* Ask Bjørn Hansen <[email protected]>

> There are only 8 countries with more than 20 IPv6 servers and 17 with
> more than 10. For IPv4 the numbers are 21 and 35. The load sharing
> system works much much better with more servers. Maybe there are
> enough IPv6 servers, but just focusing on the percentage of users vs
> the percentage of servers doesn’t necessarily cover the trade-offs.

I do believe there are enough IPv6 servers in the NTP pool, yes. Fewer
than 1 out of 10 users on the internet are capable of using IPv6
(source: http://www.google.com/intl/en/ipv6/statistics.html), so it's
not like enabling AAAAs will cause a mass migration on the spot. In any
case, about 3 out 10 pool servers are IPv6 ones, so it should be
sufficient even if every single IPv6-capable user does migrate to IPv6.

Of course, as Dag Erling rightly points out, many of these are probably
dual-stacked servers that simply appear twice, one in each list. In
that case, I guess it makes no difference which IP version the client
is using.

> The other (maybe bigger) issue is that the client support for IPv6
> is, to me, uncertain. At a minimum I want to have an option for the
> vendor zones so they can enable and disable IPv6 as needed if they
> know their clients don’t do the right thing (whatever that is — as
> was pointed out in the thread here even ntpd isn’t obviously doing
> the right thing).

Yep, identifying and fixing client bugs is obviously very important. Do
you have any data about how the dual-stacked 2.pool.ntp.org service has
performed in this regard? Has it uncovered any showstoppers, or has it
for the most part not caused any problems?

> As mentioned in another thread I’m planning to change how the various
> zone names work to better support countries with few servers. When
> I’ve added the vendor zone options I might also enable more of the
> sub-zones for IPv6 at that point.

Sounds great! Slow and careful progress is completely fine to me, as
long as there is progress. :-)

> If you want IPv6 for your ntpd instance, just use “pool
> 2.pool.ntp.org” as the configuration now. On ntpd 4.2.6p5 it will add
> both the IPv4 and IPv6 servers (so about doubling the load on the
> servers, in that instance anyway).

Ack, but I'm not too concerned with my own instances to be honest. My
interest in this is to help make IPv6 get to the same level of
production readyness as IPv4. The default behaviour of a public service
such as the NTP pool is important in this regard. The point is to avoid
situations where a random device connected to a random IPv6 network is
not able to get correct time, while it would be if it was connected to
a random IPv4 network.

That is, IPv6 must eventually Just Work(tm) to at least the same extent
that IPv4 currently does. Otherwise IPv6 is pointless because we'll end
up being stuck with IPv4 forever. There's lots of work to be done here,
so it would be great to get the "dual-stacking the NTP pool" off the
list of outstanding issues. :-)

Tore
_______________________________________________
pool mailing list
[email protected]
http://lists.ntp.org/listinfo/pool

Reply via email to