Le 1 avr. 2010 à 17:31, Tore Anderson a écrit : > Hi again, > > * Rémi Després > >> These 95%+ problems, being related to 6to4 and Teredo, don't concern >> ISPs that offer native IPv6 addresses (independently from whether >> they offer them with 6rd, like Free did in 2007, or otherwise). Note >> that this is consistent with Free's users not encountering such >> problems. > > Indeed, providing end users with native IPv6 will make the problematic > transitional techniques obsolete. However, this is not something that > content providers such as Yahoo or myself have much influence over.
At least, they should: - get native IPv6 addresses - make sure they advertise both a native-IPv6 and 6to4 addresses. - transmit IPv6 packets not exceeding 1280 octets. Note also that the proposed hack, which is in ISP DNS servers, in not more in Yahoo's hands. >> Would you have some information about the 5%- remaining losses? > > I haven't been able pinpoint any common denominator(s). I have some > ideas, but it's mostly speculation. In no particular order: > > 1) Higher latency with IPv6, less developed peerings, etc. If a client > navigates away from the test page exactly when the IPv4 PNG has > loaded, but while the IPv6 request is for instance 50ms away from > completion, he will be (falsely) counted as a lost client. > 2) Linux clients with 6to4/RFC 1918 as described in my previous message. > 3) Users with CPEs/SOHO routers that do "evil things" to DNS responses > containing AAAA records. > 4) Tunnels (HE, SixXS, etc.) that the user configured and forgot about, > and that broke at some point for some reason (like his IPv4 address > changed). > 5) Software that (like Opera did earlier) does not behave according to > RFC 3484 and prefers transitional/any IPv6 connectivity over IPv4. > > In any case it's hard to determine it accurately. Also I think I'm > approaching the statistical margin of error (on some days the client > loss measurement turns out negative). Thanks, I will think about these. > Oh, and by the way - #3 in the list over could conceivably be a big > problem for Yahoo or other global players, since there could be huge > geographical differences. For instance, if an ISP in some other country > far away from here distributes have distributed a CPE to all its > subscribers that breaks AAAA responses, it would hardly be noticable for > me. But again, that is pure speculation. > > I'll be sure to let you know if I do discover more problems, though. Thanks. >> A "way of knowing" that doesn't work all the time is not a real "way >> of knowing", right? > > I'd call it a heuristic estimation or something like that. Right, and to increase operational costs, nothing is better than heuristics, especially if they are supposed to solve ill identified problems, and are known to create new ones for sure! Regards, RD _______________________________________________ DNSOP mailing list DNSOP@ietf.org https://www.ietf.org/mailman/listinfo/dnsop
