IPv6 support on Virgin media

Sunday update --- and the rise in VM IPv6 counts continues, APNIC permitting.

If anyone is counting the weeks between my (usually) fortnightly graphs, they'll have noticed that the last one failed to appear on schedule. There was a good reason for that: APNIC had a little hiccup on the 19th January, and its data became questionable.

Being an engineer, data capture and analysis is stock-in-trade, and so is watching out for data anomalies. One such anomaly is common-mode changes in variables that are considered independent, which usually indicates that there is a problem with the data capture or its transmission. In some situations it is possible to perform Common-Mode Rejection to recover the wanted data safely by removal of the common-mode contribution, as long as certain conditions are met. I gathered another week of daily samples to check, but unfortunately the required conditions are not met. We're going to have to live with this glitch in the curves, a valuable reminder that indirect proxy measurements are often less robust than direct ones.

The following graph shows the signed magnitude of the increments in each day's sample value relative to the day before, for all four of the UK ISPs monitored, starting one month before the anomaly for reference. Virgin Media's curve appears as a straight line hugging the x-axis because VM increments are dwarfed by the common scale and much larger increments of the other three, but the same anomaly is apparent in VM data when examined at a smaller scale.  Namely:  on 19th January 2019, APNIC's IPv6 counts for all four ISPs dropped simultaneously by a large factor (negative increment), and remained low for four days:

By itself, this large drop would not have prevented Common-Mode Rejection and data recovery, but unfortunately after the equally sudden rise 4 days later, none of the curves showed that the increments masked by the 4-day common-mode drop had accumulated to provide continuity after the common-mode fault ended. As a result, continuity is missing and cannot be recovered by interpolation across the fault. We are in effect in a new data sequence now: the measurement apparatus has changed.

Here is the current graph for all four ISPs, visually suggesting a common-mode fault and loss of statistical continuity with the data before the event:

And finally, our usual plot of Virgin's IPv6 activity at finer resolution:

Fortunately the continuity remaining in the curves is easily sufficient for eyeballing purposes (it could have been *much* worse), so APNIC's hiccup hasn't altered our ballpark analysis. IPv6 activity is clearly still at the highest level ever seen in Virgin's AS5089 and is still growing, so we can be confident that something interesting is happening.

Today's IPv6 sample count for VM stands at 20,531 per day.

Morgaine.