►
From YouTube: IETF104-V6OPS-20190328-1350
Description
V6OPS meeting session at IETF104
2019/03/28 1350
https://datatracker.ietf.org/meeting/104/proceedings/
A
A
I'm
going
to
Mickey's
mouth,
you
know:
Mickey's
handed
no
sticking
up
so
I'm
going
to
start
with
a
little
bit
couple
of
questions
and
some
thoughts
that
might
be
useful
to
you,
guys
I
hope
they
are
about
every
IETF
meeting,
I
wind
up
explaining
to
somebody
how
Ron
and
I
choose
our
agenda.
So
let
me
walk
through
just
a
little
bit
how
we
choose
an
agenda.
A
This
thing
that
you're
looking
at
is
a
little
file
that
I
have
a
script
that
I
put
together.
It's
the
same
information.
That's
in
the
data
tracker,
it's
organized
in
a
way
that
I
find
useful.
If
somebody
cares
I
can
make
it
available
to
them.
But
but
basically
you
have
a
list
here
of
the
internet
drafts
that
we
have
in
the
working
group
and
what
their
status
is
as
of
about
eight
o'clock
this
morning,
so
we've
got
one
that
the
RFC
editor
is
banging
away
on.
A
We
have
a
working
group
document
that
didn't
get
updated
since
the
last
I
kept
meeting
and
so
on
and
so
forth.
Various
individual
submissions
that
did
or
did
not
get
updated
him
since
last
meeting
what
Ron
and
I
wind
up
looking
for
his
documents
that
are
somewhere
in
the
working
group,
which
is
to
say
working
group
documents
or
individual
submissions
that
have
been
updated.
You
know
so
that
wind
up
being
a
subset
of
the
drafts
that
were
looking
at
here
and
which,
by
the
way.
A
So
what
we're
looking
for
is
an
updated
draft
or
one
one
that
is
new
since
last
night
cap
meeting
and
for
which
the
comments
that
we
see
on
the
mailing
list
first
off,
we
see
comments.
Can
that
they're,
supportive
that
that
people
are
saying
this
is
something
that
I
wanted
to
play.
My
network
I
think
this
looks
like
a
good
idea.
Can
we
chase
this
down
can't?
A
And
so
you
know
so
those
are
the
the
drafts
that
we
wind
up
looking
at
in
meetings
here,
fYI
I'm,
pretty
sure
most
of
you
heard
that
speech
once
before,
but
but
depth.
That's
how
Ron
and
I
do
it
where's
Ron?
Well,
we
have
virtual
Ron.
Ron
is
sitting
next
to
me,
but
he's
often
some
other
meeting
right
now.
A
A
So
this
is
the
agenda
that
we
have
this
morning.
We
had
one
guy
that
came
in
and
he
wanted
a
asked
for
time
this
morning
and
he
forgot
to
ask
and
so
I
included
him
as
any
other
business
and
that's
actually
an
opposite
draft
they're
asking
for
review
operational
review
on
it.
So
that's
there,
but
but
listed
as
AOB
you'll
notice
that
there's
one
draft
that
there
has
been
discussion
on
the
list
in
the
last
couple
months
that
isn't
on
the
agenda
reason.
A
The
ten
at
sixty
four
recommendations,
deployment
recommendation
deployment,
guidelines
draft
so
I'm
willing
to
do
that.
Do
you
guys
think
that
let
me
just
ask
for
a
sense
of
the
room?
Do
you
guys
think
we're
ready
to
go
to
work?
Need
replies,
call
on
that.
If
so
come,
if
not
come
okay,
so
fine
I'll
ask
for
a
working
group.
Last
call
Hey.
It
probably
won't
do
it
next
week.
I'll
probably
do
it
in
two
weeks
and
we'll
do
that.
A
D
A
Want
data:
okay,
Jen,
wants
data,
we
keep
hearing
or
we
hear
from
time
to
time
that
operators
don't
show
up
here
and
funny
thing:
we've
got
an
entire
operational
area.
We've
got
operators
that
show
up
in
various
working
groups
in
there
she's
just
kind
of
curious.
How
many
people
in
this
room
represent
operators
or
how.
A
G
You
person
standing
at
the
mic,
Jarrod
much
come
on
networks.
We
run
a
network
amongst
other
things.
If
people
are
looking
for
data,
it
would
be
helpful
if
somebody
could
quantify
that
and
there's
I
think
plenty
of
research
opportunities
and
collaboration
opportunities
available.
So
if,
if
people
can
identify
what
sort
of
information
they
want,
you
know
we
certainly
publish
data
about
what
we
see
in
the
ipv6
space
and
we'd
be
interested
in
publishing
more.
If
there's
something
that
is
of
interest.
Okay,.
A
A
H
I'm
Michael
and
to
quantify
I'm
speaking
for
DTS
we're
running
a
project
not
necessarily
like,
what's
in
the
production
Network
at
this
point
of
time,
the
largest
one,
but
we
have
been
doing
at
36
for
quite
a
long
time
in
in
this
project.
H
We
have
get
lab
and
we
have
quite
a
few
other
things
that
I
interact
with
on
a
daily
basis
and
I'm,
also
going
to
mention
some
things
that
I
have
noticed
coming
up
in
various
activities:
six
forums
where
people
come
and
ask
questions
like
I
have
problem.
I
want
to
do
this.
Why
could
and
so
on?
I
spend
a
bit
of
my
leisure
time,
helping
people
you
know
run
into
be
successful,
so
just
an
introduction
to
the
the
what
we're
doing
here
in
the
project.
So
the
I
need
to
choose
one
to
point.
Okay,
so.
H
Okay,
so
will
you
run
like
a
v6
only
here,
there's
a
lightweight
four
over
six
tunnel
going
from
here
to
the
data
center,
and
that
provides
the
v4
connectivity
and
then
with
its
regular
dual
stack
here
and
then
we
are
dual
stack
over
here,
but
our
data
center
is
basically
only
ipv6
for
most
of
the
services.
H
So
one
of
the
problems
we've
seen
over
time
has
been
bugs
or
misbehavior
in
the
relays
of
two
of
the
vendors
that
were
using.
We
have
four
different
vendors:
we've
spent
a
non-trivial
amount
of
time,
debugging
various
dhcpv6
relays
and
the
states
that
they
need
to
keep
in
order
to
do
the
routing
from
from
here
to
here.
H
So,
like
the
the
the
link
local
next
hop
route
for
the
delegated
prefix,
so
some
of
them
need
to
keep
state
here,
and
this
has
not
been
trivial
for
some
vendors
to
get
right
to
specialist,
since
we
actually
do
three
prefixes
here
that
are
algorithmically
tied
together,
which
has
been
one
thing
that
I
guess
the
implementations
here
didn't
really
expect.
So
the
client
asks
for
three
PD's
of
difference:
I,
don't
remember
what
the
identifier
is,
but
there
is,
there
are
different
identifiers.
H
We
have
machinery
here
in
the
DSP
v6
server
that
that
uses
this
and
it
gives
them
three
different.
Prefixes.
H
So
we've
seen
realize
that
act
as
DHCP
servers
and
sentry
rejects,
and
they
didn't
even
relay
it,
which
is
fun.
We've
seen
the
relay
eat
response
packets,
because
what
came
back
from
the
server
didn't
agree
with
what
the
thing
had
in
its
existing
state
table
and
there
are
other
very
interesting
behaviors
that
are
that
have
been
going
on,
and
so,
if
you
see
on
my
list
when
I
discussed,
like
v6
I'm
kind
of
PTSD,
I've
posted
my
distress
in
the
syndrome
from
from
having
to
interact
with
this,
there's
been
quite
a
frustration.
H
So
we
looked
into
this
I've
discussed
this
multiple
times
again
Ferrer
my
colleague
he
dug
up
this
and
I
talked
to
all
it
Terron
and
then
so
on
and
this
how
this
is
should
be
done
was
ever
never
specified.
Some
people
say
this.
Is
it's
obvious?
You
don't
need
to
specify
it.
There's
been
a
lot
of
discussion,
so
I
also
seen
others
that
have
gone
to
their
vendors
and
said:
look
we
want
to
team,
implement
the
prefix
delegation
relay
the
vendor,
says
sure.
H
Comes
back
six
months
later,
yeah
I,
just
really
like
it
just
doesn't
in
solarops
kind
of
meaningless
with
the
pieces
having
a
round
to
the
the
speedy
relay
and
doesn't
install
a
route
to
make
the
PD
actually
do
anything.
This
has
happened.
I've
heard
this
story
three
or
four
times
from
different
vendors.
I
Bernie
volts
just
a
quick
question
for
you
are
you,
you
know
what
what's
the
technology
you're
using
from
the
home
to
relay
or
whatever
you
know,
cuz
the
cable
environment
for
DOCSIS
has
gotten
this.
You
know
done
a
pretty
good
job
at
defining
this
stuff.
