►
From YouTube: IETF101-DHC-20180319-1740
Description
DHC meeting session at IETF101
2018/03/19 1740
https://datatracker.ietf.org/meeting/101/proceedings/
A
C
D
A
A
A
Anybody
have
any
agenda
bashing.
They
need
to
do,
or
is
everybody?
Okay
with
that
order?
Alright,
just
want
update
people
that
we
did
get
the
recharter
approved.
Thank
you
Suresh
and
the
isg
for
that.
We
also
have
updated
our
milestones,
and
so
we
have
a
few
things
that
we
would
like
to
potentially
get
in
for
this
month.
A
You
know
the
review
that
was
done
for
the
during
the
ITF
last
call
that
there
were
some
director
reviews.
They
were
also
the
iesg
reviews
ended
up
with
about
a
hundred
and
forty
comments.
They
there's
a
lot
of
comments,
but
you
know
some
of
them
were
like
there's
one
that
it
could
have
just
been
one
comment,
but
it
ended
up
being
like
30
comments.
Okay,
so
we
we
did
the
update
to
that.
You
know.
Most
of
them
were
minor
a
lot
we're
just
typos
were
like
cleaning
up
references
and
things
like
that.
A
One
major
change
was
suggested,
which
was
to
reduce
the
hop
limit
count,
which
is
how
many
relays
you
can
chain
together
from
32,
to
8
the
32.
You
know,
especially
they
got
32
going
one
way,
32
going
the
other
way.
Somebody
thought
that
was
kind
of
ridiculous
and
there
could
be
some
latency
issues
and
other
things.
So
you
know
and
I
don't
think
I
don't
think
relay
training
is
most
sites.
Probably
only
do
do
one
relay,
you
know,
I,
don't
know
of
many
that
do
more
than
one.
A
E
Yeah
so
I
searched
Krishna,
so
I
think
he's
likely
just
minor
things
like
just
some
checks
and
get
it
on
and
I
really
want
to.
Thank
you
guys
for
like
doing
the
spreadsheet
for
tracking
this
I.
Don't
know
if,
like
it's
like
visible
to
everybody,
but
it's
like
I
think
you
posted
on
the
list
right
and.
F
E
A
A
A
We
hopefully
will
get
one
approved
for
publication
and
there
are
two
working
groups
that
are
currently
active
that
are
both
on
the
agenda.
One
working
group
document
recently
expired
which
isn't
on
the
agenda,
which
is
the
bootstrap
options,
and
the
author
of
that
document
has
changed
employment,
and
you
know
he
he
may
revisit
the
work
in
the
future,
but
for
now,
unless
somebody
else
is
strongly
interested
in
taking
it
on,
it'll
probably
be
put
aside
and
just
be
a
dormant
working
group
document
to
be
awoken
at
some
point
in
the
future.
A
A
A
B
Hackathon
update
so,
as
you
know,
there
was
a
hackathon
over
Saturday
and
Sunday
at
this
time,
so
Bernie
next
slide.
This
time
we
had
two
projects,
the
first
one.
We
thought
that
we
would
try
to
implement
the
server
model
and
that
is
specified
in
the
in
the
draft
that
we
have
as
a
working
group
item,
we
made
a
lot
of
progress.
There
was
some
sharp
edges,
we'll
put
it
a
couple
issues.
B
They
are
in
the
issue
tracker,
so
I
will
be
talking
about
them
briefly
and
when
we
get
to
the
yak
model
discussion,
and
but
the
overall
perception
is
that
it
was,
it
was
reasonably
good.
So
we
wrote
some
code,
it's
not
yet
functional,
but
well,
it's
so
much
functional,
not
usable
yet,
but
we
managed
to
extract
some
of
the
information
and
the
other
project
that's
been
happening
was
that
there
was
a
first
implementation
of
the
three
to
15
beasts
started
from
scratch.
It
was
for
a
riot
system.
B
G
Okay,
however-
and
this
is
for
the
DHCP
for
over
six
so
suggest
option-
we
have
just
post
the
latest
version
on
the
dealer
tracker,
so
you
guys
could
just
find
it
on
the
website.
