> On Mon, 04 Jan 2016 14:17:56 -0800, Owen DeLong said: >> Further, 8.8.8.8 actually fully supports EDNS0 Client Subnet capability, so >> if the geo-IP balancer in question wants, they can eliminate the failure mode >> you are describing in that case. > > Which only helps for people using 8.8.8.8. Client Subnet does help the issue, > but it doesn't actually fix it until support is near ubiquitous across > intermediate nameservers that have clients in other geographic locations…
Well… that and any other DNS server that supports EDNS0 client subnet. > > (I believe that the fact that Google found a need to create EDNS0 Client > Subnet *at all* is proof that using the DNS address for localization is > problematic…) Sure, but anycast is even more problematic and those are basically the only two alternatives currently known for solving the problem in question. > > And again - it's still something that needs work upstream to support, and > you *still* have to deal with the case where the intermediate DNS server > doesn't do Client Subnet. Or accept that no solution is perfect, make this one as good as we can for now and move on. > >> I say slightly pessimistic because there aren’t all that many 3XX responses >> being reported. > > OK, that's a slightly different kettle of fish :) To the nearest 10% or > so, how many are answering with a 3xx of any sort? Well… I’ll post a separate message detailing my findings. It’s more interesting than I previously realized because none were reporting 3xx results and all 3xx results were getting hidden behind 5xx results which weren’t (all) entirely valid. Owen