I
think
that
maybe
you're
using
BMG's
and
other
things
and
those
have
not
done
as
good
a
job.
Okay,
so.
H
This
there
is
one
VLAN,
there
is
a
one
broadcast
domain
and
then
it's
a
VLAN
interface
over
here.
It's
it
goes
on
tagged
here.
It
get.
The
ghost
on
tag
here
goes
into
these
nets,
which
then
comes
tagged,
and
so
then
there
is
one
sub-interface
per
customer
here,
which
is
so
there
is
one
broadcast
domain
per
customer
facing
port.
So
this
is
IP
over
Ethernet,
okay,.
I
H
H
Okay,
so
yes,
I've
seen
this
so
we've
had
problems
with
it.
Other
people
have
had
problems
with
this
right,
so
different
problem
spaces,
so
we're
using
different
prefixes
to
mean
different
things.
So
we
have
video
voice
and
best
efforts.
The
best
effort,
this
internet
access,
video
and
voice.
You
cannot
talk
to
the
internet
on
that.
That's
that's
not
part
of
the
internet
offering.
So
basically
we
do
not
want
known
aware
clients
to
configure
addresses
from
them.
So
we've
done
this
in
different
over
time.
H
In
different
ways,
we've
used
deep
CP
and
looked
at
the
the
identifiers
that
the
clients
are
using
and
say:
oh,
should
they
get
an
address
from
this
prefix
or
not?
So
we
are
currently
not
for
the
general
use,
we're
not
sending
PIOs
with
the
video
and
voice
prefix
to
own
like
to
general
devices,
because
they
will
just
have
leave
configure
themselves.
The
addresses
choose
one
and
potentially
doesn't
like.
A
H
H
So
did
this.
This
work
is
now
going
on
in
int
area.
There's
a
draft
they're
discussing
these
six
things,
v6
related
things
in
int
area,
is
very
different
from
discussing
it
on
p6
ops.
This
tells
me
that
people
who
are
very
interested
in
v6
ops
are
not
monitoring
into
our
inte
area.
In
the
way
they
should
so
I
urge
more
people
to
pay
attention
to
interior
there's
stuff
going
on
there.
There
is
actually
of
interest
that
is
relevant
and
you
should
monitor
it.
J
H
Into
this
in
the
data
center,
so
we
use
the
same
kind
of
multi
prefix
stuff
for
approach
to
do
have
different
management
in
one
domain
for
management,
one
for
like
internet
facing
things
and
one
for,
for
instance,
the
voice
service.
So
there
is
like
the
whatever
it's
their
home
indicate.
A
gateway
is
talking
sip
to
will
have
an
address
out
of
the
voice
prefix,
and
that
is
not
reachable
from
the
internet
and
we've
had
to
do.
H
We
had
to
beat
the
operating
systems
in
very
careful
manners
in
order
to
get
it
to
do
what
we
needed
to
do.
This
has
changed
between
operating
systems,
the
versions
of
the
operating
systems
as
the
different
network
managers
or
whatever
is
being
used.
What
they
do
so
this
has
not
been
optimal.
I
am
seeing
the
same
thing
in
other
contexts
where
the
so
the
Linux
kernel
now
can
do
source
based
routing.
H
So
if
you
run
opened
up
your
tea
can
do
that,
there's
just
a
lot
of
things
going
on
and
you
can
include
the
resolver
you
supply
to
it,
and
this.
This
also
applies
to
when
the
host
has
a
VPN,
for
instance,
which
should
you
choose
what
destination
address
you
choose
what
resolver
should
have
choose.
This
is
a
mess
we
should
work
more
on
this.
H
This
is
causing
me
pain,
daily,
I
would
say,
and
it's
not
just
for
me-
I've
seen
others
so
the
ula
even
mentioned
there
was
one
guy
I
was
a
help
trying
to
help.
He
wanted
to
use
Yule
A's
for
a
certain
for
some
some
traffic.
He
was
doing
that
66
on
the
exit
to
do
that.
Didn't
really
go
too
much,
but
anyway
he
wanted
his
devices
to
use
the
ula
address,
to
talk
to
the
internet
and
like
the
whole
source
and
destination,
address
selection
that
that
doesn't
work
on
all
devices.
H
So
we
can
discuss
all
the
didn't,
but
that's
what
he
tried
to
do
and
the
the
device
only
had
usual
a
address
didn't
have
a
give
you
a
address
and
it
didn't
want
to
use
it
this
the
v6
space
at
all,
and
then
he
could
change
the
guy
conf
on
some
operating
systems.
But
some
didn't
have
this
option,
so
you
couldn't
even
beat
the
operating
system.
He
said
he
looked
into
Android
and
in
Android
you
just
couldn't
do
this.
H
H
Turn
put
give
the
hosts
some
kind
of
policy
like
a
hint
from
the
network
like
please
do
this
I,
don't
think
we're
even
close
to
everything
that
we
need
in
order
to
help
the
host
to
do
what
it
needs
to
do
in
in
certain
you
know,
situations
this
is
also
like
PVD
and
permission.
You
know
domains
connected
again,
I
would
say
the
part
of
this.
Is
there
a
here
as
well
any
to
you
so
we're
one
of
those
radical
people
who
want
to
do
empty
empty
you
non
positive
to
to
everything.
H
Currently,
today,
within
the
home,
we
only
do
1500
because
the
some
of
the
some
devices
I've
seen
so
many
different
ways.
This
is
handled
some
some
devices.
The
wall
like
in
the
Nick
driver
we
compiled
without
jumper
frame
support,
so
the
device
will
be
1500,
I've,
seen
cases
where
a
device,
the
kernel
thinks
the
device
does
at
9
K
when
it
receives
the
packets,
that
is
over
2300.
The
package
silently
dropped
on
Wi-Fi.
As
far
as
we
can
tell
the
standard
only
allows
for
2300
byte
packets.
H
H
That
was
this
line,
so
this
is
also
a
case
in
internet
exchange.
Points
like
some
people
don't
want
to
enable
Java
frames
on
their
devices.
Others
do
you
can
have
to
I
spaced
on
one
exchange
jumbo
frame
traffic
while
some
don't
so
this
means
again
you're,
mixing
and
differently.
Configured
MTU
on
the
NIC
and
M.
Are
you
so
you're
silent
to
dropping
packets
if
you're
sending,
because
this
is
typically
done
together,
so
the
maximum
receive
unit
and
the
maximum
transmit
unit
is
usually
follows
each
other
on
a
lot
of
platforms?
H
H
So
when
people
are
saying
you
know,
the
only
problem
is
that
all
ISPs
are
stupid
and
filtering
ICMP.
There
are
so
many
pathological
cases
where
this
doesn't
happen.
I've
seen
it
like
in
my
20
years
them
to
like
deploying
an
MPLS
network.
There
were
so
many
MT
related
problems
in
the
beginning.
It
was
ridiculous,
and
this
is
still
a
problem,
still
not
solved
so
there's
work
in
on
packet
level
of
Pithom
to
discovery,
work
it
on
in
transport,
right,
T,
3wg,
that
is
applicable,
okay
and
then
we
have
the
Richard.
H
It
also
detects
if,
if
this
is
for
whatever
reason,
if
you've
been
moved
to
a
different
PNG
or
whatever,
you
can
figure
out
that
you
need
to
restart
basically
treat
this
as
an
interface
down
event
and
restart
again
do
solicit
or
whatever
you
need
to
do
in
order
to
establish
the
state
again
that
you
need
for
the
forwarding
to
work.
H
H
This
is
not
a
problem
for
PPP,
because
they're
there
keep
a
lot
of
timers
and
I
guess
session
identifiers,
and
so
on
so
I
mean
it's
not
only
that
nobody's
answering
you.
You
can
actually
tell
that
the
other
guy
has
lost
state
that
for
a
while,
within
minutes
rate,
I'm
looking
at
for
sure
he's
not
in
case
Barbra's,
said:
okay
Parvez,
not
commenting
so
I,
don't
think
people
have
mentioned
BFD
here.
The
problem
is
that
if
you're
running
VFD,
the
both
both
participants
need
to
understand
50.
H
So
the
current
proposal
here
is
to
actually
pin
yourself,
so
the
the
home
gateway
here
sends
a
packet
to
one
of
its
own
addresses
or
in
the
within
the
PD
space.
With
that
guy
MAC
address,
so
you
ping
yourself
there.
This
actually
works
a
lot
of
people,
so
the
Linux
stack
here
is
not
is
not
happy
to
receive
packets
from
the
outside
with
its
own
source
address.
But
this
is
the
thing
you
can
turn
off
and
so
we've
we've
done
tests
and
this
can
be
made
to
work.
H
This
works
for
v4
as
well,
but
let's
not
discuss
that
in
turn,
I've
also
had
had
ideas
about
using
I
mean
we
have
a
neighbor
on
reach
ability.
This
mechanism,
why?
