►
From YouTube: IETF99-LISP-20170717-1330
Description
LISP meeting session at IETF99
2017/07/17 1330
https://datatracker.ietf.org/meeting/99/proceedings/
A
B
B
B
B
B
So
well,
we
have
to
talk
about
the
Lister
security
document,
but
you
didn't
move
that
along
and
so
will
unblock
also
the
introduction
of.
So
we
are
our
fault
a
little
bit
late
with
the
signal
team,
because
us
the
working
with
last
document,
but
we
actually
get
sloppy
and
we
just
pushed
there
for
the
publication
or
not
yet
for
publication.
We
need
a
shepherd.
I
will
do
with
the
data.
There
is
no
problem,
but
we
will
swim
push
it
to
the
highest
year
for
publication.
B
Okay,
just
think
we
came
a
little
bit
late
and
then
we
have
a
few
documents
that
that
you
see
with
the
purple
dots
on
which
we
we
are
working
with.
We
work
on
and
finally
publish
them.
Okay,
there's!
The
latest
adoption
now
asks
for
the
agenda.
Machine
I
will
start
with
us
a
little
bit
about
the
these
documents.
B
Okay,
we
will
skip
the
to
a3
yeah,
the
mobility
presentation,
because
finally,
there
is
no
okay,
okay
and
we
will
try
to
have
a
remote
participation
with
Rashad
trying
to
present
from
talking
I
guess
the
younger
okay,
whatever
maybe
yeah
yeah,
should
our
mouth.
We
want
to
keep
face
to
face
with
in,
but
for
today
evening
then
dinner
will
give
an
update
on
their
predictive
art
locks
document,
and
then
we
need
to
the
main
work
with
the
pythons,
and
you
know
we'll
continue
with
the
Eid
anonymity
document.
B
C
C
D
D
And
then
we
also
had
commented
me
from
being
that
we
were
a
mistake
and
we
changed
the
message
from
two
weeks
to
unreachable
fragmentation
needed
which
I
write
one
for
a
team
so
next
night,
then
that
for
the
control
plane,
we
have
four
iterations
seems
last
idea
and
then
to
discuss
again
all
the
changes
so
first
module.
We
clarified
so
now
the
control
plane
does
some
SMR
and
after
proving
our
part
of
data
plane,
which
means
that
the
control
brain
does
not
offer
any
services.
D
So
if
another
the
plane
wants
to
use
the
list
control
plane,
they
need
to
either
implement
those
by
themselves
or
use
or
reuse.
The
existing
data
plane
there's
a
margin
of
profit
mechanic.
So
we
clarified
that
that's
a
requirement
for
any
I
have
been
willing
to
willing
to
your
second,
then
we
also
change
at
the
register
file.
Now
it
is
I
instead
of
a
16-bit
field
to
a
bit
field,
while
the
hello
it
did
represent
the
key
ID
and
the
high
in
to
present
the
algorithm.
The
algorithm
which
is
used
for
entity.
D
And
then,
since
we
moved
all
the
IANA
considerations
from
that
event
to
the
control
plane
because
they
were
actually
pointing
to
controlling
this
is
how
it
is
right
now.
So
the
under
consideration
for
the
control
plane
would
have
a
four
type
of
message.
So
first
is
the
load,
control,
plane,
type
type,
which
is
not
registered
map
and
defied
request
a
free
transfer.
D
So
we
remove
the
listing
for
request,
which
was
not
on
the
agenda
of
original
RFC,
then
type
code
19
to
14,
they
are
not
assigned
and
they
are
at
Vienna
to
assign
them
if
needed.
And
then
there
is
the
leaf
shark,
essential
message
which
is
now
on
an
eye
on
a
registry
with
the
first
come
first
serve
policy,
and
this
is
to
allocate
subtypes
for
experimentation
with
it
with
Newman.
Then
we
have
the
list
act
and
black
teal,
which
allocating
new
one
means
that
you
need
to
vote
idea
view
or
idea.
D
So
this
means
that
if
you
want
to
operate
and
you
would
have
to
meet,
you
have
to
write
a
mistake
and
then,
since
we
change
that
the
name
of
the
field
for
PID
which
changes
the
name
for
the
registry,
but
we
requested
I
amateur
changeling
for
the
registry
to
this
algorithm-
maybe
a
number
which
is
first-come
first
ex-police.
If
someone
wants
to
allocate
any
so
next
leg.
D
So
it
is
how
is
trying
to
summarize
so
packet
subtype
9
for
dinner,
I
on
assign
and
then
type
16
subtypes
are
assigned
by
iana.
First
come
first
act
on
flag
type.
Her
idea
review
is
the
approval
ELCA.