Okay,
next
act,
please
something
about
the
document
status.
It
is
currently
an
active
working
group
document
and
it
tends
to
update
the
RFC
75
a
night,
if
approved
and
actually
about
this
drafts
lifetime.
G
The
first
presentation
is
maybe
four
or
five
years
ago,
whatever
we
just
get
the
draft
back,
so
we
have
done
some
major
updates
since
the
working
group
adoption-
and
here
are
six
for
Bernie's-
reviewing
comments.
Okay
next
slide
well,
I
will
just
give
a
very
quick
overview
of
this
document.
So
the
motivation
is
that
the
current
software
provisioning
model
is
deterministic.
G
For
example,
the
ipv6
source
address
have
to
use
predetermined,
prefix
and
the
the
C's
ipv4
or
ipv6
and
ipv6.
These
have
to
be
bound
together.
So
the
conclusion
is
that
it
is
very
limited.
So
in
this
document
we
just
use
the
DHCP
four
over
six
transport
defined
in
RFC,
734
t1,
and
the
detailed
process
is
a
very
typical
two-way
communication.
G
The
client
will
just
request
the
server
the
DHCP
for
v6
server
is
ipv4
and
ipv6
configuration
I
includes
the
ipv4
address
the
PRS
ipv6
address
and
also
the
ipv6
spanning
prefix
hint,
which
is
optional,
and
then
the
client
we
just
communicate
the
ipv6
also
just
back
to
the
DHCP
for
v6
server,
and
this
mechanism
is
applied
to
the
post
light
we
four
six
and
maybe
so,
to
fulfill
the
process.
We
have
to
update
and
define
new
options.
G
The
option
that
we
have
to
update
is
the
option
s
for
6pr,
because
in
the
original
RFC
it
has
been
defined
that
it
should
be
used
with
specific
software
container
options.
So
in
this
document
we
just
remove
this
restriction,
so
it
could
appear
directly
in
the
in
the
process
and
we
also
need
a
new
TRC
pv6
option
to
carry
the
preferred
prefix
hint
for
the
client
to
bind.
G
G
H
G
Torture
comm
they
have
to
some
implementations
and
found
that
if
we
just
use
a
SCP,
v-6
option
is
very
complicated.
So
we
just
changed
make
this
change
next
slice.
Well,
here's
a
outstanding
issues.
We
want
a
new
title
here.
Currently
we
just
call
the
district
dscp
v4
over
TV
music
source
address
option.
This
is
because
it's
a
laksa
from
the
initial
version.
G
We
just
use
this
a
single
source
adjust
option
to
carry
three
different
kinds
of
information,
but
currently
we
have
two
new
options
and
one
updated
option
and
draft
defines
a
whole
provisioning
process,
not
just
the
options
so
our
course
or
sink
that
better
teto
is
required,
and
here
is
what
we
propose
to,
as
the
new
title
is
called
a
dynamic
soft
wire
provisioning
using
th
CPV
for
over
tcp
b6,
and
so
I'm
just
here
to
ask
the
working
group.
Are
there
any
comments
on
this
new
title.
A
G
A
F
We
have
that
info,
so
I
think
the
first
couple
of
paragraphs
are
something
in
the
introduction:
make
the
point
about
the
RC,
75,
98,
sort
of
pre
provisioning
options
and
and
how
this
differentiates
a
year.
The
differences
between
this
that
mechanism
and
this
mechanism
so
III,
do
take
your
point
about
the
double
die
now
or
three
dynamics
in
there.
If
you
count
all
of
them,
so
maybe
we
could
come
up
with
a
better
word
instead
of
dynamic,
I
guess
the
wider
question
is
I
mean.
A
A
G
G
For
the
next
steps,
actually
we
have
proof
of
concept
implementations
by
Deutsche
Telekom.
The
client
is
based
on
the
open
to
party
and
there's
always
the
Kea,
and
even
just
told
me
that
this
will
be
published
as
soon
as
possible
and
the
learnings
from
those
implementations
has
been
incorporated
into
the
latest
version.
