidential info.
No one (outside the manufacturers) seems to really understands how depth of the interleaving is decided upon.
I did do some comparisons and it seems you can have the same depth - but with different framing parameters.
Perhaps interleaving depth is a bit like the router attenuation figure quoted in that it is a single figure which everyone 'uses' wheras in reality this figure is merely indicative of a lot more complex stuff going on in the background.
I suspect you also need a degree level education in data transmission and compression techniques to have the capability to understand the concepts - I don't!
From what I have seen I'd suspect in the case of the upstream interleaving question having reduced the depth from 8 to 4 and removed the interleaving cap of 888 and let the sync 'run free' at whatever it can at 6 db upstream target margin ------ it will leave it there.
I know you lot on here are obsessed with fast path and I know why - however I tend to the opinion that all lines should be interleaved unless there is a very good reason why not.
Some lines/routers do seem to have a problem with uncapped upstreams. When uncapped the upstream bits/tone are full loaded rather than only being part loaded when capped. Those fully loaded higher frequency parts of the upstream frequency spectrum seem to bleed through into the adjacent downstream lower tones, thereby reducing the bit loading capacity of the lower downstream tones and hence sync speed.
I personally suspect it the the less than perfect routers rather than the lines!
A bit like pre DAB some radios are better at segregation/discrimination of adjacent radio stations than others are/were. Aviation grade radios have very strict specifications to only receive/transmit very precisely on frequency and not to affect adjacent ones - and they cost a bomb as a result. and we/the ISP's don't want to pay for expensive routers.
On training..........
It certainly does not depend on the amount of data being downloaded. ADSL runs at I think 4000 frames per second whether or not you are downloading stuff. If you are not, then the frames contain dummy pseudo-random data. This can best be seen leaving on your computer overnight on idle and running full routerstats with telnet access and a suitable router. look at the errors graph and you will see most probably loads of CRC errors occuring overnight unless your line is exceptionally quiet/good. How so? you say as I'm not transmitting any data up or down so there should be no errors. Well the router/DSLAM are still communicating at the 4000 frames per second regardless - they are just sending dummy data and registering the CRC's on that instead.
The SNR margin target may be set to (for example) 3, however it seems on my line anyway that the sync aims to actually initialise at around 3.3 to 3.5. A great deal goes on in terms of analysing the line in those first few seconds before it comes to a decision on the correct sync to use. Once it has sync'd of course the actual SNR margin may go up or down depending on the amount of "noise" appearing on the line when compared to the amount at the instant of sync.
Training on 21CN seems less relevant as the line is managed more actively than 20CN which seemed to set the line parameters during the 10 days and then stick to them. 21CN seems more willing to change things should conditions change.
Message 7 of 9
(2,320 Views)
Re: 10 day training period - what's the exact process?
11-08-2012 9:29 AM
- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Report to Moderator
Big thanks to x47c - that all made a lot of sense.
Actually I do have a degree in Electronic Engineering (not specifically encoding/compression but I did focus on telecoms and networking)... That said, I graduated in 1998 and since then have been mucking around doing IT so have forgotten all the useful stuff.
I suspect you're right that the training process is DSLAM specific, fairly proprietary and confidential. I can live with that.
I did manually kick my connection on Thursday evening. It had been stable for 5 days so I thought it wouldn't hurt. That caused upstream interleaving to drop to 1 (effectively off - and my M, T and R values all dropped to 1/1/0 for upstream too). Downstream stayed the same at D=64.
Unfortunately I got a bit greedy and yesterday evening (~22 hours after the last kick) I cycled the router again and upstream interleaving jumped back up to 4. I guess I shouldn't have been so impatient.
Interestingly my router reports that with an upstream interleave depth of 1 it's adding 1.13msec of upstream latency. With D=4 the delay is 5.95msec. Downstream D=64 is adding 7.97msec latency.
So I'd like to get rid of it... if my line can support it. But I'm still unclear as whether the training process will try to do that for me (on downstream) or whether PN need to push a button to request it. In case it's the latter, I've logged a ticket.
Anyway, given how lovely the weather is, I shall be leaving well alone and doing outdoorsy stuff this weekend.
Cheers,
Bobsta
Actually I do have a degree in Electronic Engineering (not specifically encoding/compression but I did focus on telecoms and networking)... That said, I graduated in 1998 and since then have been mucking around doing IT so have forgotten all the useful stuff.
I suspect you're right that the training process is DSLAM specific, fairly proprietary and confidential. I can live with that.
I did manually kick my connection on Thursday evening. It had been stable for 5 days so I thought it wouldn't hurt. That caused upstream interleaving to drop to 1 (effectively off - and my M, T and R values all dropped to 1/1/0 for upstream too). Downstream stayed the same at D=64.
Unfortunately I got a bit greedy and yesterday evening (~22 hours after the last kick) I cycled the router again and upstream interleaving jumped back up to 4. I guess I shouldn't have been so impatient.
Interestingly my router reports that with an upstream interleave depth of 1 it's adding 1.13msec of upstream latency. With D=4 the delay is 5.95msec. Downstream D=64 is adding 7.97msec latency.
So I'd like to get rid of it... if my line can support it. But I'm still unclear as whether the training process will try to do that for me (on downstream) or whether PN need to push a button to request it. In case it's the latter, I've logged a ticket.
Anyway, given how lovely the weather is, I shall be leaving well alone and doing outdoorsy stuff this weekend.
Cheers,
Bobsta
Message 8 of 9
(2,320 Views)
Re: 10 day training period - what's the exact process?
11-08-2012 8:17 PM
- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Report to Moderator
PN will probably want to wait until the end of training before doing modifications to the line settings.
D=64 seems a common setting downstream.
Once, post a fixed line fault I was left with a very high target margin and low sync and I noticed my downstream D setting was reduced to 32.
when the target was reset to 6 - back D went to 64.
But as I said above - who knows how it is set....
Sometime the 21CN/ADSL2+ DLM does seem to go nuts on some lines.
I have heard of one instance where the DLM switched interleaving off during the day and then switched it on during the evening/night then back off during the next day and so on: not what was probably intended methinks.
Anyway since you have the electronics degree and my degree is electrical from a decade (and more) earlier than yours - you can answer all the difficult questions on here
D=64 seems a common setting downstream.
Once, post a fixed line fault I was left with a very high target margin and low sync and I noticed my downstream D setting was reduced to 32.
when the target was reset to 6 - back D went to 64.
But as I said above - who knows how it is set....
Sometime the 21CN/ADSL2+ DLM does seem to go nuts on some lines.
I have heard of one instance where the DLM switched interleaving off during the day and then switched it on during the evening/night then back off during the next day and so on: not what was probably intended methinks.
Anyway since you have the electronics degree and my degree is electrical from a decade (and more) earlier than yours - you can answer all the difficult questions on here
Message 9 of 9
(2,320 Views)
Topic Options
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page
- Plusnet Community
- :
- Forum
- :
- Help with my Plusnet services
- :
- Broadband
- :
- Re: 10 day training period - what's the exact proc...