The
specification
required
and
are
very
maybe
is
again
first
comfort,
and
these
are
two
small
changes
of
you
are
planning
for
the
next
version,
which
is
the
first
one,
is
changing
world
citations
to
24:16
with
82
all
which
is
the
new
ipv6.
F
Do
you
have
recommendations
on
the
ones
that
are
curly
about
to
be
published?
Should
they
point
to
the
best
documents
or
no,
let
them
move,
and
they
point
us
at
sixty,
eight
thirty
and
sixty
okay
and
any
okay,
so
the
problems
I've
had
is
any
new
internet
drafts
that
want
to
point
to
something.
That's
specific
to
this
should
point
to
these
right.
G
So
this
is
the
update
on
the
young
model
track.
I
think
the
best
thing
I'm
coming
back
three
weeks
on
vacation,
so
I'm
kind
of
disconnected
the
best
thing
is
that
probably
the
detail,
the
victim
will
be
done,
probably
in
the
next
meeting
or
maybe
possibly
in
the
mailing
list
and
I
just
give
you
of
the
update.
So
this
version
has
a
dress
that
changes
with
the
change.
G
D
H
G
B
F
Can
hear
me
in
the
back
okay,
something
to
talk
about
predictive
outlets?
It
just
became
the
working
group
document,
the
document
status
he
originally
submitted
at
last
spring
2016.
We
updated
the
timer
in
November
and
also
updated
again
in
May
of
this
year,
and
he
needed
a
working
group
document
in
June
column
statements.
F
So
the
mobility
problem
is
pretty
simple.
When
the
knee
ID
moves,
it
just
simply
sends
packets
to
a
new
location
right.
Well,
not
really.
This
packets
that
are
already
in
the
network
are
going
to
the
old
location
where
the
PPID
no
longer
is,
and
the
key
ideas
arrived
at
the
new
location,
but
the
signalling
system
doesn't
know
it's
very
F
in
packets.
The
sender
doesn't
know
about
the
move,
yet
so
packets
are
are
not
moving
yet,
even
though
the
Eid
is
there.
F
F
So
mobile
IP,
you
really
can't
send
in
a
home
agent
because
it
doesn't
know
where
the
location
is,
and
you
have
to
do
that
as
quickly
as
possible.
Post
routes
can
tell
you
the
location
of
what's
being
moved,
but
it
requires
the
host
route
to
be
propagated
throughout
the
network,
and
so
here's
officer
slow
because
the
Eid
host
route
has
to
go
everywhere
because
queuing
delays
it
as
processing
delays.
F
Locator
ID
separation
can
be
used
if,
if
signaling
can
be
fast,
so
we
tried
to
make
signaling
as
fast
as
we
can
Lisp
in
general,
but
we
think
we
can
make
it
go
faster
by
not
doing
it.
So
what
we
want
is
we
really
want
near
zero
packet
loss?
You
really
don't
want
to
drop
any
packets,
and
you
want
these
four
handoffs
to
be
instantaneous.
How
do
we
do
it?
Well,
the
future
can
be
somewhat
clearer
if
you
know
where
all
the
new
locations
are
going
to
be.
F
So,
let's
have
a
source
send
packets
to
all
the
new
locations.
The
roaming
Eid
is
going
to
be
we'll
search
and
find
where
the
CID
has
Rome
to
by
getting
feedback
when
it
sends
back
so
we're
exercising
here.
A
fan
with
signaling
trade-off.
We're
not
gonna
signal
we're
going
to
use
the
data
plane
and
bandwidth
to
find
where
the
new
location
is.
This
is
how
it
can
work.
So
let's
say
we
have
these
2d
IDs
a
and
B
they're
going
down
the
road.
Let's
call
the
road
main
street.
F
Let's
say
that
Main
Street
is
an
eID,
that's
registered
with
mapping
database
with
our
Lopes
a
b
c
and
d.
Those
are
boxes
that
are
along
the
street
cut,
let's
just
call
them
roadside
units
like
the
IP
way,
people
call
it.
We
each
have
our
looks
a
b
c
and
d.
Let's
say
e
ID
points
the
main
street
and
so
does
a
ID
b,
which
means
they
both
can
share.
This
particular
replication
list
note
that
Main
Street
is
mapping
to
an
our
local
record.
That's
called
our
elite.
That's
replication
list
entry.
F
F
F
So
what
our
look
s
would
do
when
the
server
side
Eid
source
is
a
packet,
is
it
will
send
the
packets
to
all
four
places
in
the
case
where
it's
looking
for
Eid
a
if
the
server
sunny
di
da
the
packets
replicated
all
four
places?
So
what
happens?
Is
the
packet
arrives
at
all
the
XT
RS,
and
if
the
ID
is
discovered,
the
packet
is
forwarded
by
that
x,
TR
to
the
Eid,
so
in
this
case
a
sees
that
Eid
a
is
attached
to
it,
so
it
just
forwards
the
packet,
B,
C
and
D.
