IPv6 mistakes, was: Re: Looking for an IPv6 naysayer...

Fri Feb 11 17:20:59 UTC 2011

> On Fri, Feb 11, 2011 at 10:00 AM, Scott Helms 
> wrote:
> > Agreed, V4 traffic levels are likely to drop and stay at low levels
> for
> > decades.
> 
> I seriously doubt v4 traffic is going to fall off a cliff.  That would
> require IPv6 adoption on a large scale over a relatively short period.

The thing is that a very few networks account for a very large amount of
traffic.  So it depends on what you mean by "adoption on a large scale".
<1% of the networks account for >50% of the traffic. If a handful of
networks move to v6, then we have a very large amount of v6 traffic and
a significant decrease in v4 traffic.  It depends on if you mean large
numbers of different endpoints or large numbers of packets when you say
"adoption on a large scale".

Joe's Fish Farm might and all the other Fish Farms might stay v4 for a
decade or more but that traffic accounts for an insignificant portion of
traffic in the context of internet traffic as a whole.

>  To date, nothing in the v6 verse has happened *quickly*.  Replacement
> or software upgrades to millions of CPEs in hundreds of network is not
> something that will happen overnight.

What is the natural "churn" rate for CPE for one of the large MSOs? What
portion of the MSOs have v6 capable CPE in place right now but v6 just
isn't in use but is planned to go into v6 service soon?  You don't need
to migrate "hundreds" of networks to account for the majority of eyeball
internet traffic in North America, you only need about five.  It could
be that v6 capable CPE has been in the process of being rolled out
already and has been for months to possibly years.

> Even then, that will not
> instantly switch everyone and every device to IPv6.  How many
> "connected" devices do you think there are in the average home?  TV?
> DVR (stb)? Game console(s)? Netflix streaming thing?

Ok, we have been watching our DNS servers for who is requesting AAAA
records.  The vast majority of our connections come from a very small
number of networks.  We are seeing requests for AAAA records.  The next
step is to put a v6 only DNS server into whois but hand out only A
records for a while.  But the idea is to see what of the requests for
AAAA records actually arrive via IPv6.  Once we profile that for a
while, we will return AAAA records for the largest requester but only
for requests arriving by IPv6 requesting AAAA records.  The next step is
to see that the requests actually result in connections to the service
address handed out by the AAAA records and let that "bake" for a while
and see if any service oddities are noticed.  We happen to be in a
unique position in that requests from different remote networks request
a unique service address for that remote network and most others don't
have that luxury.  So if one remote network is v6 clean, we can change
one IP address to AAAA records and "migrate" that remote network to v6
without impacting others simply by changing the DNS record for their
service IP.  If another network has issues, is requesting AAAA records
but can't really talk over v6, we can roll back to A records for that
service IP associated with that particular remote network.  Other
providers don't have that luxury and I understand that.  But still, once
two of those remote networks switch to v6, that is a very significant
portion of our traffic.  It will be possible, depending on which remote
networks migrate and at what speed, for traffic to migrate in "chunks"
as we migrate those AAAA records.  We might go from 0% of traffic on v6
to 25% of traffic on v6 in less than a calendar quarter depending on the
behavior of the remote nets.  Also, once THEY see more successful
traffic migration to v6, it gives impetus for them to move faster in
that direction for additional services.

> 
> >> Facebook ... So that's 55% of Internet traffic right there.
> 
> and making a dent in it means residential transition.  50mil (or
> 83mil) devices is a lot of shit to replace or reprogram.  Not to
> forget the thousands of devices that feed them.

Yes, and I mentioned that.  So once you have >50% of the potential
content sources v6 capable and >50% of the potential eyeballs v6
capable, you have potentially 25% of internet traffic on v6.  And that
can be done with the migration of enough networks to count on your
fingers.  So again, are we talking number of networks or number of
packets when we say "large scale adoption"?

> > ... mobile devices
> 
> i.e. cellphones... the two largest groups there (iPhone and Android)
> support IPv6 already. (in certain versions)

And are already being given native v6 IP addresses in some markets.
Some markets are already doing NAT64 or something to get these devices
to v4 content.

George