Why
aren't
we
using
this
like,
and
this
also
could
trigger
the
whole
thing
about-
was
like
okay,
so,
like
I,
don't
know
if
I
mentioned
right
so
with
dhcpv6
and
the
Raa
and
endi
are
just
completely
decoupled.
So
what
do
you
do
if
your
default
gateway
that
you
received
a
pre
PD
from
becomes
unreachable
by
detection
of
nd?
What
do
you
do?
H
You
do
nothing
as
far
as
I
know
today
you
might
deprecated
your
default
route,
but
you
keep
the
prefix
and
say
lalala
I,
guess
if
you
do,
if
you
remove
your
default
product
at
least
you're
gonna
send
ICMP
unreachable,
but
it's
like
you're,
not
gonna,
say:
oh
I,
probably
should
you
know
try
to
on
the
sender,
router
solicitation
or
you
know,
ask
for
a
renewal
of
the
PD
or
you
know
check
it.
Is
everything
okay
up
there
I,
don't
know
it's
it's
I,
don't
think
we
have
defined
exactly
what
you're
supposed
to
do.
K
D
Jenko,
thank
you
very
much.
That's
awesome.
I
I,
totally
agree
right,
I,
think
you're
doing
such
an
unusual
thing
here,
I
think
terms
of
how
many
interesting
things
you're
doing,
which
most
of
the
v6
deployment
just
haven't
experienced
the
other
right.
That's
why
you've
seen
so
when
I
was
a
shoes
right
and
I
agree.
We
need
to
probably
do
some
kind
of
gap,
analysis
and
document.
What
needs
to
be
done
to
make
the
protocol
more
about
the
multi
variable
and
yeah
I'm
like
would
be
happy
to
help
and
I
agree.
D
E
D
H
There
is
one
more
thing
also
about
the
PV
DS
I
have
four
VMs
running
at
home.
Just
because
I
have
three
different
VPNs
I
need
to
talk
to
VPNs,
don't
interact
nicely
with
other
VPNs
running
on
the
same
host.
This
is
this
is
a
problem
that
many
people
must
have
and
we
just
haven't
solved
it
and
it
could
be
solved
by
the
PVD
work.
Yes,.
D
D
H
A
Well,
if,
if
we
have
something
that
tells
us
there
are
there's
a
parameter
and
you
know
might
have
five
values,
then
yes,
what
what
is
the
current
setting
or
what
is
the
appropriate
setting
might
be
a
basic
substance
if
you're
asking
for
a
parameter,
yeah
I
think
you
need
to
talk
and
Vava
noise.
If.
H
L
But
what
I
am
what
I
am
actually
meaning
is
implementations
in
the
sense
of,
for
example,
open
wrt
to
a
lot
of
these
things
correctly,
but
many
vendors
are
not
doing
well.
So
maybe
we
need
to
say
hey
what
this
document
is
saying.
Is
you
should
do
your
implementation?
Probably
this
way,
because
other
implementations
work
well
so.
A
M
H
H
Of
my
proposal,
people
I
think
we
have
discussed
this,
but
I
would
like
to
be
able
to
say
that
route
has
that
empty.
You
so
basically
to
say
my
land
has
this
empty
you
and
the
implicit
default
route
that
you
learned
by
receiving
the
RA
could
have
a
different
deputy
yeah,
because
I
mean
the
Linux
kernel
supports
it.
I,
don't
know,
but
other
operating
systems,
but
that
would
help
belong
I'm,
not.
M
M
H
If
we
should
do
that,
we
should
kind
of
come
up
with
the
best
common
practice
and
so
on
and
come
up
words
criteria
for
when,
should
you
consider
this
an
interface
down
event,
bring
it
bring
it
down
and
start
going
back
into
deep
CPV
SiC
solicit
and
you
know
restart
the
entire
machinery
right
right.
So
yeah.
M
A
long
time
ago,
this
was
an
issue
that
we
were
talking
about
with
the
the
RS
resiliency
stuff
it
was,
there
was
actually
three
parts
to
it.
There
was
a
resiliency
on
the
on
the
front
end,
and
then
there
was
like
re
arcing
every
after
you
haven't
heard.
Any
re
is
for
a
while,
and
we
never
did
and
we
did
differ
if
it
has
that
work.
So
so.
H
M
N
Hi
martynuk,
you
have
mentioned
at
the
beginning
of
presentation
that
dhcp
or
it
doesn't
always
put
her
out
in
the
routing
table.
You
have
similar
problems
and
when
we
used
just
a
DHCP
server,
which
also
doesn't
usually
include
there,
I
think
and
we
actually
seen
some
operators
sewing
this
through
the
hooks.
But
there's
also
add
row
before
when
the
devices
are
not
every
time
using.
Link
local
address
from
generated
from
mac
address.
So
I
just
wanted
to
mention
that
this
problem
is
also
with
the
DHCP
servers.
Great.
H
I've
had
that
happen
to
me
myself
as
well,
when
I
try
to
start
doing
this
ten
fifty
years
ago
and
I
still
see
people
joining
this
social
myth
in
the
IRC
channel.
I
mean
complaining
about
that
exact
thing
so
opened
up.
Ruti
has
a
lot
of
these
bindings
between
things
going
on,
so
the
PD
install
route
do
allocate
addresses,
and
so,
if
you
just
download
at
each
p,
v6
server,
it
has
none
of
these
hooks
that
you
were
talking
about.
Then
people
were
implementing
themselves.
If
you're
running
a
standard,
Linux
box,
yeah.
N
H
B
B
L
So,
to
give
you
a
little
bit
of
context
on
this
on
this
document
we
have
a
colleague
in
Argentina
Alejandro,
which
started
deploying
ipv6
in
his
network.
The
provider
name
is
Stella
Centro
and
they
are.
They
have
a
cable
cable
network
and
they
started
with
stool
stack
deployment,
but
of
course
they
have
the
problem
of
running
out
of
ipv4
addresses,
so
they
started
investigating
and
actually
starting
with
small
pilot
for
for
64
X
lat,
and
they
told
me,
hey,
I
have
a
problem
with
this,
because
I
am
a
big
operator.
L
I
have
CD
ins,
I
have
caches
from
different
providers
and
if
I
use
4
6
4
X
lat,
it
means
all
the
traffic
from
ipv4.
Only
devices
will
be
translated
by
the
not
64
which,
of
course,
it
happens
as
well
in
the
summer
networks,
if
you
have
tethering
and
ipv4
only
devices.
So
this
is
a
problem
that
it
has
been
here
and
we
didn't
realize
it
before.
So
he
asked
it
in
the
mailing
list
in
b6,
ops,
I
think
it
was
beginning
of
March.
I
need
a
solution
for
this,
and
I
was
thinking.
L
L
Okay,
so
I
keep
going
meanwhile.
So
so
the
thing
is
that
in
the
case
of
category
not
there
is
a
prefix
which
is
hundred
dot.
64
is
less
than
I
think
if
I
recall
correctly,
yeah
164
does
10
and
some
Syrian
provides
interfaces
to
their
customers
for
for
not
needing
this
traffic
to
be
translated
by
the
career,
Reina.
Okay,
so
I
was
thinking.
Let's
try
to
do
the
same
in
in
in
the
case
of
4,
6,
4,
X
lat
and
actually
it
may
work
as
well.
L
I
still
need
to
investigate
a
little
bit
more,
but
it
may
work
as
well.
In
other
cases,
that
use
also
sit
for
the
net
46
like
map
T
right.
So
this
is
a
typical
4
6
for
X
lat,
and
if
we
have
a
dual
stack
CDN,
it
will
be
somehow
connected
because
most
of
the
Syrians
today
and
caches
have
already
dual
stack.
It
will
be
connected
to
the
isp
network,
the
operator
and
also
to
the
ipv4
network,
they
sorry
too,
but
today
pv6
and
ipv4
networks
of
the
operator
right.
L
So
we
have
a
customer
which
has
lands
with
dual
stack.
We
have
the
CP
with
the
salaat
and
then
we
have
the
ipv6
only
network
and
as
an
option.
It
may
be
also
dns64
in
addition
to
the
regular
DNS
we
have
here
the
necessity
for
so
that's
that's
the
global
architecture
of
this,
and
this
is
very
similar
or
mopti,
but
in
that
case
we
have
here
a
border
relay
right.
So
what
happens?
If
we
have
an
ipv6
device
in
the
customer
lands,
everything
will
be
going
which
ipv6
so
they
flow
into
any
site.
Bb.
L
Six,
only
that's
perfect!
That's
an
optimal
situation.
What
happens
if
I
have,
which
is
very
common
today,
I,
add
I
have
actually
asked
at
Samsung.
Can
you
provide
ipv6
in
the
model
of
a
smart
TV
that
you
release
it
last
week?
No,
we
don't
plan,
it
will
be
a
new
model,
maybe
two
years
from
now.