F
Don't
know
that
Eid
a
is
attached
to
it,
so
it
just
drops
the
packets
in
the
case
where
B
the
server
sending
to
be
it,
notices
that
it's
receiving
packets
back
from
it
from
B.
So
it
can
actually
do
a
low
optimization,
do
not
have
to
replicate
packet
state
only
to
where
it's
at
and
where
it's
about
to
go
to,
so
we
can
have
a
pruning
decision
now.
This
is
all
done
without
any
type
of
signaling
whatsoever.
The
I-x
TRS
looks
up
math,
Main
Street
or
looks
up
AI
da
or
B
finds
Main
Street.
F
F
F
So
we
think
we
can
maybe
even
work
for
intersections
okay.
Let's
say
we
have
this
imp
this
on
the
street
called
Wall
Street,
which
is
basically
an
intersection
between
Main
Street
and
let's
say
that
the
Wall
Street
mapping
entry
is
this
distinguished
name
called
Wall
Street
and
it
has
a
nested
RLE.
So
that
means
when
the
Eid
a
starts
off.
We
assume
that
it
might
be
going
straight.
So
you
see
a
and
B
is
the
first
element
or
the
first
two
elements
in
the
r-la,
and
then
you
see
at
the
bottom
at
C
and
D.
F
F
F
So
what
were
the
list?
Protocol
changes
required
to
make
this
happen?
Actually
there
were
no
protocol
changes.
Those
no
spec
changes
at
all
to
the
existing
specs,
because
the
AR
le
l
calf
is
already
used
for
multicast
at
end
replication.
We
were
just
now
using
it
for
a
unicast,
the
ID,
rather
than
a
multicast
as
comma
G
and
multi
multicast
will
just
work
the
roaming,
the
ID.
F
When
the
roaming,
the
ID
is
a
receiver,
it's
just
a
GMP
reports
and
then
those
guys
just
signal,
but
now
they
don't
have
to
signal,
because
if,
if
you
know
where
the
the
Eid
is
roaming-
and
you
know
what
groups
it's
going
to
go-
and
you
can
just
do
this
ahead
of
time
now,
when
the
roaming,
the
ID
itself-
is
a
source
for
a
multicast,
it's
the
SE
ie.
It
cannot
be
prefetched.
I'll
talk
about
prefetching
in
a
minute.
F
Okay,
so
this
smoke,
this
comment
on
your
lower
XD,
ours,
concrete
prefetch,
our
look
s.
So
when
a
is
roaming
down
here
and
B,
C
and
D
are
getting
packets,
they're
dropping
packets
on
the
floor.
But
there
was
a
usefulness
of
this
packet.
It
actually
identified
the
source,
the
ID
and
what
it
can
do
now
is
look
up
that
source.
So
it
has
a
map
cache
entry.
F
F
F
F
Let
me
explain
4
types,
that
we
have,
that
the
directional
one
is
the
one
I
just
explained,
but
let's
say
you
have
an
airily
with
ABC
in
DNA
and
the
way
it
works
for
all
usage
types
is
that
you
would
replicate
to
all
four
of
them
and
then
you
would
solicit
some
feedback
to
prune
or
decide
which
one
of
ABC
or
D
you
should
do.
You
should
replicate
to
next.
F
F
The
ITR
has
a
list
in
its
map
cache
of
a
B
and
C
and
D,
but
there's
flag
set
that
says
it's
only
gonna
replicate
the
C
and
D
a
and
B
is
kept
there
because
there
might
be
a
usage
type
where
it
may
have
to
use
a
and
B
later
in
the
random
usage
case,
when
you
send
to
all
of
them.
But
then
you,
when
you
get
feedback
only
from
one
of
them.
F
F
Okay,
we
did
not
document
the
prefetch
that
I
just
discussed,
but
we'll
document
that
in
0
1
and
that
that
means
when
RS
use
get
packets
from
the
initial
replication.
They
do
a
look-up
on
the
source
Eid
to
populate
their
map
cache.
So
there
are
no
pocket
packet
drops
when
they
appear
on
the
new
XTR.
F
Other
words,
this
is
kind
of
research,
you
sort
of
work,
but
maybe
what
can
we
do
more
things
to
prune
that
list
because
of
feet?
If
the
hourly
list
is
going
to
be
long,
if
there's
roadside
units
every
you
know,
one
kilometer
and
cars
are
going
really
fast.
Will
we
replicate
too
long
I
mean
so
we're
looking
at
ways
of?
Should
we
use
multiple
our
local
records?
That's
the
third
bullet
and
concatenate
them,
or
should
we
use
overlapping?
