Broadband Usage Graphs - something's not right

[#1]

http://www.plus.net/support/service/network_performance/broadband_bandwidth_usage.shtml
There's something wrong with the Total traffic on the network graphs for WBC in that the maximum figures don't match the graph. E.g.

That's clearly peaking at over 12G but the Max in the totals below is nowhere near that. I'd guess that figures for some gateways are included in the graph but not the totals below. The 20CN IPSC graphs look OK.
What is also strange is the ratio of upload compared with download. Colindale WBC is a lot higher than Faraday WBC and both of them are much higher than the 20CN IPSC.
Here's the rest of the graphs:

20CN


NB All of the images in this post are links to the live graphs - not snapshots.

[#2]

I wonder if the uncapped uploads are having an effect on 21CN as opposed to 20CN but it does seem a bit extreme

[#3]

Many 20CN business users will be on Max Premium and there's the residential users like myself who pay the extra for the 832 kbps upload on Max Premium.
OK, with the uncapped uploads they will be faster, but I can't think of any reason why 21CN users should upload more data. If anything I'd have expected it to be the other way, data coming down faster and then discarded (e.g. YouTube videos not completely watched).

[#4]

I was thinking more about things like P2P which aren't speed limited on uploads

[#5]

What proportion of users use P2P? I'd suggest it's way, way lower than those who watch things like YouTube and iPlayer.
I'd also suspect the majority of users who were put on a capped upload don't know they can ask for it to be uncapped and I don't believe PN have been doing it without asking. It's not that long ago the policy changed to provision uncapped.

[#6]

I can see a darker green line in the WBC graphs that corresponds to the numbers in totals - in the Colindale WBC graph it's just above the blue, in the Faraday WBC graph it's nearer the top of the green.

[#7]

I suspect that's more to do with the way the graphing is done, rather than having any significance. I'm sure Kelly will put me right if that's not correct.
I wonder if the differences between the graph and the figures has something to do with the move to the new E100 Traffic Management platform?

[#8]

I wondered that, but the IPSC connect graphs look right and some of that goes through E100's.

[#9]

I hope Kelly and co. haven't forgotten the comments and suggestions in http://community.plus.net/forum/index.php/topic,105696.0.html

[#10]

Right,
There certainly is some oddness in the totals, so will look into the config tomorrow of those graphs.
Ejs is right, you will see some darker green lines, but these are not what is being reported on in the totals. These are actually image artefacts, as a result of the way multiple datasets are compiled together.
In unit form, the total value of the data sets combined is meant to be the text output, but in graphical form each data point is plotted, with the next point stacked on top. RRDtool causes these artefacts in the stacks display.

[#11]

Morning all,
I have found the source of the problem with this graphing. It appears that the system that generates them is doing something odd with some of the values that are input.
The legend and values contained are accurate if my reading of RPN math is correct. It is certainly doing math against the correct data sets.
It is just the stacked lines in the graph that are incorrect. It seems to be double counting some data sources, and completely not mapping others. It's actually quite odd.
Problem 72160 has been raised on this.
In regards to the colours in the graph, it is configured to have a 60% "Opacity/Alpha Channel". The darker artifacts are the overlapping plots.