So
that
means
that
we
have
set
the
boxes
and
a
Smart
TV
stand
all
kind
of
devices
that
will
be
IP
before
only
for
a
while
okay.
L
So
in
this
case,
if
we
use
the
standards
that
we
have
today
for
four
six
four
X
lat
and
map
and
so
on,
we
will
have
this
ipv4
traffic.
Here
it
gets
translated
by
the
salaat.
It
goes
to
the
net.
64
green
is,
of
course
ipv6
red
is
happy
before
and
then
it
go
using
the
ipv4
interface
of
the
dual
stack
CDN
okay,
so
we
have
this
kind
of
ipv4
to
ipv6
to
ipv4
flow,
and
we
have
yes
a
stateless
translation.
L
If
the
salaat
is
well
design
it,
and
then
we
have
a
stateful
translation
in
the
nest
64
now
what
I
am
trying
to
do
is:
let's
try
to
do
this,
because
this
boy
is
already
translating
us
to
ipv6.
Why
cannot
reach
directly
the
CDN
which
ipv6,
okay
and
an
obvious
possibility
for
doing
that?
Okay,
this
is
this
is
a
summary
of
the
of
the
different
situations.
We
have
the
situations
where
devices
are
using
ipv6
and
that's
optimal
situations
where
we
have
ipv4
only
devices
which
need
two
translations,
and
then
that
will
be
the
optimal
situation.
L
L
One
of
them
is
exactly
the
same
is
being
done
for
Carol
grenade
boxes,
so
using
the
private
addressing
space
which
in
this
case
is
not
private,
is
they
not
64
prefix,
either
they
well-known
or
the
network
specific
prefix,
and
then
confirm
the
interfaces
of
the
CDN
with
that
prefix
this
bulwark,
but
require
Gore
from
the
operators
of
the
Syrians
okay,
so
some
of
them
I
already
doing
for
APB's
484,
but
not
all
of
them
agree
in
doing
this
as
well
for
a
pv6.
They
said
this
means
work.
We
will
charge
the
customers
for
that.
L
Okay,
so
that's
set
by
one
of
these
operators.
I
have
not
been
able
to
talk
which
one
of
them,
even
if
I
ping
at
all
of
them,
didn't
got
too
many
responses
at
the
moment.
So
this
case
also
problem.
If
the
CDN
is
sitting
on
the
network
of
the
brighter,
that
will
work,
but
what
happens
if
the
city
omission
and
IX,
because
the
ISP
don't
have
sufficient
traffic,
we
cannot
run
the
not
64
or
as
in
theory,
we
should
not
do
that
through
the
connection
with
the
X
right.
E
L
Is
approach
number
one
approach
number
two
we
have
in
sit,
which
is
what
4
6
4
X
lat
is
using
in
the
silat,
which
is
the
translation
from
not
4
to
6
or
not
4
6.
We
have
already
a
shoot
for
another
standard
which
is
explicit,
address
mapping
table
okay.
So
that
means
that
instead
of
telling
the
translator
to
use
the
net
64
prefix,
we
can
tell
the
translator-
hey
I,
have
already
here
a
quite
a
record
in
my
Syrian.
Why
you
doing
you?
Don't
use
that
record
for
the
translation?
L
You
understand
that
you
got
that.
So
the
idea
is,
for
example,
if
I
am
trying
to
reach
a
website
from
a
smart
TV
that
has
this
a
wrecker
and
also
the
website
is
dual-stack
it
and
has
this
ipv6
address.
Let's
have
a
way
for
the
salaat
to
make
automatically
an
entry
in
the
explicit
addressing
mapping
table
for
this
connection.
L
So
what
this
means
is
that
this
Blissett
address
mapping
entry
will
be
having
the
destination
for
this
ipv4
address,
going
to
use
the
real
quad
a
record
for
the
Sirian
instead
of
using
day,
not
64,
and
that
means
that
by
the
for
routing
it
will
reach
directly
without
needing
an
additional
translation.
Now
in
this
approach,
what
I
am
suggesting
is
let's
make,
because
the
cities
usually
have,
and
in
fact
in
the
in
the
new
document
that
we
approve
it
on
it's
in
the
in
the
SG.
Sorry
is
in
the
RFP
Kiwi
for
de
P.
L
Before
as
a
service
cities,
we
said
they
must
have
a
DNS
proxy
okay.
So
why
not
asking
the
CPE
when
it
detects
a
device
that
is
asking
for
an
a
record
and
not
asking
for
a
quite
a
record
what
that
means.
It
means
that
the
device
is
ipv4
only
so,
let's
make
automatically
the
CP
to
verify
if
there
isn't
quite
a
record
and
make
that
entry.
L
What
is
the
impact
on
that
not
but
impacts?
The
only
thing
is,
if
somebody's
changing
the
DNS
of
the
Smart
TV,
it
will
not
work,
but
usually
nobody
changed.
The
DNS
of
the
Smart
TV,
if
somebody's
changing
the
DNS
in
the
CPE
is
still
working
okay.
So
there
is
not
a
negative
impact
at
Lee.
I
can
not
see
that
yeah.
H
A
H
G
Yeah
sure
it
March
Akamai,
there's
ongoing
effort
to
go
and
take
a
lot
of
this
stuff
away
from
host
level
control
and
push
it
into
application
level,
control
for
DNS
queries
and
so
this
technique.
Well,
it
works
for
host
level
controls
and
I'm
kind
of
wondering.
This
is
a
this
is
maybe
off
topic
for
this
working
group,
but
this
is
going
on
kind
of
in
the
in
the
general
multi
DNS
area.
There's
a
number
of
DNS
working
groups.
G
Application
behaved,
the
application
and
host
layer
interactions
are
currently
changing,
and
certainly
the
standards
work
that's
happening
here
at
the
IETF
is
transitioning
in
that
direction.
This
is
something
that
I
think
people
should
generally
be
aware
of,
because
it
may
have
implications
in
other
areas,
such
as,
if
you're
implementing
these
types
of
technologies,
yeah.
L
I
understand
that
we
we
need
to
do
for
interactions
which
quids
TNS
changes.
That's
that's
clear,
okay,
so
that's!
Let's
call
this
approach
number
two
and
then
I
have
number
three
number.
Three
is
basically
the
same
as
number
two,
but
instead
of
being
the
CP,
the
one
creating
the
EMT
it's
created
or
push
it
or
pull
it
by
the
CP
from
the
operator.
Okay,
so
that
bandage
is
the
operator,
has
more
control
but
increased
complexity.
L
Okay,
because
it
means
the
operator
need
to
do
something,
and
probably
we
need
to
find
which
protocol
to
use
to
send
this
table
to
the
CP
or
to
be
able
to
get
the
CP
getting
that
table
from
the
operator.
Okay,
so
I
really
think
that
the
best
choice
and
the
one
that
is
not
breaking
anything
but
at
least
is
optimizing.
L
L
That's
I
think
it's
interesting
at
least
even
if
we
can
only
solve
that
that
point
what
it
means,
basically,
that
the
document
that
we
have
today
for
464xlat
should
be
updated
to
say:
hey,
you
need
to
support
explicit,
address
mapping
tables
which
it's
already
a
shoot
in
sit.
Ok,
so
it's
not
that
much
difficult
in
my
opinion
and
because
we
have
just
released
the
document
waiting
for
the
RFC
number,
that
is
telling
the
operators
sorry
babe
benders
do
dip.
The
sea.
Peas
like
this
I
think,
is
the
right
time
to
say:
hey.
L
L
L
Now
one
further
idea-
and
this
is
maybe
too
crazy-
is
this-
a
record-
may
be
pointing
to
specific
address,
hosted
by
ITF
I.
Can
the
registries
whatever
to
say:
hey,
you
are
trying
to
access
an
ipv6,
only
service
grid,
which
you
write,
ISPs,
not
offering.
Okay,
that's
that's
an
additional
crazy
idea,
but
come
on
at
the
end,
this
user,
which
I
pee
before,
is
not
going
to
to
access
this
service
anyway.
So
it's
it's
somehow
a
way.
Also
to
push
people
hey
knock
the
ISP.
Do
you
know
offer
ipv6?
L
K
K
L
H
Clear
Michael
Abraham's:
yes,
so
you
can
probably
create
heuristics
and
figure
out
how
to
do
this,
but
and
I
think
the
the
reason
you
want
to
do
this
is
the
downside
is
that
you
lose
the
optimization
and
if
you're
right,
then
things
will
work
better.
So
the
worst
case
is
the
Curtiss.
What
we
do
today-
and
the
better
case
is
some
optimization
right,
so
there
is
very
little
downside
to
be
wrong.
Yeah.
Q
Jeff
used
to
just
has
really
said
that,
because
you're
wrong
okay,
this
isn't
your
mother's
DNS
and
if
you
think
two
consecutive
queries
will
invariably
go
to
the
same
resolver
engine.
That's
what
your
mother
thought!