F
So
we
don't
have
to
have
these
long
hourly
lists
where
we're
replicating
the
more
places
than
we
knew
you
need
to.
Okay,
can
we
use
geo
prefixes
and
see
where
the
roaming
Eid
is
right
now,
but
probably
using
geo
prefixes
when
it
changes
its
location?
That
location
has
to
be
reported
with
a
control
plane
to
some
database
and
if
we
have
to
look
it
up
its
defeating
our
purpose
right
if
geo
coordinates
can
be
part
of
the
data
packet,
then
that
that's
something
that
we
could
use,
but
there's
no
specification
to
do
that.
F
Yes,
and
not
necessarily
we're
not
necessarily
recommending
yet
can
we
put
our
key
ours
close
to
the
EP
RS,
so
the
replication
from
the
head
end
doesn't
have
to
be
N
and
we
just
replicate
once,
which
is
closer
to
Main
Street
and
then
have
it.
Replicate
there
to
minimize
the
head,
end
replication
problem,
and
can
we
use
a
level
of
indirection
with
distinguished
names
for
grouping
roaming,
ie
IDs
to
reduce
the
predictive?
F
Our
look
what
I
mean
there
is
in
my
example,
we
had
both
a
ID,
a
and
the
ID
be
points
of
Main,
Street
and
Main
Street
had
the
IRL
e,
rather
than
putting
that
same
hourly
in
both
a
and
B.
So
that
was
just
the
way
of
not
having
to
duplicate
the
information
in
multiple
mapping
entries
and
then
cuy
useless,
crypto.
F
Well,
we
can
use
loose
crypto
by
default,
just
like
we
specified
it
in
the
signal
for
multicast
days,
where
you
encrypt,
with
a
different
key
to
each
head,
end
replication
and
if
you
want
it
to
only
encrypt,
wants
and
replicate
too
many.
You
need
a
shared
key,
which
is
not
specified
and
is
typically
viewed
as
being
not
very
secure.
Evasion
of
keys,
shared,
not
share
keys
between
two
points,
shared
keys
between
more
than
two
points.
J
Victorian
so
the
earliest
seem
to
have
directionality
in
the
examples
you
gave
how,
if
you're,
going
on
Main
Street,
east-west
and
now
you
come
in
same
car,
it's
going
to
go
back
home
any
thoughts
on
how
you
can
actually
express
that,
with
with
your
mapping,
requests
and
get
if,
in
hourly,
that
is
going.
F
If
you
use
this
circular
one
or
the
back-and-forth
usage
type,
then
you
can
just
put
ABCD
in
there
and
replicate
all
of
them
and
find
out
where
you
first
hear
from
it
would
start
speaking
you
hear
from
the
D.
You
know
it's
going
west,
you
hear
from
it
from
a
you
know:
it's
going
Eve's,
meaning
it's
sending
packets
with
a
source,
a
source,
our
look
of
D
or
a
so
you
know
where
it's
starting,
so
there's
all
kinds
of
things
we
could
do
there.
K
I
just
wanted
to
add
something:
is
there
something
else
that
we
would
thinking
about
these
thumb,
Alex
being
actually
added
by
a
third
party
which
might
actually
also
be
having
more
information
to
watch?
What
is
useful,
if
necessary,
it's
not
as
read
the
Eid
point
itself
as
registering
all
the
promotion.
F
J
J
F
Say
that
my
intersection
case
was
three-dimensional,
neither
no
I
was
drawn
on
a
two-dimensional
slide,
so
you
could
have
a
plane
that
goes
in
the
z
direction
and
then
you
would
have
more
nested,
er
least
hourlies
like
me,
when
you
came
over
a
certain
direction,
you'd
have
to
send
down
here
and
down
there
as
well
as
there
and
there.
So
you
could
do
now.
People
have
asked,
can
you
use
this
in
airplane
applications
where
you
don't
know
exactly
where
you're
going
and
what
you
do
is.
F
F
K
K
I
I
B
F
L
F
F
We,
what
it
meant
was
we
replicate
to
a
3d,
and
we
don't
know
that
the
direction
that
the
Eid
is
going
in
if
it
was
at
a
we'd,
have
a
good
opportunity.
That's
gonna
be
at
be
next
for
me,
assume
directionality,
but
is
when
it's
random.
It
could
be
a
see
any
that's
what
we
meant
Padma
tadmor,
and
what,
if
you
could.
K
They
said
I'm
not
an
interesting
case
about
this-
is
that
even
though
we
say
we
have
a
long
list
and
you
can
replicate
to
all
of
them
either
most
of
the
time,
even
if
you
look
at
if
we
look
at
solar
phones
or
anything
like
that,
don't
actually
handing
over
from
one
Sun
tower
to
the
other.