So
we
think
is
ready
for
the
working
group
last
call.
A
A
Actually
I
wrote
some
notes
down
come
on
up,
but
about
the
draft,
and
that
is
that
in
the
3315
biz
we
have
a
new
table
for
and
that
we're
gonna
ask
I
Anna
did
to
do
which
just
says,
for
example,
whether
an
option
should
be
in
the
Oro
and
whether
it's
a
unitary
or
not
option.
Okay,
so
I
think
one
of
the
things
that
you
should
update
the
draft
to
say
as
to
update
that
table
based
on
3315
biz.
A
You
know
some
reference,
3315
biz
table
for
and
a
direct
I
Anna
to
change
those
entrance,
because
you
got
a
change
option.
92B
now
be
available
in
the
Oro
and
you
should
add
the
definitions
for
your
two
new
options
to
tell
I
Anna
what
to
put
in
the
table
for
those
entries.
Okay,
so
that's
that's
I
can
send
a
note
out
to
the
mailing
list,
but
I
think
we
should
you
know
that
should
be
done
just
so
that
it's
ready
when
3315
biz
goes
through.
J
Hello,
everyone
I'm
from
Tsinghua,
University
and
I
recently
took
over
the
dhcpv6
llamada
work.
Next,
please,
and
since
since
last
meeting,
we
hadn't
made
a
lot
of
progress
and
the
two
versions
have
been
published,
saying
and
thanks
for
en,
and
it
called
Cheers
for
setting
up
the
report
on
github
and
we
resolve
most
issues
from
Ian,
Burnie
and
mossy
next
place.
J
J
We
also
rican
structure,
the
network
range
definition
any
and
now
it
only
contains
address,
pause
pupils
and
hostile
reservations.
The
previous
reserved
addresses-
and
there
is
a
prefix
emerged
into
hostel
reservations
because
they
same
the
function
of
hostel
reservations
and
we
remodel
the
the
DUID
definition
and
now
it
contains
four
types
defined
in
three
three
one
five
days
and
a
divide
unknown
type.
J
Rapid
calm
to
pool
level
and
removed
some
and
necessary
paulien
nose
and-
and
this
will
be
further
discussed
in
later
slide
and
we
had
some
interactions
with
ITF
interfaces,
also
remodeled,
some
remodeled
most
option
definitions
x,
place
and
now
this
is
what
our
current
model
structure
is:
the
server
relay
and
client
import
options
and
types
next
place,
and
here,
thanks
for
thanks
for
Tomic,
who
had
who
discovered
the
data
tracker
feature
by
which
we
can
view
the
model.
The
tree
model
interactively
next
place
is.
A
Kind
of
nice:
it's
a
good
way
to
explore
the
model,
because
you
can
click
on
these
and
then
you
keep
going
down.
You
can
drill
and
expand
the
model,
and
it
shows
you
all
the
relationships
and
things
pretty
nicely.
I
think
this
is
a
this
is
tooling
that
is
being
done
for
all
yang
models.
It's
really
kind
of
nice
and.
J
F
In
fur
so
yeah,
this
is
a
common
originally
raised
by
Marcin
in
a
review
from
probably
18
months
or
longer
ago
about.
Why
is,
does
the
PD
functionality
have
an
enable
option
in
there,
whereas
the
address
pools
for
non
temporaries?
Don't
I'm
slightly
torn
on
this,
because
I
am
I.
Thinking
behind
this,
when
I
kind
of
you
know
looked
at
this
again
was
you
could
have
a
fully
compliant
RFC
3315
implementation
that
doesn't
do
PD
right.
A
F
F
F
Here
are
my
capabilities,
so
young
young
does
this
I
mean
this
is
part
of
it.
You
know
when
the
the
session
is
established,
then
there
will
be
this
capabilities
exchange,
but
the
capabilities
exchange
assumes.
If
you
say
you
support
this
model,
you
support
everything.
That's
in
this
model.
