Reply to comment:

George MIchaelson
Thanks for this. We have difficulties understanding why the skew in South Korea is so bad. We have known for some time the distribution of adverts to ISP in this economy probably doesn't reflect the true in country distribution (we have been told by research scientists from KISA our figures are off) and we also suspect some dynamics of the Korean peering behaviours distort the measurements. (We are told a lot of domestic IPv6 is not well peered or lacks transit.) SK Telecom are said to have significantly more market share than we impute. I would love to discuss this with Google advert placement specialists but it's a pretty opaque placement model inside the economy. Recent under-measurements in Thailand related to intermediate transit issues to our linode which have just been resolved. Black holes in IPv6 routing are often hidden by dualstack connection behaviours but in our case if an ipv4 served client attempts to fetch our configuration for adserve over IPv6 with an invalid route there is no fallback and we can lose samples. It is interesting to consider the akamai figures too. I tend to characterise sonic as an under-measurement and others as optimistic over measure!