So
at
one
point
or
another,
you
could
restrict
them
to
only
the
Quonset
of
the
real
target.
K
So
all
this
implies
that
you
have
you,
don't
more
machine
me
behind
this,
which
is
actually
updating
it,
but
this
is
very
doable
because
they
do
have
the
information
anyway.
But
I
don't
want
to
go
specifically
for
that
mobility
games,
but
you
can
think
about
other
cases
where
you
might
use
that.
So.
F
F
F
J
F
F
F
F
Yeah
but
I
would
say
if
it's
truly
random
or
unpredictable,
you
should
send
to
all
of
them
and
never
prune
the
list,
because
you
don't
you
don't
know
where
it's
gonna
come
from,
but
anyways,
but
we
could
do
that
too,
but
so
it
sounded
like
there's
some
input
that
maybe
random
is
not
useful
or
not
understandable.
We
could
certainly
prune
it
hasn't
been
written
up.
Yet
some.
M
I
think
because
think
about
what
the
Google's
the
Facebook's
is,
they
just
have
a
does
in
the
base
of
sure
locations.
Also
in
the
simpler
SNMP
area,
we're
putting
in
geolocation
information
should
be
a
Stata
into
the
SNMP
strings,
to
figure
out.
They
are
the
devices.
On
the
other
hand,
all
your
mobile
devices
nowadays
they
all
have
GPS
in
any
way.
M
So
I
don't
know
if
it's
really
hard,
like
with
us,
a
title
implementation
to
pull
the
GPS
information
out
of
the
GPS
deck
and
put
it
into
the
list,
and
then
I
have
GPS
data
to
GPS
data,
so
just
move
there
are
links,
take
from
GPS
the
GPS
and
always
have
directions
already.
We
are
using
this
for
GPS
navigation
systems,
yeah.
M
F
M
F
It's
a
URI
100%,
but
how
does
the
ice
ITR
get
the
information
when
the
Eid
moves
from
our
log
be
to
see
how
does
it
know
that
if
a
physical
move
takes
place?
Yes,
these
parameters
are
updated
in
the
mapping
database.
Now
the
ITR
has
to
know
to
encapsulate
to
see
it's
got
to
query
the
mapping
database
that
we're
avoiding
that
whole
process
altogether
B.
We
are
claiming
it's
too
long.
F
You
go
put
the
GPS
in
the
if
you,
but
if
all
those
parameters,
if
a
B
and
C
and
D
can
be
programmed
ahead
of
time
like
we're
specifying
here
and
this
ABA
and
its
geo
coordinates
its
to
your
coordinates
as
you're
saying
then,
when
somebody
moves
and
the
data
packets
being
said
in
the
opposite
direction.
If
the
Geo
point
was
put
in
the
data
package.
F
F
K
To
you
know,
I,
don't
what
he's
saying
the
way
I
see
it
is
if
we
had
those
our
lives,
which
are
those
IP
addresses,
but
actually
have
a
cooperator
of
his
certain
range,
then
de
jure
appropriate
hazard
sense.
You
can
say
he
shot
out
of
this
range
of
your
coordinates.
Your
vertical
move,
as
you
know,
which
direction
you're
moving.
Then
you
can
actually
use
it
and
saying.
Okay
and.
K
K
F
The
the
the
I
think
the
missing
piece
that
a
solution
is
that
the
moving
entity
has
to
advertise
directly
to
the
replicator.
What
it's
geo
position
is
and
if
we
can
put
it
in
a
data
packet,
that's
part
of
the
encapsulation,
then
it
can
work.
But
you
know
that's
gonna
increase
the
header
size.
If
you
could
send
me
a
packet
telling
me
where
your
geo
position
is
at
this
point
as
part
the
encapsulation
payload,
then
I
know
exactly
that.
Where
you
move
right
and
then
I
can
use
that
information
for
something
else
very.
M
F
F
Okay,
so
the
document
status
here
is
this
was
posted
in
April
of
last
year
as
well,
and
in
October
we
did
a
timer
update
and
then,
in
April
of
this
year
we
did
an
update
to
describe
that.
Efemer
IDs
can
use
public
key
hashes
as
well
as
a
random
number
I'll
talk
about
that
a
little
bit
more
oh,
and
we
can
also.
F
F
Solution:
well,
if
a
source
creates
an
ephemeral,
a
ID
and
start
sending
packets
from
them,
that's
how
we
can
do
it.
In
other
words,
a
particular
house
doesn't
have
any
ID,
that's
permanent
to
it.
It
changes
it
over
time.
An
ephemeral,
Eid
is
going
to
be
an
ipv6
address
and
the
reason
is
is
to
make
that
random
enough
or
to
carry
hashes.
