@davefiddes:
I think I've discovered what's wrong with the graphic you linked. Take a look at the image below, and compare it with your curve after you've set the Average Interval (days) to 1 day, and clicked on the 3m (3 month) button to place the start of October 2018 just a little to the left of centre. It's not just vaguely similar, it's almost exactly identical! Well guess what I plotted --- oh, you know already because I described it in the caption, it's not the IPv6 counts but the rate of change of IPv6 counts, namely their first derivative!
I'm including below the daily samples from mid-September onwards so that you can match the peaks and troughs yourself numerically. To make this easier I've included an additional column, "Increments", which makes any daily excursions from steady growth very easy to spot. The graph above is simply a plot of the Increments column from 2018-09-30 until today.
This almost perfect match is far beyond any possibility of accident --- they either intended to plot the rate of change or else they have a first derivative term in their plotting code and triggered it through some programming error.
It does at least give us a chance to check the time-sync between their graphics and my CSV tables taken from APNIC's Javascript. The cleanest place to check is the start of October, because it was preceded by a long period of relative quiet, and so a sudden increase will be equally distinguishable in both the counts themselves and in their first derivative. And sure enough, my counts start to skyrocket with the first large increase on 2018-10-04, and their graph spikes too, hence confirming that we are in time-sync on 4th October. Well you can't be in sync on one day and not in sync later on in a daily sampling sequence, it doesn't work that way. 😛
In summary, I'm completely convinced that we're both looking at exactly the same data. They've just done something really odd when trying to plot it, either accidentally or by intent.
PS. The fact that despite time-sync at the start of October, the daily counts rise monotonically for almost 2 whole months, from around 2,000 to well over 8,000, and yet that isn't reflected in the graphs you linked, highlights that they're doing something very wrong. Given the time-sync, you can't make one half of a 2-month monotonic rise vanish under any processing I know, unless you're plotting the derivative and so the gradient goes negative as the growth slows down a little.
Morgaine.
Since the graph that @davefiddes linked at https://stats.labs.apnic.net/ipv6/AS5089?a=5089&c=GB&x=0&s=1&p=1&w=30&s=0 will change continually and could lose the extremely close correspondence to my plot of the first derivative of the APNIC daily counts, I'm recording a copy for us here:
It's the portion from the Oct 2018 line just left of centre towards the right that matches very closely, as this is the 3-month view. You can match the peaks and troughs, one by one. 😛
Sunday update -- well the IPv6 counts sure are shooting up!
We're in uncharted territory now, beyond the peaks of IPv6 activity ever seen before by APNIC for AS5089. Although we can only guess how to interpret these figures, two things seem clear:
• These numbers are not yet signs of the release of a public IPv6 service, since they are too small by one or two orders of magnitude, depending how release is done. At the same time it should be mentioned that the counts are far above the noise level in the data, so they do indicate something with statistical significance. We just don't know what.
• The growth in these daily counts is pretty linear, which I interpret as some kind of ongoing equipment rollout process, because that is one of the things that is commonly linear in growth since it is limited by human manpower. For example, it might be the number of deployment engineers available to visit equipment installations around the country that is constraining growth in the APNIC counts to a near-constant amount.
If it is indeed some kind of resource-limited deployment process that is responsible, then that deployment will eventually come to an end, and it's not unreasonable to guess that we might be due for it any time now. After all, Virgin's network won't have grown too much since the end of September 2017, and the similar growth rates now and then do give some confidence that there is no significant new multiplying factor involved.
Despite the grim reality of last weeks' posts, so strongly anchored in the very bleak precedent of Virgin's past actions, at least things are moving in the right direction.
Morgaine.