The
features
function
in
there
is
the
way
that
you
can
get
the
the
variance
from
that.
So
if
you
say
you
know
it's
perfectly,
you
know
this
part
is
optional.
You
can
still
be
perfectly
compliant
in
your
implementation,
but
not
to
do
that.
Then.
F
That's
why
the
features
function
is
in
there.
It's
why
we
move
to
doing
that
with
the
with
the
options
modeling
of
saying
you
know,
there's
whatever
150
options
in
there,
there's
probably
only
20
of
those
that
you
have
to
actually
do
in
order
to
say
I'm,
3315
compliant
or
whatever.
The
number
is,
but
it's
quite
a
subset
of
that.
So
we
went
down
the
road
of
saying.
Well,
if
you
don't
implement
some
of
the
later
ones.
Well,
you
can
turn
that
off
with
our
feature
still
be
a
compliant.
F
Dhcp
server
still
be
compliant
to
the
a
model
and
we
have
to
have
to
worry
about
any
of
those
things,
so
the
thinking
extends
beyond
that
to
say.
Well,
if
it's
done
with
prefix
delegation
again,
you
could
be
a
3315
server
and
not
do
PD,
so
maybe
it
makes
sense
to
move
that
to
being
a
feature
function.
Maybe
we
can
also
make
this
this
argument
about
address
pools
if
you
were
doing
purely
stateless
DHCP.
In
that
case,
you
know,
there's
no
reason
why
we
can't
do
that.
F
K
F
So
the
only
other
thing
that
I
thought
about
around
this
is
whether
you
would
want
to
have
a
situation
where
you've
had
pre
config
in
place.
I've
built
all
of
my
pools
here,
but
I
don't
yet
want
to
turn
this
on,
for
whatever
reason,
but
I'm
not
sure
whether
in
the
yang
world,
that's
actually
necessary,
I
mean
it's
it's
commonly
done
with
Reuters,
for
example,
pre
deployment
of
configuration.
So
when
someone
plugs
in
a
new
interface,
for
example,
it
pops
up
and
starts
working
I'm,
not
sure
if
there's
a
DHCP
equivalent
to
that,
but
yeah.
A
F
I
mean
yeah
I
think
we
are
getting
more
towards
this.
You
know
ironing
out
the
the
wrinkles
in
there
and
the
things
that
you
know
we
actually
experience.
It
looks
from
the
testing.
That's
been
done
so
far,
at
least
that
the
core
of
the
model
is,
it
seems
to
be
solid,
which
is
you
know,
gotta
be
the
first
step.
Okay,.
J
Thanks
next
sorry,
sorry
and
the
relocation
of
Duty
UID
and
a
rapid
comedy
issue
currently
for
server
and
client,
the
UID
is
defined
in
server
attributes
and
the
client
attributes.
But
actually
the
UID
is
exchanged
between
server
and
the
client
in
options.
Should
we
just
move
that
to
option
live
away
from.
G
The
seminary
City
as
for
the
location
of
the
2id
and
becoming
I,
would
prefer
to
put
all
of
them
into
the
options
module,
because
you
know
we
just
happen:
split
option
module
why
we
need
to
put
them
under
another
contender,
it's
a
little
bit
confusing
for
me,
so
I
personally,
I
would
prefer
to
put
all
these
things
in
the
standard
option
module
you
have
just
defined
as
another.
The.
B
Reason
why
to
treat
wrapped
option
is
that
it's
not
a
simple
option
that
is
being
sent
back
by
the
server
it
effects,
how
the
server
operates.
So,
in
my
opinion,
the
best
place
for
this
would
be
the
network
range
or
subnet.
So
because
this
is
more
like
a
property
of
the
network
that
is
running
the
server.
When
you
use
rapid
community,
you
say
that
I
am
the
only
server.
There
is
no
other
server.
B
G
A
It's
kind
of,
like
you
know,
I
kind
of
word
we
use
is
sort
of
policy
right.