You
need
enough
bits
in
the
address,
so
maybe
part
of
the
ipv6
address
space.
F
So
if
I
was
a
client
accessing
Google,
the
Google
I
ID
would
obviously
be
registered
in
DMS,
but
I
could
change
every
connection
to
make
to
Google
I
could
change
my
ID
and,
of
course,
the
extr
deaths
directly
connected
to
me.
Just
discovers
new
ID
registers,
the
mapping
system,
when
it's
our
love,
so
return
traffic
and
come
back.
It's
pretty
it's
as
simple
as
that.
What
were
the
list
protocol
changes?
None
all
we
had
to
do
is
just
use
the
mechanisms
in
Eid
mobility
and
the
ID
mobility
draft.
F
When
an
IP
address
comes
and
goes,
the
xtr
just
knows
that
it
moved
on
to
it.
It
doesn't
know
that
it
was
a
stationary
node
that
has
created
a
new
Eid,
it's
the
same
machinery
that's
being
used,
and
so
when
the
source
stops
sending
from
ephemerally
ID,
it's
just
like
when
the
source
moves
away
from
the
next
er.
It's
a
moved
away
event,
so
it
can
be
register
as
well,
and
then
all
traces
of
the
idea
removed
from
the
mapping
system
and
map
caches
and
remote
ITRs
are
invalidated.
It's
pretty
simple
situation.
F
F
F
F
So
what
we
need
to
do
on
our
to-do
list
is
we
want
to
document
ephemerally
ID
address
collisions.
So
if,
if
the
number
of
bits
of
the
random
number
is
small
in
the
field,
so
you
have
enough
hierarchy
in
the
higher
of
the
bits
where
you
really
don't
have
a
long
random
number
there
might
be
collisions
and
what
happens
with
collisions.
F
The
problem
is
here:
is
that
every
time
you
change
in
a
family,
ID
de
as
a
hash
in
it,
you
have
to
change
the
key
pair
for
the
system.
Is
that
practical?
If
you
want
the
be
temper
to
temporal
aspects
of
Eid,
does
it
make
sense
and
is
it
deserve
the
management
of
doing
key
pairing
all
the
time,
because
once
you
rekey
a
system,
you
have
to
make
its
public
key
available
to
other
people
that
want
to
verify
signatures
or
encrypt
to
you.
So
there's
a
bit
of
an
overhead
there.
Okay
questions
this.
F
A
policy
in
the
source
whenever
it
wants
to-
maybe
maybe
a
source,
says
I-
want
to
use
an
e
ID,
/
TCP
connection.
So
every
new
TCP
connection,
I
established
I'll,
go
ahead
and
create
a
new
ephemeral
e
ID
and
then
and
then
create
the
TCP
connection
based
on
that
and
then
e
TCP
connection,
I
terminate
then
I
stop
using
it.
So
whatever
the
source
wants
to
do
it,
it's
that
it
doesn't
have
to
be
a
list
defined
policy.
F
D
G
F
C
D
F
B
B
F
F
You
know
all
the
rate-limiting
stuff
in
the
mapping
database
still
applies,
but
what
it
will.
Typically,
what
we
can
do
is
on
the
guy
who
beat
ku
so
so
Victor
is
a
co-author
of
the
IP
mobility.
What
we
could
think
about
is
when
somebody
moves
on
to
you
as
an
ex
TR,
and
you
register
these
dynamic
e
IDs.
F
The
question
is:
is
how
many
will
you
register
per
second,
and
so
it
might
not
be
a
mobility
situation,
because
the
number
of
VIPs
already
exists,
but
they're
just
moving
back
and
forth,
like
I
guess
at
1
billion
IOT
devices
move
on
to
the
next
er.
This
is
a
problem
which
is
no
different
than
the
comment
you
made.
1
billion
needs
to
be
created
in
short
time,
so
maybe
we
have
to
look
at
scaling
problems
in
general
on
how
to
do.
G
Well,
you
know
it
depends
from
what
you
want
to
do
in
the
sense.
For
example,
there
are
some
solutions
for
anonymous
routing
right,
like
Onion
Routing,
that
they
go
to
a
very
long
extent
right
and
they
try
to
use.
You
know
different
level
of
encryption
to
kind
of
prevent
that
anyone
in
the
system
can
infer
anything
about
the
identity
of
the
sender,
so
those
I
think
would
be
interesting,
consideration
right
and
saying:
okay,
let.
F
G
D
G
G
F
I
mean
Oh
a
solution,
it's
not
an
end-all,
but
it
helps
the
problem
is
you
could
have
a
set
of
arlok's
on
the
xtr
yeah
a
large
set
and
then
the
ID
random
e
IDs
could
be
encapsulator
supported
by
the
different.