It's
not
what
happens
today.
You
are
really
far
off
the
way
we
now
do
dns
in
bulk,
even
down
near
the
user.
So
this
idea
that
somehow
a
memory
about
the
nature
of
query
a
can
also
reflect
the
new
query.
B
does
not
apply
anymore,
so
you're
right,
but.
L
Q
L
L
D
Seems
only
way
you
can
solve
this
Dena's
problem
is
to
make
sure
you
CP
is
your
resolve,
whereas
NCP
might
have
some
logic
about
knowing
which
client
is
before
only
but
providing
cow
broke
and
CPUs
I.
It's
not
gonna
happen.
On
the
other
hand,
come
on
if
before
works,
roars
and
wishes,
that's
okay,
I,
don't
mind.
Okay,.
H
Microwave
Renzo's,
so
what
about
doing
traffic
statistics
I
mean
if
you're
talking
to
if
you're
talking
to
an
ipv4
address
a
lot?
If
this
is
for
streaming,
video
I
mean
you
might
be
talking
to
a
few
places.
You
could
look
into
correlating
this
on
just
the
IP
layer
and
just
looking
at
the
traffic,
because
I
mean
you're
instead
of
using
the
DNS
correlation.
Okay,
you
could
do
that
as
well.
I,
don't
know,
but
I
think
you
mean
prototype
this
and
see
if
it
works
with
a
bunch
of
devices
and
see
how
well
it
works.
Well,.
L
D
L
Goes
in
the
slides,
I
didn't
mention,
but
I
think
that
what
is
missing
today
in
the
empty
table
is
a
TTL
I
think
we
should
assume
the
TTL
from
the
VNS.
Obviously
okay,
so
if
you
get,
if
you
create
an
entry
in
the
in
the
M
the
table,
what
is
the
time
life
for
that
entry?
They
want
from
the
DNS
quite
a
record
response.
What.
D
G
L
The
second
person
of
the
document-
that's
zero,
one
already
included
Alejandra's
Walter,
because
he
also
said
that
he
may
be
able
to
try
in
his
in
his
test
bed
as
well.
I
am
considering
that.
Maybe
we
need
to
change
the
title
because
actually
is
talking
about
four
six
four
X
lat
and
I
think
it
can
apply
to
other
environments
where
you
are
using
an
ad
for
six
terms
later
in
the
CP
and
and
that's
it
I,
also
the
question
of
becoming
working
or
item.
A
A
R
A
A
S
B
T
T
The
overall
goal
for
this
document
is
to
collate
all
this
information
around
the
various
methods
into
one
one
place.
It's
to
then
highlight
the
the
benefits
and
the
detriments
of
each
of
them
and
to
provide
provide
an
operator
with
all
the
information
they
should
need
to
to
make
an
informed
decision
around
which
technology
suits
their
network.
The
best,
so
we've
decided
to
focus
on
the
five
most
popular
or
the
current
technologies
that
we
see
and
just
plain
ignore:
the
the
alternatives
and
the
legacy
ones
that
we
we
deem
unpopular.
T
T
Before
going
into
a
bit
more
bit
more
of
the
architecture
and
then
the
detail
of
each
of
these
technologies,
we
didn't
go
down
into
even
further
detail
around
the
differences
between
them.
The
current
supports
so
what's
what's
available
today
in
the
current
CPAs
or
not
sort
of,
and
possibly
the
regulatory
concerns
that
one
may
have
over
the
other
v4
address,
sharing,
isn't
a
separate
section
on
itself.
There's
an
important
topic
that
does
get
discussed
throughout
the
document.
T
As
mentioned,
ipv4
address
sharing
is
quite
prevalent
through
this
document.
It's
quite
a
common
we
commonly
used
when
providing
ipv4
service
because
who
wants
to
give
everybody
a
single
ipv4
address
these
days.
We
talked
about
the
location
of
where
that
the
traditional
stateful
napped
for
for
happens,
be
it
in
on
the
CPA
or
further
in
the
network.
T
T
We
talked
to
talked
about
load,
balancing
load
sharing
between
the
different
devices,
the
gateways
which
ones
can
be
deployed,
and
then
anycast
topology
we'd
also
talked
about,
which
ones
are
how
they
are
typically
used
today.
So,
for
example,
for
6
for
X
left
in
a
mobile
environment,
but,
as
Jory
mentioned
before,
you
can
also
deploy
that
in
a
in
a
CPU
gateway.
T
We
we
highlight
the
provisioning
options.
Each
one
of
them
has
whether
there
are
existing
dhcpv6
options
to
configure
the
devices,
whether
there's
a
yang
model,
where
there's
radius,
attributes
that
sort
of
thing
and
then
a
brief
brief
security
considerations
mentioned
certain
things
like
the
code
size
of
each
implementation
and
how
they
differ,
and
that.
T
So
those
are
the
overall,
the
key
things
that
I
think
we
mentioned
in
the
document
and
I
think
a
key
we've
had
some
changes
since
the
last
revision
and
that
you
may
have
heard
this
being
presented
about
the
last
ITF
Ian
has
joined
us.
As
a
co-author,
there's
been
some
additional
details
and
figures
added
to
the
description
of
the
of
each
of
the
technologies.
T
There
was
some
restructuring
around
the
description
moves
into
a
separate
section
as
well,
and
then
with
completely
rewritten
the
high-level
architecture
and
their
consequence
section
and
also
whole
bunch
of
other
additions
throughout
the
detailed
analysis.
Section
there's
still
quite
a
bit
that
we
want
to
do
in
this
document.
There's
a
bunch
of
performance
and
scalability
analysis
that
we
want
to
do.
T
There's
that
81:9
methodology
of
get
Bors
I
think
Gabor
has
a
PhD
student
I
believe
that's
implemented
a
stateless
now
6:4
testa
I
believe
he's
he's
going
to
try
and
find
another
student
to
do
some
stateful,
nat64,
tester,
told
him
as
well,
but
I
think
he
has
also
made
a
request
for
anybody
who
would
like
to
help
develop
some
tooling
to
help
tests
these
transition
technologies.
He
would
be
very,
very
keen
to
hear
from
you
a
bit
more
work
in
the
security
side
of
things
around
there
as
well.
T
So
he's
also
got
another
methodology,
that's
referenced.
They
are
they
we're
planning
on
using
I,
also
like
to
have
a
small
section
around
why
we
just
excluded
the
the
other
aforementioned
legacy
technologies
and
anything
else
that
you
guys
may
feel
that
we
should
include
in
the
document.
It
would
be
happy
to
hear
from
you
now
or
on
the
list
and
I
think
that's
it
yeah.
That's
it
so
yep
still
gonna
keep
working
on
it.
T
There's
gonna
be
some
more
revisions
coming
through
I
do
think
the
authors
were
looking
to
ask
if
there
was
support
for
this
as
a
worthwhile
document.
I
think
Fred.
You
actually
posed
the
topic
of
this
document.
Initially
I'm.
Sorry
I
think
you
proposed
the
totnes
draw
this
document
topic
list
originally,
so
it
came
out
of.
A
T
So
I
think
two.
If
the
original
goal
at
the
start,
the
slides
was
to
provide
a
single
reference
point
for
an
operator.
I
went
through
myself
a
while
ago
trying
to
do
this
analysis
myself
and
work
out
which
ones
we
should
be
looking
at
the
point.
So
if
I
had
had
this
document
two
years
ago,
it
would
have
made.
T
My
well
would
have
saved
me
a
lot
of
time
and
helped
me
out
quite
a
lot
so
I
think
that's
my
personal
goal,
for
it
is
to
help
people
like
me
who
had
to
make
this
decision
so
I
want
to
make
sure
we
capture
everything
that
we
can
so
I
would
like
to
see
some
more
discussion
around
if
people
think
we
are
missing
some
key
points
and
help
us
get
all
that
information
into
this
document
that
you
think
is
useful.
So
maybe
it's
a
question
for
the
other
13
operators
in
the
room.
A
Seems
like
at
least
part
of
what
we're
I,
just
a
random
bird
sitting
on
the
wall
and
thinking
about
such
things,
I
might
say
so,
which
one
should
I
be
using.
Please
and
you
wind
up
with
the
discussion
of
the
trade-offs,
for
example,
with
like
light
weight
for
over
six
and
what
terror
stream
is
using.
If
all
of
your
all
of
your
CD
CDP's
are
or
CDs
are
in
a
data
center
and
that
all
works,
if
you're
doing
translation
such
as
Alejandro
is
doing
that
might
not
work
and.
E
T
Absolutely
I'll
just
so.
There
are
there's
also
a
lot
of
subjective
things
in
here
as
well,
so
we
mean
like
I
like
it
yeah
pretty
much
I
mean
so
if
you,
if
we
were
to
ask
Michael
Laurie
and
they
prefer
the
encapsulation
version.