This
is
whether
you
will
support
rapid,
commit
or
not
or
whether
you
want
to
send
a
set
of
options,
it's
kind
of
a
policy
type
of
thing,
and
so
we
in
our
server
we
kind
of
a
policy
container
that
has
all
of
the
stuff
about.
You
know
how
we
should
behave
for
this
set
of
clients
pools
server
wide
right.
We
have
policies
that
contain
options
that
contain
rapid,
commit
yes/no.
A
The
server
line,
the
DUID
for
the
server
I
think,
should
be
a
server
attribute,
because
that
I
don't
think
there's
an
immodest
as
I
mean
there's
nothing
that
prohibits
saying,
use
this
server
UID
on
this
network
segment,
this
other
one
there
but
I,
don't
know
what
benefit
that
would
ever
have,
and
so
I
think
that
should
be
a.
You
know
that
that
should
be
a
property
of
the
server
as
it
is
a
property
of
the
client
right.
A
To
go
back
and
look
at
the
model
to
see
what,
where
you
do
certain
things
and
how
they're
they're
done
I
mean
I
think
again
that
they
should
be
sort
of
in
this
option
bucket,
because
you
should
be
able
to
say
you
know
for
this
client
I
want
you
to
use
this
option
bucket
for
this
class
of
clients.
However,
you
want
to
define
that
class
I
want
you
to
use
a
set
of
options
for
this
pool
that
set
or
as
a
default
for
the
server
wide
here's
the
option.
J
A
My
mind
it
should
be
contained
in
and
things
whatever
container
we
hold
we
use
to
hold.
You
know
whether
the
options
is
the
right
word
or
not.
I
can't
say
right
off,
I,
don't
remember
the
model
that
into
intimately,
but
it
should
be
at
the
same
sort
of
level
where
you
can
define
the
set
of
options
that
are
supposed
to
be
different.
Okay,.
J
J
F
We
haven't
tried
yet
I
think
it
would
be
fair
to
say,
I
agree
that
it's
going
to
be
a
complicated
thing
from
a
modeling
perspective
and
there's
nothing
really
that
the
springing
to
mind
as
to
you
know.
This
is
a
potential
solution
that
we
can
try.
I
can't
believe
that
there
is
that
much
variance
in
'bitter
in
what
are
the
things
that
we're
making
selections
from
you
know
it's
a
sore
subject,
it's
an
interface
that
the
relay
originates
from
it's
relay
identifier.
It's
a
kind
of
client
identifier.
You.
A
Because,
like
you
know,
this
is
burning
goals
for
our
server.
You
can
actually
you
know
you
can
you
can
actually
write
an
expression.
You
know
bug,
there's
a
whole
bunch
of
data
together
to
produce
something
and
it
can
use
conditional
logic
that
says
you
know.
If
the
relay
is
this,
and
this
option
is
that
so
it's
very
complicated
and
also
for
the
client
class
in
this
pool
is
kind
of
too
simplified
because
we
use
a
level
of
indirection.
You
know
you,
we
do
have
the
concept
of
a
client
class,
but
for
pool
selection.
A
F
I
still
think
you
know
that
there's
a
lot
of
commonality
here
that
I'm
sure
your
implementation
could
do
many
more
things
and
there's
no
reason
why
you
couldn't
create
models
that
will
augment
with
specific
functionality
that
doesn't
exist
elsewhere.
But
if
we
can
get
you
know
the
things
that
will
work
for
90%
of
people,
then
it's
it's
worth
the
effort
and
and
I
don't
think
you
know
I,
don't
know
how
to
do
it
at
the
moment,
but
I'm
sure
it
is
possible
and
so
I.
B
So
I
suppose,
if
you
keep
the
description,
generic
enough,
you
could
do
it
probably
every
implementation
that
does
this
has
a
concept
of
class
name
and
some
sort
of
class,
quantifier
or
expression,
but
there
might
be
a
lot
of
additional
features
that
are
very
specific
to
the
server.
But
still,
if
you
provide
this
in
the
model
as
put
any
kind
of
string
here
it,
it's
gonna
work.
