SNR Target? stick or try to lower?

[#1]

Howdy all,
I was considering dropping my SNR target so decided to monitor the connection(i used to have a 3db profile many moons ago with my DG843 V3 but now im stuck with a 9db using DGND3700 oh how silly was i...)
Never the less i noticed when i first got into busy box that i had lots of ES and HEC's

Since Link time = 18 days 12 hours 50 min 19 sec
FEC:            0               0
CRC:            27586           3913
ES:             13776           3404
SES:            171             0
UAS:            0               0
LOS:            0               0
LOF:            0               0
LOM:            0               0

This was over 18 days so decided to reboot and get a fresh full log,

Since Link time = 1 days 48 min 52 sec
FEC:            0               0
CRC:            3212            239
ES:             1813            217
SES:            3               0
UAS:            0               0
LOS:            0               0
LOF:            0               0
LOM:            0               0
~ # 

Now i know this is a fairly large snapshot again and there doesn't seem to be alot of errors(i have seen it much MUCH worse when i had a line fault back when i first signed up pretty much 9-10 years ago)
I believe interleaving is off because link depth shows as "1" which is probably why errors show
My question is if i were to drop the profile are many more errors likely to creep in and in turn have a much worse effect? it would appear i cant go lower than a 6db profile as i seem to vary +/- 3.5 SNR but i also noticed through running router stats that there appears to be a steady pattern to CRC error numbers is this normal? the pattern seems to drop really low after midnight even though connection is used throughout night.
Currently looking to get shot of this modem and change from netgear all together maybe see if a different one can help stop such a large variation in SNR...
Any advice would be helpful, hopefully haven't made a fool of myself this time 

[#2]

I'm not sure what the graph is supposed to show, the pattern just seems to be the CRC count going to zero every 15 minutes, but the label says it's the CRC count in the "latest 15 minutes". I think the pattern is just something wrong or unclear with how the graph is being drawn.
If the pattern drops really low after midnight - that's good isn't it? Low CRCs.

[#3]

That's what i initially thought, however it doesn't seem to reduce every 15 minutes though if this is true it simply shows a fairly consistent increase in errors during peak time and significantly less in night?
After midnight the connection is usually used more so wouldn't we see more errors rather than less? 
I'm not overly familiar with the software or what the tolerances are.. and i don't want to mess up and go back to 5MB profiles...

[#4]

I think if the CRC rate decreases after midnight, that would be because most people have switched off various electrical items and gone to bed, so the level of noise decreases. I used to think the error rate depends on how much you use the connection, but now I think that's not the case.
Obviously the DLM isn't going to reduce the target SNR margin, not without interleaving anyway, or it would have tried it eventually. Lowering the SNR margin probably would increase the error rate and increase the range that the SNR varies over the day, but it's difficult to know exactly how much they would change by. If you tweak the SNR down yourself, the DLM might then need to increase the margin it sets.
Perhaps it would be more useful to graph how the SNR varies over a day, or graph the number of errors in the last 1 minute (not 15), and see it any changes correspond to certain things being switched on and off, and how much things change when the sun sets.

[#5]

If what you are plotting is a variable summed over the last 15 values then it is mathematically impossible for it to suddenly drop to zero.
For example, say the counts in each minute were
1, 1, 1, 1, 1, 1, 2, 1, 2,1, 1, 1, 1, 1, 1 , 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
Then the summed counts would be:
1, 2, 3, 4, 5, 6, 8, 9, 11, 12, 13, 14, 15, 16, 17 , 16, 15, 14, 13, 12, 11, 9, 8, 6, 5, 4, 3, 2, 1,0
Any real sudden change would become smoothed out over 15 minutes by the integration process.
I imagine what you are really plotting is the number of counts in the last 15 minutes but with the total count set to zero after 15 minutes.  So the 15 minute periodicity is just an artefact and your variable is decreasing over time as indicated by the decreasing height of each triangle.

[#6]

Quote from: ReedRichards

I imagine what you are really plotting is the number of counts in the last 15 minutes but with the total count set to zero after 15 minutes.

The counts in question are from the start of a 15 minute window, not necessarily the latest 15mins. These counts are reset at the end of the period.
RouterStats only has the option of plotting 'Errors during latest 15 minutes', 'Errors during latest 24 hours' or 'Errors since last reboot' and obtains its data directly from the router.
The set of available error counters are typically

Total time = 1 days 13 hours 11 min 31 sec
Latest 15 minutes time = 11 min 31 sec
Previous 15 minutes time = 15 min 0 sec
Latest 1 day time = 13 hours 11 min 31 sec
Previous 1 day time = 24 hours 0 sec
Since Link time = 8 hours 59 min 30 sec

The times quoted are normally the time within the period that is covered by the set of counters

[#7]

Aye, This sounds more representative, some were about 10 mins long others were 20.. though still not sure if errors will drastically increase on the line or if its best leaving it until i grab a better router
My old DG843 had line attenuation down between 40 and 41 but this router holds out at 43 i cant help but think a decent router & Drop in SNR(if possible) may help get at least an additional �½ - 1 MB out of my profile with a stable line. I'm resigned to the fact that even if we do get fiber here in the next decade it wont make the 3KM trip to update 1 cab so i may as well make the best of this line as i can...