So
if
I
wait
for
over
six
as
well,
they
don't
always
I'm
less
against
translation,
so
I'm
looking
for
a
map,
t
approach,
so
it's
just
some
ideology
things
they'd
have
to
well.
We
haven't
really
captured
but
I'm,
not
sure
how
to
capture
those
who.
H
Make
over
MC
so
some
of
the
practical
decisions
one
has
to
make
also
has
to
do
with
platform,
support
for
different
methods
and
so
on.
I
think
that's
gonna
be
hard
to
capture
because
now,
all
of
a
sudden
you're
starting
to
talk
about
a
consistent,
on-chip
persons,
you
know
from
different
vendors
and
so
on.
I
still
think
it's
a
very
useful
document
and
I
agree
with
with
Richard
here.
H
H
L
Of
course,
we
are
now
looking
into
more
details
that
I
didn't
realize
it
at
the
time,
but
I
I
think
if
we
don't
have
this
document.
What
it
means
basically,
is
that
any
operator
that
want
to
choose
a
transition
mechanism
for
ipv6
only
an
eye
peeled
for
the
service
need
to
root,
to
read
and
understand
all
the
details
of
every
one
of
the
documents
that
we
have
on
those
transition
mechanism.
L
This
document
clearly
facilitates
a
lot
that
that
work-
and
that
was
my
reason
to
say
when
fret
or
who,
who
was
in
the
mailing
list
about
this
que
I,
will
join
the
effort
because
I
already
started
that
effort
on
my
own.
A
few
years
ago,
I
really
think
it's
very
useful
and
I
got
a
lot
of
people
reading
the
document
and
telling
me
yes,
this
is
this.
Is
this?
R
Yeah
so
I
think
that's
that's
a
useful
document
in
some
some
degree,
so
I
disagree
with
the
majority,
someone
who
is
doing
his
work.
He
has
to
read
all
the
pointers
who
are
in
this
in
only
I
would
say
a
starter
and
there's
some
and
of
index,
and
then
you
have
to
zoom
into
the
details
technologies
to
understand
what
is
behind.
So
it
provided
just
a
piece.
R
We
don't
know
how
yeah
it's
just
a
first
first,
a
first
view
of
the
available
I
would
say
alternatives,
but
if
you
need
to
I
would
say
to
make
your
own
decision,
you
have
to
understand
yeah
yeah
yeah
yeah.
So
you
have
you
have
to
read
that
you
have
to
understand
the
subs
of
this
between
all
these
I
would
say
this
proposals.
Actually
in
the
proposal.
They
are
not
five,
then,
or
only
two.
They
are
there's
only
the
net
based
one
and
E
Plus
P.
R
So
if
you
have
this
kind
of
category
there,
people
won't
be
surprised
when
you
open
the
document.
You
say:
there's
a
lot
of
a
proposal
there.
So
it's
just
a
matter
I
would
say
on
present
in
the
material.
The
material
is
useful
in
some
in
some
degree
it's
not
among
that
we
to
have,
but
it's
nice
to
have,
but
then
you
have
to
understand
the
subtlety
behind
each
of
them.
Yeah.
T
Yeah,
sorry,
I,
don't
I,
think
a
lot
of
operators
I'm
trying
to
be
careful
with
what
I
say:
they're
not
gonna,
spend
the
time
going
down
and
reading
all
these
individual
RFC's
exactly
half
the
time.
They
won't
even
look
they'll
just
do
what
the
vendor
tells
them.
They'll
say:
hey
yeah!
We
support
this.
That's
really
great!
You
should
use
it.
R
T
U
Come
from
the
Vanek
I
think
I'd
like
to
see
this
become
adopted
as
a
working
group
document,
because
it's
an
important
decision-making
document
for
someone
who's
about
to
want
to
deploy
over
the
last
year
and
my
team
will
be
running
some
kind
of
hot
desk
for
operators.
What
is
it
by
ATV?
6-0
of
them
start
that
decision
by
going
to
read
the
RFC
for
Marty,
none
of
them
so
I
think
this
is
going
to
be
useful
document
for
decision-making,
and
then
you
know
the
details.
Can
people
can
always
refer
to
the
respective
RFC?
V
Hello,
so
it's
one
of
the
co-authors
on
the
most
recent
version,
that
what
we've
been
trying
to
do
here
is,
as
mentioned,
is
to
remove
anything
around
objective
kind
of
things,
and
just
and
just
boil
it
down
to
this
is
these
are
the
king
points,
and
this
is
the
stuff
that
you
need
to
know
as
part
of
the
decision-making
and
things
which
are
demonstrable
facts
as
much
as
is
possible
inherent
in
the
different
approaches
and
I.
V
Think
if
you
can
do
this,
you
know
I
mean
that
the
things
like
logging
are
they're
going
to
be
deal
breakers
with
some
of
the
things.
If
you
you
know,
if
your
regulatory,
your
local
law
regulation,
says
you
need
to
do
this
kind
of
logging
here
that
will
exclude
certain
mechanisms
from
there.
There's
no
point
in
going
on
reading
the
RFC's
there
and
trying
to
puzzle
through
how
do
you
know
what
the
implications
of
all
of
those
kind
of
things
is
they're,
just
fundamentally
not
capable
of
it?
V
And
so
you
know,
the
idea
was
to
try
and
keep
the
comparison
between
things
as
things
which
are
just
demonstrable
facts,
and
they
see
these
are
the
characteristics.
This
is
the
stuff
that
you
need
to
know.
If
you
can
use
that
to
thin
out
what's
worth
considering,
then
you,
you
know
it
would
probably
save
you
a
load
of
work
and
a
load
of
thinking.
It
would
otherwise
have
to
do.
D
P
Many
Cisco
from
NTT
communications
are
down
operator,
I
love,
reading
RLC,
then
yeah,
so
I
support
these
tourists
regard
also,
if
it's
very
useful
to
decision-making,
then
that
one
quick
comment
so
RFC
six,
eight
eight
eight,
which
is
the
CPU
one
to
seven.
If
the
common
requirements
for
Cali
agreed
not
so
he
took
it-
was
to
be
rethought
from
this
debate.
So
sorry,
I
missed
that
will
kiss
repeat
the
question:
RAC
six
888
about
como
requirement.
Oh
great.
T
W
Hey
Richard
young
George,
looking
for
Internet
Society,
but
speaking
on
Mike
in
my
own
capacity,
and
also
as
as
the
best
color
operational
practice,
Task
Force
co-chair
a
tribe
I
think
this
is
really
really
interesting.
Work
that
you
are
doing
there
and
I
would
suggest
if
you
can
join
us
in
Reykjavik
meeting
that
we
also
discuss
this
this
list
at
the
personal
operational
practice,
Task
Force,
where
you
have
where
they
have
even
more
operators
than
again
then
in
this
room
but
I,
think
it's
it's
good.
A
So
now
let
me
ask
a
question:
we
actually
have
a
document,
the
title
of
which
is
guidelines
for
using
night
bg6
transition
mechanisms
during
ipv6
deployment.
Yari,
Arco
and
I
wrote
that
about
eight
years
ago,
and
it's
probably
due
for
an
update,
but
it
raises
the
question:
should
this
draft
perhaps
be
considered
to
obsolete
and
replace
that
draft?
If
so,
then
I
think
you
need
to
address
at
least
the
issues
that
are
in
that
draft.
Don't
learn
at
that
RFC
and
that's
something
that
we
can
discuss.
A
What
I'll
do
is
I'll
send
a
note
to
the
list
saying
what
do
you
think?
I
should
just
be
working
group
document
that
that's
the
way
we
you
try
to
do
these
things
anyway,
so
that'll
happen
in
a
few
weeks
and
you'll
see
that
when
it
comes,
it
sounds
like
at
least
here
on
the
floor
that
there's
a
fair
amount
of
support
for
that.
A
A
L
So
this
is
Fernando
percent
in
the
document.
Flux
reaction
to
remember
event,
events.
This
is
a
document
that
we
co-author
with
young
source
and
discussion,
so
far
has
happened
mostly
on
both
mailing
lists
on
six
men
and
be
six
ops
for
the
most
part
we
wanted
to.
You
know
press
in
the
document
on
p6
ops
to
get
feedback
from
operators.
In
particular,
we
try
to
do
our
best
to
post
an
announcement
of
the
document
on
on
relevant
operator
mailing
lists.
L
So
what's
this
document
about
I
mean
there
have
been
like
a
lot
of
discussion
on
the
mailing
lists
over
thirty
three
hundred
messages,
so
the
short
version
of
it
is
that
there
are
a
number
or
scenarios
in
which
the
you
might
have
slack
hosts
and
the
slack
network
configuration
information
may
become
stale
without
the
host
on
the
network.
Becoming
aware
about
that.
This
is
just
one
of
the
scenarios
in
which
this
might
happen.