A
H
A
It
should
work
right,
whether
you
yep,
so
there
might
be
some
level
of
indirection
or
the
possibility
to
do
indirection
that
might
be
useful
I.
Don't
know
why
that's
just
thinking
off
the
top
nice,
so
I
think
it's
still
an
open
thing
that
we
have
to
debate
a
little
bit
about
how
to
do
those
things.
Yeah.
J
Yeah
I
think
so
next
place
and
next
will
first
we
will
continue
to
work
on
issues
and
github
and
and
notice
that
atomic
has
just
post
some
issues
from
the
from
the
Hexen.
Thank
you
and
we
will
further
check
the
correctness
and
the
completeness
of
the
option-
definitions
this.
This
is
easy,
but
it
it
it
consumes
a
lot
of
time
and
we
also
need
some
input
from
the
vendors
okay,
any
other
comments
or
suggestions.
F
One
very
well
I
found
a
very
useful
exercise
was
taking
some
real
config
that
we
have
and
know
that
we're
using
at
the
moment
and
trying
to
push
that
into
an
XML
rendering
of
of
the
model
as
it
stands
at
the
moment.
So
I
generated
a
skeleton
using
one
of
the
beiong
tools
and
basically
did
a
mapping
into
there
and
it
very
very
quickly
showed
out
there's
some
glaring
holes
in
there
of
things
that
are
missing
at
the
moment.
F
So
I
think
it's
a
worthwhile
exercise
there
to
look
at
config
examples
from
you
know
as
many
implementations
as
we
possibly
can
do,
do
a
similar
kind
of
exercise
there
and
I
think
the
inner
front.
From
that
there's
a
lot
that
can
be
learned
about.
You
know
how
well
it
works.
What
is
possible,
what
what's
missing
so.
B
Yeah
there's
also
a
next
step.
I
happen
to
get
the
XML
file,
which
the
configuration
and
then
try
to
actually
load
it
into
the
configuration.
There
was
a
bunch
of
other
issues,
so
yeah,
so
I
reported
a
couple
of
them
in
the
issue.
Tracker
I
just
want
to
briefly
mention
two
of
them,
so
the
first
one
is
that
by
default,
the
server
didn't
support
any
options.
So
if
I
wanted
to
specify
those
options,
they
were
rejected
by
the
model,
because
the
features
were
not
enabled.
This
is
not
how
DHCP
service
operates.
B
So
by
default,
the
server
can
sell
any
option
that
is
configured.
There
is
no
extra
switch
that
you
need
to
turn
to
to
define
those
options.
So
if
this
is
a
requirement
from
the
yank
perspective,
maybe
it
could
be,
but
it
was
just
very
annoying
that
even
the
basic
things
like
specifying
DNS
server
I
needed
to
turn
the
functionality
on
first
and
then
enable
the
the
option
itself.
I.
F
Mean
I
followed,
there
is
the
PM
tool.
Has
a
one
of
the
output
formats
is
an
XML
skeleton
which
it
will
just
go
through,
or
you
know,
pull
in
the
model
and
it
will
generally
generate
an
xml-based
template
that
you
can
potentially
go
and
fill
in
in
the
current
version
of
p.m.
that
is
just
completely
broken,
so
I
had
to
drop
back
to
an
earlier
version
in
order,
so
you
can
get
that
thing
to
work.
F
What
it
seems
in
there
is,
it
doesn't
come
out
with
any
of
the
that
it
doesn't
build
a
skeleton
that
has
everything
that's
in
there
and
if
certain
features
that
are
mandatory
to
true
in
the
model
didn't
make
it
over
into
there
and
you
know
so,
there
was
a
lot
of
hand
cranking
that
needed
to
be
done
to
get
that
thing
to
there.
The
other
thing
I
was
unable
to
do
is
find
any
way
of
validating
the
XML
I
had
created
has
been
against
the
yang
model.
F
To
say
you
know,
is
this
actually
and
so
yeah
I
mean
any
bugs
that
are
in?
There
were
probably,
as
a
result
of
you
know,
problems
in
the
tooling
and
problems
and
the
fact
I
had
to
hand
crank
it.