Our
looks,
you
know
and
I
don't
and
then
you
don't
know
if
all
those
our
looks
are
signs
at
the
same
place
or
there's
and
different
our
looks
or
n
different
XT
are
spread
across
the
network
right.
So
those.
J
F
Yeah,
it's
purely
a
random
number
in
the
IP
address,
and
the
only
thing
that
makes
it
attach
to
location
is
the
source,
our
log
in
the
encapsulated
packet
right.
Now
you
don't
know
if
there's
you
all
you
know
is
that
there
could
be
multiple
e
IDs
behind
it.
You
know
if
they're
separate
systems
or
one
system
doing
what's
presented
here,
but
if
we
could
mash
it
up
a
little
bit,
it
would
be
better
and
certainly
obvious.
F
N
F
F
A
G
Okay,
so
this
is
a
draft
Lisbon
Doyle
capillaries
natales
metal
cap
version
zero.
So
the
motivation
for
this
art
is
very
simple.
While
we
have
been
implementing
leaves
in
various
use
cases,
there
are
some
metadata
that
we
need
to
map
that
are
kind
of
very
specific
to
the
use
cases
that
we
are
long
and.
G
E
G
So
we,
what
we're
proposing
here
is
a
well
okay,
let's
define,
let's
just
define
a
veined
or
organizationally
unique
l
cap
space,
and
we
can
basically
use
that
private
l
cap
space
within
each
organization.
So,
if
I'm
okay,
so
as
this
kind
of
requirement
can
can
define
and
use
internally
to
that
same
event
taken
that
specifically
and
we
avoid
collisions
across
different
angles.
So
this
thing
are
very
private
close
and
are
used
within
one
particular
domain.
A
few
application.
G
As
I
said,
you
know
that
the
most
urgent
request
that
we
see
is
for
fiber
deployment,
but
also
you
know,
experimentation
and
research
or
rapid
prototyping
is
something
that
you
know.
If
you
want
to
play
with
in
your
own
space
of
a
cup
types,
you
can
do
it's
very,
very
easy
and
from
the
proposal
encoding
point
of
view,
we
leverage
the
UI
the
Icicle.
G
D
F
G
B
B
I
I'm
Fred
temple
and
I
work
for
Boeing
I've
been
attending
a
standards
body
known
as
the
International
Civil
Aviation
Organization
working
group
I
and
at
working
group
I
we're
talking
about
an
ipv6
network
for
civil
aviation
for
the
future
of
manned
aviation.
So
this
is
where
I'm,
coming
from
from
the
perspective
of,
what's
known
as
the
aeronautical
telecommunications
network
or
the
ATN
IPs,
is
called
so
again.
The
International
Civil
Aviation
Organization
is
building
an
ip-based
network
system
for.
I
Air
traffic
management
they're
deploying
something
that's
called
aeronautical
telecommunications
network
with
Internet
Protocol
services
or
etienne
IPS.
It's
under
the
development
of
ICAO
working
group
I,
it's
ipv6
based
and
mobility
capable,
and
it's
going
to
accommodate
aircraft
that
have
multiple
available
data
links.
Aircraft
typically
have
many
data
links,
onboard
satellite
links.
L
Dax
is
an
LVN
cell
tower
based
service,
VHF
data
links
and
others.
I
So
we're
looking
at
three
different
candidate
solutions
for
mobility
right
now,
proxy
mobile
ipv6
is
one
of
the
candidates,
something
that's
called
ground-based.
Lisp
is
another
candidate
and
then
there's
also
a
BGP
based
overlay
system,
which
is
the
subject
of
this
document
now
BGP
and
lists
by
no
interact
and
I'll
get
into
that
later.
So
it's
not
like
BGP
is
different
from
Lisp.
It
might
be,
including
less
so
I've
written
a
document
called
a
simple
BGP
based
mobile
routing
systems,
we're
not
a
communications
network.
I
It's
an
BGP
overlay
network
that
does
not
interact
with
the
global
public
internet
BGP,
routing
system,
it's
based
on
a
hub
and
spokes
arrangement,
with
stub
autonomous
system,
border
routers
and
the
datalink
provider,
sub
networks
and
core
autonomous
system
about
border
routers
in
the
in
the
core
of
the
network,
the
stub
autonomous
system,
border
routers
advertise
and
withdraw
airplane
mobile
network
prefixes.
This
would
be
the
same
thing
as
an
eID.
Perfect
use
the
list
of
terminology,
the
core
autonomous
system
border
routers
in
a
hub
autonomous
system.
I
Ford,
it's
between
the
sub
autonomous
system,
border
routers
and
route.
Optimization
can
remove
the
core
of
this
kind
of
system,
border
routers
from
the
path.