L
There
are
others,
this
is,
let's
say
the
most
common
one
in
which-
and
it's
obviously
a
quite
common
ipv6
deployment
scenario
in
which
you
have
the
CPU
router
that
does
DHCP
prefix
delegation
with
the
ISP
and
the
CPU
router
announces
some
prefix
of
the
list
prefix
on
the
on
the
local
network.
So
nothing
complex
about
that
now.
What
happens,
for
example,
in
scenarios
where
the
CPU
runner
crashes
and
reboots,
for
example?
L
L
The
CPU
router
might
actually
get
a
different
prefix
done
before.
Ok,
if
the
CPU
rather
didn't
actually
record
the
previously
leased
prefix
on
stable
storage,
it
might
have
you
know.
No,
no,
no
state
information,
no
idea
about
you
know
the
prefix
that
has
been
leased
before,
so
it
will
essentially
advertise
the
new
prefix
on
the
local
network,
meaning
that
you
know
the
notes
on
the
local
network.
We
just
configure
another
is
for
the
new
prefix,
but
nothing
happens
with
the
previous
address
the
previous
computer
address
or
the
previous
configuring
prefixes.
L
L
The
old
addresses
might
be
preferred,
that's
implementation
dependent,
but
they
might,
you
know,
employ
the
addresses
that
you
know
were
first
configured
and,
of
course,
the
result
of
that
is,
for
example,
that
v6
connectivity
might
fail
if
it's
a
v6,
only
scenario
that
might
be
just
you
know,
play
failure
and
that's
it
if
it
said
we'll
start
scenario,
and
you
know
that's
an
application
that
implements
happy
eyeballs
well,
that
might
mean
that
things
work
over
before,
but
not
over.
Ipv6.
L
L
The
problem
is
that
it
gets
different
profits
without
knowing
that
you
know
there
was
it
previous
another
prefix
being
used
on
the
local
network
or
don't
use,
leave
the
perfect
doesn't
change.
You
know
the
problem
doesn't
even
come
up.
There
are
multiple
reasons
for
which
that
might
not
be
the
case.
I
mean
there
are
many
among
all
of
the
possible
ones.
L
Are
there
that
there
are
is
peace
that
want
to
charge
for
persistent
or
stable
prefixes,
but
there
are
others
and
another
case
that
you
know
my
kind
of
like
avoid
this
problem
to
some
extent,
is
the
case
of
cps
that
record
on
stable
storage,
the
prefix
that
has
been
leased.
So
the
idea
is
that
if
you
have
a
CP
it
records
on
stable
storage,
the
prefix
that
was
leased
before
then,
if
it
crashes
and
reboots,
it
has
to
some
extent
the
ability
to
actually
deprecated
the
previous
prefix.
L
Okay,
of
course,
the
question
here
is
to
what
extent
this
is
actually
the
case.
In
many
cases
the
CP
is
don't
just
don't
just
record
the
prefix
that
has
been
leased.
There
are-
and
this
is
also
tricky
for
a
number
of
reasons
like,
for
example,
if
you
were
to,
if
the
CP
were
to
you,
know,
crush
and
reboot,
you
might
be
able
to
actually
deprecated
or
prefer
the
previous
prefixes,
but
you
are
not
able
to
actually
completely
disable
the
addresses.
L
So
that
means
that
the
nodes
that
are
connected
to
the
network,
we'll
still
keep
your
abdiel
addresses.
They
will
not
be
preferred,
but
the
addresses
will
be
configured
and
there
are
ipv6
implementations
that
limit
the
number
of
configure
addresses.
So
that
could
mean
that
if
there
is
like
multiple
crashes
and
reboots,
for
example,
you
might
be
able
to
enter
the
addresses,
but
you
might
end
up
hitting
the
limit
on
the
maximum
number
of
addresses
that
are
configured
okay,
I
mean
this
is
just
something
that
is
implemented
and
has
been
deployed.
L
L
Okay
in
the
idea,
we
say
something
along
the
lines
off:
okay,
if
you
receive
it
a
couple
of
those
arrays
and
after
the
second,
you
should
prefer
the
address,
and
if
you
receive
a
few
more
then
you
should
completely
eliminate
the
address.
Okay.
But
of
course
there
are
multiple
things
that
you
could
do
here,
like
you
could
say:
okay,
when
I
have
received
a
couple
of
these
prefixes
I
might
just
prefer
the
addresses
and
then
keep
the
addresses
there.
L
L
Well,
there
are
multiple
things
that
have
been
proposed
when
it
comes
to
the
their
response
of
these
packets,
or
you
know
when
you
detect
that
the
information
might
have
become
stale.
For
example,
some
foods
have
suggested
to
perform
BFD
in
response
to
this
I'm,
not
saying
that
that's
what
I
think
should
be
done.
I
mean
this
is
among
the
options.
I
would
say.
The
bottom
line
is
that
at
the
moment
in
which
you
infer
that
the
information
might
be
stale,
there's
something
to
be
done
to
actually
be
able
to.
L
You
know
mitigate
this
problem,
all
the
things
that
might
help.
If
you
look
at
our
SC
48
61,
it
essentially
says
that
if
you
receive
a
prefix
information
option
with
a
life
with
the
valid
lifetime
smaller
than
two
hours,
essentially
you
shouldn't
update
it
by
the
lifetime.
The
idea
is
to
prevent
denial
of
service
attacks.
Okay,
our
take
is
that
there
are
so
many
vectors
when
it
comes
to
neighbor
discovery
that
you
know
just
ignoring
those
particular
packets
because
of
one
specific
vendor
doesn't
make
much
sense.
L
As
long
as
you
have
multiple
venues
to
do
the
same
thing,
it
doesn't
really
matter,
there's
lots
of
things
that
you
could
do
with
arrays
disabled
routers.
You
know
configuring,
correct
prefixes,
so
many.
So
what
we
think
is
that
that
rule
should
be
removed,
and
if
that
was
the
case-
and
you
had
a
CPE
that
is
aware
that
you
know
the
information
has
become
stale
and
the
CPE
has
recorded
an
information
on
stable
storage.
It
would
be
able
to
actually
completely
you
know
clean
up
the
mess
on
on
the
network.
L
Another
thing
that
we
suggest
and
discussed
in
the
idea
is
to
reduce
the
lifetime
of
the
prefix
information
options.
The
current
values
are
that
PIOs
have
a
default
valid
lifetime
of
one
month,
okay,
meaning
that,
if
you
don't
have
explicit
signaling,
the
information
is
going
to
be
there
for
one
month
they
prefer
lifetime
if
I
remember
correctly,
something
along
the
lines
of
seven
days.
L
Okay,
what
we
say
in
the
document
is
that
the
lifetime
should
be
in
the
same
order
of
the
route
like
them
and,
if
you
think
about
these
in
the
context
of
RFC
a
t28,
essentially
it
says
that
a
prefix
is
tied
to
the
router
that
advertise
it.
Why?
Because
if
you
have
multiple
routers
advertising
prefixes
on
the
local
network,
then
if
you
just
source
packet
from
any
of
those
prefixes
to
any
of
the
routers,
then
the
packets
might
be
a
grass
filter.
H
H
L
H
H
We
should
we
just
discussed
this
yesterday
that
we
wanted
to
test
what
hosts
already
do
and
I
think
that's
where
worthwhile
to
get
some
data
there,
because
I
have
done
boo-boos
at
home
and
sent
RAS
and
have
you
know
all
the
kinds
of
devices
configure
address
affordable
for
me,
but
for
itself,
I
have
surprisingly
little
problem
because
of
this
and
I
I
was
under
the
impression
that
the
preferred
lifetime
the
high
the
PID
address
with
the
highest
preferred
lifetime
was
used
for
new
connections.
I
cannot
find
this
in
the
RFC's
anywhere.
L
D
Journaling
kova
someone
who
is
heaven
hedgehog,
a
Groundhog
Day
feeling
a
force
is
sweet,
so
Michael.
So
the
source
of
this
selection
algorithm
saying
that
before
you,
when
you
get
to
the
longest
prefix
match,
you
might
don't
use
it.
If
you
have
other
means
of
finding
out
which
addresses
better
so
so
application
a
host
can
do
it
because
definitely
can
do
what
I
actually
proposed
some
time
ago.
So
Fernanda
I
sent
her
three
pages
email
this
morning.
I
got
a
comment
in
kanji,
so
I'm
not
gonna,
repeat
that
I
totally
agree.
D
D
L
O
L
Think
is
well
they're,
the
only
one
that
I
was
a
lift
which
no,
the
purpose
of
the
presentation
was
to
get
the
feedback.
Then
there
was
the
question:
while
we
were
discussing
this
in
on
both
v6
of
sand
and
six-month,
there
were
false,
saying
that
one
hand
with
to
document
the
problem
and,
on
the
other
hand,
like
do
solutions
I
do
think
that
the
solutions
belong
for
the
most
part
in
in
six-month.
So
that's
what
I
would
say
when
it
comes
to
the
documentation
of
the
problem.