I
need
to
speak
to
the
PIO
authors
I
think
around
this
week,
so
and
kind
of
make
sure
you
know
ask
them
about
it:
okay,
cool
so.
B
And
my
second
comment
is
that
the
feature
that
is,
in
my
opinion,
missing
in
the
model
is
the
ability
to
configure
the
storage
for
the
server.
So
every
server
has
to
store
the
information
like
leases,
reservations
and
other
things,
so
the
implementation
I
am
involved
in
Kea.
It's
pretty
flexible
in
this
regard,
so
you
can
point
whether
you
want
to
store
them
in
local
file
in
a
database
or
similar
places,
which
you
know
the
credentials
for
the
database
and
so
on.
B
B
I'm
asking
I'd
like
to
introduce
a
new
draft-
that's
been
published
recently
like
two
weeks
ago,
so
on
so
first
some
background.
So
the
idea
behind
this
draft
is
to
be
able
to
configure
MAC
address
over
the
HCP
v6,
so
first
impression
that
people
get
is
what
what
so
bear
with
me
and
it
will
make
sense
in
a
minute.
So
a
couple
of
background
informations,
there's
FC
17
to
41.
It
defines
cooperation
between
I,
Triple,
E,
802
and
ITF.
B
There's
there
recurring
meetings
and
there
is
some
cooperation
between
those
two
entities
so
the
norm:
I
Triple,
E,
802,
dot,
C,
splits,
the
local
MAC
addresses
into
four
quadrants
and
one
of
those
quadrants
is
dedicated
to
local
marketers
assignment.
There's
a
working
group,
802
CQ
that
is
working
on
different
allocation
mechanisms,
so
couple
people
that
are
very
respected
and
experienced.
They
thought
about
this
and
they
came
to
a
conclusion
that
dhcpv6
might
actually
be
a
useful
mechanism
used
as
a
delivery
for
this.
B
So
this
cow
which
to
Bernie
Bernie,
talked
with
me
and
we
scratch
our
head
a
little
bit
and
said:
okay
yeah.
This
could
be
done.
So
if
there's
a
very
useful
presentation,
don't
part
by
part
Taylor-
and
it
explains
a
lot
of
the
reasons
come
on
so
but
I
still
wasn't
convinced
so
I
tried
to
convince
myself.
So
what's
the
problem?
Basically,
this
is
about.
It
means
the
paradox.
B
So
if
you
are
not
familiar
and
with
the
problems
that
right
now
in
this
room,
there's
two
to
one
people,
so
there
is
more
than
55%
chance
that
two
people
have
this
day
I
on
the
on
the
same
day.
So
but
it's
it's
all
nice.
But
if
we're
talking
about
things
like
virtual
machines-
and
you
want
to
create
lots
of
them,
there's
the
problem
that,
if
you
randomly
assign
mark
addresses
there
could
be
conflicts
not
very
frequently
but
sometimes,
and
so
if
you,
if
you
have
a
large
scale,
it
will
affect
you.
B
But
the
scale
is
not
that
big
to
be
affected
by
this.
So,
for
example,
if
you
use
a
classic
approach,
for
example
in
VirtualBox,
when
you
create
a
new
machine,
only
the
last
24
bits
are
randomized.
So
if
you
have
thousand
VMs
the
probability
of
a
collision
is
3%.
So
this
is
no
trigger
and
this
is
starting
to
go
up
so
to
reach
50%.
B
You
only
need
for
less
than
5003
abs.
So-
and
this
is
this
is
significant,
so
you
could
try
to
go
further,
so
use
almost
the
whole
market
less
space.
There
are
two
bits
that
are
reserved
the
the
first
one
determines
whether
this
is
unicast
or
multicast
others,
so
it
has
to
be
unicast
whether
this
is
global
or
local.
B
So
this
is
also
determined
and
if
you
obey
the
the
quadrants
that
are
defined
by
I,
typically,
you
get
with
you
get
45
or
44
beats
so
that
with
large
deployment
it
gives
you
probability
of
1.41
of
collision.
