VDSL Performance & Tuning

Forums › New Zealand Broadband › VDSL Performance & Tuning

1 | ... | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | ... | 285

1578 posts

Uber Geek

Trusted

#680029 1-Sep-2012 10:55

Ok, so it's been a week and I'm still on DLM-2 (also tried a manual resync). Time to take action!

I've increased my SNRM target by 1db via the Fritz 'Line Settings' tab:

My stats now look like this:

So I've dropped ~1mbit from my receive line speed, but my errors seem to have dropped quite a bit (FEC is about half of what it was before and CRC errors seem to be less although I've only had it running in this config for a couple of hours so it's early days yet).

I also had an idea last night which I think may help us to work out what Mean Time Between Errors (MTBE) rate is required to transition to a better DLM profile. If we record what our CRC per minute (CRCPM) a day or two during a particular sync period and then make a note of what DLM profile we get changed to (or left on) and we can get enough people involved, we may be able to spot a trend.

What do you guys think about that idea?

sidefx

3711 posts

Uber Geek

Trusted

#680040 1-Sep-2012 11:17

SamF:
I've increased my SNRM by 1db via the Fritz 'Line Settings' tab:

So I've dropped ~1mbit from my receive line speed

yup that's pretty much exactly what I'm trying too. Though I decided to drop another notch this morning on the line settings tab. So I've traded about 3mbit for a decent reduction in errors (13db = about 1.5mbit decrease and a drop from 0.14 to 0.1) now on 14dB SNRM the CRC error rate is still dropping. And doing a speedtest the reduction in throughput is even less noticeable - it's still plenty for me; tbh I'd be happy dropping another 3mbit or more to improve latency.

Hopefully all I need to do is avoid resyncs now, but that's never been a problem as I've never had a resync except where I've initiated them!

I also had an idea last night which I think may help us to work out what Mean Time Between Errors (MTBE) rate is required to transition to a better DLM profile. If we record what our CRC per minute (CRCPM) is at the end of a 7 day sync period and then make a note of what DLM profile we get changed to (or left on) and we can get enough people involved, we may be able to spot a trend.

What do you guys think about that idea?

Excellent idea!

"I was born not knowing and have had only a little time to change that here and there." | Octopus Energy | Sharesies
- Richard Feynman

SamF

1578 posts

Uber Geek

Trusted

#680042 1-Sep-2012 11:25

Cool, good to see there's someone else doing what I'm doing so we can compare notes!

So how long were you on 13db SNRM target and how long have you been on DLM-4? Frankly I'm surprised that you've not been moved to a better DLM profile based on even your previous error rate.

richms

28168 posts

Uber Geek

Trusted

Lifetime subscriber

#680043 1-Sep-2012 11:26

Check this out:

Richard rich.ms

SamF

1578 posts

Uber Geek

Trusted

#680044 1-Sep-2012 11:29

Nice!! :D

When did that change over?

richms

28168 posts

Uber Geek

Trusted

Lifetime subscriber

#680045 1-Sep-2012 11:31

Went on that a few weeks back, but im more impressed at the corrected error count for 15 mins. Not sure what would have caused that since its normally only a few 100, not in the millions.

Richard rich.ms

SamF

1578 posts

Uber Geek

Trusted

#680046 1-Sep-2012 11:33

Yeah, I did notice that!! Bit odd!

So you're on DLM-1 with an ongoing CRCPM of 0.14... Interesting.

Then again, Napster was on DLM-1 with a CRCPM of 0.86, so dunno what's going on there!? Maybe FEC errors are counted after all (despite what the Chorus docos say)? Both yours and Napster's FECPM are quite low.

Equinox IT

Cloud spending continues to surge globally, but most organisations haven’t made the changes necessary to maximise the value and cost-efficiency benefits of their cloud investments. Download the whitepaper From Overspend to Advantage now.

sidefx

3711 posts

Uber Geek

Trusted

#680048 1-Sep-2012 11:37

@SamF: It's DLM3 isn't it? I've been on it since 10/08 so I guess not that long. Though I had a 13 day run without touching the modem and no resyncs and a very consistent error rate around 0.14 and no change, so that's when I decided to drop the SNRM by 1. This morning I couldn't resist dropping it another 1 to get the error rate down a bit more. It's just dropped to 0.05 so the average is still going down. Now I'm just gonna leave it and see what happens.

@richms: Nice. Interesting your FEC seem quite high(? EDIT: Or not, I haven't really been paying attention to FEC in people screenshots!) but CRC quite low.

"I was born not knowing and have had only a little time to change that here and there." | Octopus Energy | Sharesies
- Richard Feynman

SamF

1578 posts

Uber Geek

Trusted

#680055 1-Sep-2012 12:14

sidefx: @SamF: It's DLM3 isn't it?

Yes, sorry you are right :)

SamF

1578 posts

Uber Geek

Trusted

#680077 1-Sep-2012 13:32

@richms: has your error rate settled down now or is it still crazy?

richms

28168 posts

Uber Geek

Trusted

Lifetime subscriber

#680081 1-Sep-2012 13:39

all looking ok now, notice bitswap has changed?

Richard rich.ms

SamF

1578 posts

Uber Geek

Trusted

#680082 1-Sep-2012 13:42

Hmm, interesting, I guess there was some line interference. Mind posting your statistics tab?

richms

28168 posts

Uber Geek

Trusted

Lifetime subscriber

#680083 1-Sep-2012 13:47

Richard rich.ms

SamF

1578 posts

Uber Geek

Trusted

#680085 1-Sep-2012 13:50

Cheers :) So looks like your CRC error rate is pretty stable & low in general.

SamF

1578 posts

Uber Geek

Trusted

#680098 1-Sep-2012 14:55

In the interests of science I've gone back through this topic and extracted relevant stats from various people's connections to see if I can see the threshold required to be swapped to DLM1. The data isn't conclusive, but from what I can see the CRCPM threshold seems to be somewhere between 0.20 and 0.40*. From the data we have, if your CRCPM is under 0.20 you're looking pretty good for a swap to DLM-1. I have also looked at FEC trends and they pretty much follow CRC error patterns so I don't think FEC has any direct influence on DLM profile selection. However, I will say that a high FEC rate may well indicate a future increase in CRC errors is likely.

* I realise this seems to be at odds with your experience sidefx, but unfortunately you seem to be the odd one out!

Summary of Stats
User      DLM CRC
SKYMAX    1    0.06
richms     1    0.14
Allan       1    0.21
SamF      2    0.39
Instinct     2   0.5
jjnz1       2 2

I also noted any trends where users were either swapped to DLM1 or off DLM1 and what their error rates were before this happened:

Swapped onto DLM1
User       CRC
SKYMAX 0.06
SamF      0.07

Swapped off DLM1
User        CRC
maxzzz    1
SamF      12
jjnz1       86

I also thought that my personal DLM history might be useful:

My DLM History
DLM CRC
3    0.04
6    0.05
2    0.07
1    12
2    0.39

1 | ... | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | ... | 285