T
T
We
have
widgets
and
things
the
moments
already
defined
that
we
can
use
to
solve,
or
at
least
mitigate
this
problem
for
a
good
90,
something
percent
of
of
the
problem
and,
as
we
spoke
about
before,
we've
we've
done
that
I
think
documenting
sort
of
proposed
recommendations
and
and
best
current
practices,
and
things
like
that
might
be
a
better
fit
for
this.
Well
quantifying
that
last
remaining.
T
L
T
X
Hi
Bob
hindered
several
points,
I'm
somewhat
skeptical
of
the
basic
problem.
It
seems
to
me
that
the
ISP
just
should
remember
the
prefix
it
gives
out
to
the
customer.
For
the
same
reason,
I
think
it
has
to
remember
that
it's
a
customer
so
having
a
having
the
additional
state
of
the
prefix
assigned
to
the
customer
doesn't
seem
too
much.
X
So
it
would
be
good
to
write
that
down
somewhere
because
I
don't
see
why
that's
seems
to
be
the
core
problem
here
is
that
the
ISP
is
not
sending
out
the
same
prefix
for
the
lifetime,
so
I
don't
see
why
I
can't
remember
that
I.
This
document
is
not
a
remembering
document,
because
you
don't
deal
with
all
the
other
aspects
of
remembering
like
updating
DNS
like
changing
firewall
rules.
It's
a
look.
It's
a
well-known,
big
problem.
This
is
not
a
remembering
solution.
This
is
trying
to
address
a
perceived
it's
specific
problem.
Yeah.
T
X
You
should
like
not
use
different
words
in
the
document
so
and
yeah
like
some
of
the
other
people,
I'm,
not
convinced
that
the
right
solution
here
is
protocol
changes
I
think
it
may
be
better
to
document
the
problem
and
use
the
tools
that
we
have
already,
which
ones
specifically
they
don't
have
a
specific
list,
but
yeah
I'm,
just
not
sure
the
approach
of
trying
to
change
its
neighbor
discovery
is
the
right
thing
for
this.
Well.
L
Iq
thing
is
that
what
my
employee-
let's
say,
the
existent
mechanisms
is
to
somehow
make
the
addresses,
become
unprepared
and
there's
really
nothing
that
you
have
there
to
actually
trigger
the
addresses
to
become
a
prefer.
If
you,
the
only
thing
that
you
might
have,
if
you
want
it
would
be
to
play
with
the
timers,
but
then
either
you
have
to
really.
You
know,
reduce
the
timer
so
much,
which
is
you
know
the
right
thing
to
do,
and
so.
X
L
There
are
others
that
don't
have
to
do
with
a
with
the
timers,
but
when
it
comes
to
that
particular
one,
the
only
scenario
in
which
you
would
run
into
trouble
is
the
scenario
in
which
the
CPE
dies,
because
what
we
say
is
that
ACP
should,
for
example,
never
advertise
a
lifetime
longer
than
the
router
lifetime.
But
as
long
as
the
CPU
router
is
alive,
it
would
keep
sending
the
erase
and
the
life
stamps
would
be
refresh.
L
Y
High-Tension
so
I
suspect
this
problem
is
relatively
well
hidden
at
the
moment
because
of
happy
eyeballs
and
the
fact
that
this
is
generally
seems
to
be
aimed
at
residential
networks
where
there
isn't
much
v6
only
content
they
P
unis,
looking
to
go
to
so
we're,
probably
not
seeing
that
the
full
impacted
at
the
moment
so
follow.
The
question
of
that
is:
has
anyone
done
any
real
tests
with
various
operating
systems
and
Reuters
and
actually
seen
what
actually
happens
and
whether
what
we
think
happens
does
happen?
I
mean.
L
Y
Y
Y
Y
A
Okay,
so
I
need
to
bring
this
to
a
close,
because
I've
got
another
presentation
in
10
minutes,
so
no
I
hope
you
got
out
of
this.
What
you're?
Looking
for
okay,
Martin.
N
N
N
N
N
N
Just
also
draft
is
actually
adopted
now
by
Jen,
which
is
using
array
option,
and
there
is
a
also
RFC
at
which
using
DHCP,
but
it's
not
listed
here,
and
there
is
also
this
draft
now
some
pros
and
cons
about
those.
So
current
solutions,
Darcy
70/50
works
with
modification
without
purification
to
protocols.
This
is
good,
actually
so
most
of
the
RFC
70
51
issues,
not
the
issue
number
3,
because
it's
dns-based,
so
it
requires
dns
to
operate
there.
Also
some
cons
about
this-
that
it
requires
dns64
with
the
same
network,
specific
graphics
or
well-known
prefix.
N
D
N
H
N
N
It
works
with
foreign
DNS,
because
when
you
do
have
a
local
domain,
then
you
might
also
search
for
the
records
in
global
DNS,
even
through
toh.
If
you
want-
and
it
might
also
be
application
level
change,
because
if
you
using
do
H
and
browser,
then
problem
can
solve
it
by
using
this,
and
there
are
no
changes
in
the
protocol.
N
All
is
already
there,
you
see,
DNS
is
already
there
and
it
requires
a
local
domain
detection.
It's
a
main
pitfall
of
the
this
draft,
because
it's
this
seems
to
be
quite
issue
now
what
you
can
do
in
terms
of
local
domain
detection,
when
you
do
have
a
just
one
segment
which,
for
example,
might
be
your
company,
then
it
is
easy.
You
can
just
use
DNS
search
list
option
in
the
are
a
DHCP
option.
24,
and
what
comes
in
problem
actually
comes.
Is.
N
There
is
a
for
example,
router
in
in
way
in
such
way.
You
can
use,
for
example,
PTR
records,
which
also
would
point
probably
to
your
operator
when
you
are
a
home
user,
and
there
are
also
some
options
in
DHCP,
mainly
option
57
or
110,
and
there
are
also
some
other
ways
how
to
this
can
be
actually
discovered.
N
N
The
AR
AR,
who
is,
is
not
thing
you
would
or
should
use,
because
this
ends
up
to
be
PTR
records,
so
no
time
and
now
about
detection
of
Gnostics
for
itself.
Basically,
the
house
would
ask
for
SRV
record
for
each
local
domain.
Srv
record
would
look
like
this
actually,
so
there
will
be
some
not
six
for
ipv6
prefix.
E
N
N
Now
a
detection
of
the
nat64
I
was
thinking
if
I
should
include
this
into
the
draft,
but
it
seems
for
me
it's
might
be
useful
because
it
might
also
be
used
by
other
means
of
detection
of
nat64
prefix.
If
you
don't
have
a
capability
of
making,
not
six
four
addresses
at
a
host,
so
it
might
point
to
some
servers
which
would
provide
either
standard
DNS,
VOT
or
whatever.
N
N
How
would
you
say
to
client
that
there
isn't?
No,
not
six
for
a
surrogate
court
allows
a
dot,
and
so
that
service
is
not
provided,
it
depends
how
it
should
be
processed
by
the
host.
One
thing
would
be
to
use
it
as
not
not
provided
to
you
specifically
when
you
are
asking
from
outside
DNS
or
saying
just
that
there
is
no,
not
sixth
form
it's
up
to
a
decision,
and
there
is
also
one
input
about
change
in
configuration.
N
Dns
records
does
have
the
time
to
live.
So
when
there
is
a
change,
for
example
in
record
itself,
then
it
would
time
out
eventually
also
whether
there
is
a
change
in
your
up
stream,
and
you
are
giving
this
records
to
your
downstream.
Then
you
might
also
consider
to
use
C
name,
because
when
there
is
a
change
in
upstream,
it
will
also
propagate
to
clients
downstream,
and
the
other
input
was
a
multiple
errors.
Pacific
prefixes.
N
Obviously,
you
can't
use
the
genus
search
last
industry,
anything
because
if
you
would
just
make
this
a
transitive
through
the
CPE,
you
might
broke
something
in
the
local
network,
but
you
can
use,
for
example,
PTR
and
then
a
strap.
This
prefix
is
from
the
PTR
from
the
answer
of
the
PTR
query,
and
this
way
you
also
came
through
to
the
sorry
buddy
third
domain
over
your
provider.
N
A
There
we
have
a
problem,
and
that
is
there
were
actually
past
the
scheduled
end
of
the
meeting,
so
I'm
gonna
have
to
take
this
out
to
the
mailing
list.
You'll
get
that
discussion
on
the
mailing
loss,
tutor
Wow.
We
have
one
more
thing
before
everybody
rushes
the
door
and
I
know
I'm
going
to
preempt
you
here.
There
is
a
discussion.
That's
happening
right
now
in
OPSEC
they're,
having
a
what
a
second
working
group
last
colony
on
draft
IETF,
OPSEC
v6,
looking
at
operational
security
considerations
for
ipv6
networks
and
they're.
Looking
for
your
review.