So,
but
this
is
a
problem,
because
this
is
a
probability
for
every
additional,
the
end
that
you
spin
up,
so
you
will
get
a
lot
of
collisions.
So
that's
why
there
is
the
need
for
for
a
central
repository
that
will
assign
the
addresses
okay.
B
So
what
are
the
use
cases
so?
The
the
fields,
the
fields
the
most
most
common,
in
my
opinion
use
case-
will
be
a
hypervisor
that
is
allocating
a
MAC
addresses
for
virtual
machine.
They
might
have,
depending
on
day
on
the
use
cases
and
the
business
models.
They
might
be
a
very
long,
lift
or
very
short
lift
and
it
be
recycled
frequently.
B
So
the
other
question
is
why
this
should
be
the
HP
v6
based.
So
the
quick
answer
is
that
there's
existing
infrastructure,
we
have
the
protocol
that
is
extensible.
We
have
the
support
in
networks
we
have
realize
deployed
that
could
already
deliver
the
traffic
and
do
some
things.
There
are
tools
that
are
built
around
the
HP
and
administrators
are
familiar
with
this,
so
it
should
be
easier
to
deploy
than
complete
new
protocol.
And
finally,
we
are
the
DHCP
guys.
So
when
the
when
you
ask
us,
the
answer
is
always
d2p
so.
B
B
This
is
slightly
different.
The
option
is
others
option,
so
it
is
used
to
convey
what
we
call
a
block
of
addresses.
So
it's
flexible
enough
that
you
can
request
and
get
a
single
address,
but
if
you
want
to
you
can
get
more
of
them
in
a
single
option,
so
the
link
like
address
type
is
just
the
type
who
was
induced.
This
is
Ethernet
or
something
else
link
layer
LAN.
This
is
almost
always
six
bytes,
but
there
are
some
exotic
architectures
like
infinite
InfiniBand.
They
are
using
20,
bytes
and
I.
B
Think
WiMAX
is
in
some
cases
are
using
8
bytes
long,
but
anyway
there's
an
address
that
you
can
specify
and
then
the
number
of
additional
addresses
that
are
objecting
to
it.
So
if
you
want
to
just
get
single
address,
extra
distance
is
0.
That's
it.
There
is
no
preferred
lifetime.
Just
about
lifetime.
B
B
B
So
and
of
course
the
hypervisor
doesn't
use
them
on
its
own.
Just
creates
the
VM
suite
with
those
addresses
and
for
the
actual
clients
like
IOT
devices,
there's
the
booster
problem
to
send
the
HTTP
packets.
You
need
to
have
Marc
address,
so
you
could
start
with
randomize
address
and
either
ask
the
server
to
assign
address
or
ask
the
server
whether
this
MAC
address
is
okay
or
not,
and
then
you
sort
this.
This
would
be
sort
of
like
a
registration
mechanism
so
and
that's
pretty
much
it.
There
is
one
more
slide.
B
So
we
have
some
questions
whether
we
should
use
rapid,
commit
or
not.
Couple
approaches
are
possible
that
it's
forbidden
it
can
be
allowed,
or
if
you
want
to
speed
up
the
process,
we
can
even
make
it
mandatory.
We
are
not
sure
whether
to
use
reconfigure
or
not,
and
what
do
what
the
hip
advisor
is
supposed
to
do.
It
server
becomes
unavailable,
it
tries
to
renew
and
finally
the
addresses
expire
so
just
to
wrap
it
up.
We
have.
B
E
B
A
L
M
A
Thank
you
and
I
think
we'll
take
the
you
know,
we'll
take
the
question
to
the
working
group
as
to
whether
there's
you
know
we
may
not
do
enough
adoption
call
right
away,
but
we
probably
will
do
one
sometime.
Maybe
once
we,
you
know,
there's
a
few
cleanups.
We
have
to
do
and
and
questions
to
answer
and
we'll
probably
send
out
a
few
queries
to
the
mailing
list
as
well.