So
how
it
looks
like
is,
is
that
you
have
these
ew
d
based
sub
networks
that
the
airplanes
connected
these
things
here
and
also
air
traffic
controllers
connected.
I
That
would
be
the
core
and
EGP
bgp
is
used
between
the
stub
autonomous
system,
border
routers
in
the
core
and
within
the
core.
I
bgp
is
used
between
those
core
routers,
so
the
details
of
the
BGP
system
is
that
stub
autonomous
system
border
routers
advertise
their
associated
a
ID
based
mobile
network
prefixes
to
the
core
autonomous
system
border
routers,
the
core
autonomous
system,
border
routers
originate
default,
but
they
denied
advertise
any
MNPs
to
the
stuff
autonomous
system,
border
routers
and
also
each
stub
autonomous
system.
Stub
sub
network
is
assigned
a
med
value.
I
The
core
autonomous
system
routers,
are
disabled,
das
path,
selection
method
and
select
routes
based
on
the
med
values,
and
that
means
that
packets
may
take
a
longer
path
in
order
to
reach
a
stub
network
with
a
lower
net
value.
So
the
reason
for
this
this
med
value
is
that
some
data
links
are
preferred
over
others.
I
For
example,
the
airplane
should
use
SATCOM
as
a
first
alternative
and
failover
to
other
data
links
when
SATCOM
is
not
available,
so
the
system
ensured
that
there
always
be
a
working
route,
even
if
it's
only
a
default
route,
I've
got
a
working
network
model
of
this
I
could
demonstrate
if
anybody
was
interested
I'm
not
going
to
show
at
the
demonstration
right
now,
but
we
can
do
that
afterwards.
If
anybody
wants
to.
I
So
how
does
this
route
optimization
work?
You
have
a
source
autonomous
system,
border
router.
You've
got
a
core
autonomous
system
border
router
that
we're
calling
the
router
in
this
diagram
and
a
target
sub
Sabbath
on
misses
from
order
router.
So
the
source
is
initially
going
to
be
sending
packets
via
the
core
router
to
the
target
router.
I
While
the
data
packets
are
flowing,
the
source
will
get
notified
that
there's
a
better
first
hop
for
it
to
use
the
source
then
does
what's
known
as
a
route,
optimization
exchange
with
the
target
and
then
future
data
packets
can
be
direct
go
directly
from
the
source
to
the
target.
So
this
is
a
little
bit
different
than
what's
happening
in
Lisp.
I
By
my
read
of
Lisp
in
that,
if
you
think
of
the
source
as
being
the
ITR
and
the
target
as
being
the
EGR
normally,
what
happens
is
that
the
the
ITR
holds
the
data
packet
yeah
and
then
does
a
mapping
table
lookup
to
discover
the
arlok
of
the
target
and
then
start
sending
the
data
packets
to
the
target.
So
the
difference
here
is
that
data
packets
continue
to
flow,
even
if
they're
going
through
a
less
optimal
route
until
the
route
optimization
can
happen,
and
this
is
where
the
the
document
is
located.
I
I
Well,
you
could
use
the
Lisp
encapsulation,
it
could
use
any
African
capsulation
format.
Okay,
so
where
this
came
from
is
a
proposal
that
I've
been
working
on
for
a
one
time
called
arrow.
An
arrow
uses,
what's
known
as
generic
UDP,
encapsulation
or
gue,
but
arrow
could
just
as
easily
use
the
Lisp
UDP
encapsulation,
the
encapsulation
formats
kind
of
you
know
agnostic,
so
I.
K
Have
a
question
I
saw
that
you
had
you
said
that
you
don't
really
do
you
care
about?
How
do
you
optimize?
If
you
don't
you
don't
care
about?
You
look
realization.
I
know
you
mentioned
that
the
packets
were
very
important
in
addition
ever
be
dropped,
but
by
actually
sending
it
on
a
not
optimized
route,
you're,
not
sure
either,
but
it's
going
to
to
make
it
back
right
because
you
might
have
circular
routing
and
then
the
loop
or
something
you
don't
really
know
either
right.
Well,.
I
I
I
Well,
the
thing
about
Lisp
is
that
the
packets
but
okay,
the
thing
about
what
I
understand
this
being
that
the
the
to
see
in
the
Lisp
is
that
the
ITR
holds
on
to
the
packet
until
I
get
something
up,
reply
back
right
and
then
it
sends
the
packet
while
a
couple
things
can
happen.
First
of
all,
when
it
gets
the
map
reply
back,
it
can't
know
without
testing
at
first
that
the
path
from
the
ITR
to
the
ETRS
working
path.
I
F
I
K
The
reason
why
we
have
seen
this
is
that
you
actually
are
relying
on
the
bishop
II
cheese-eating
mechanism
to
ensure
that
you
packet
make
sense
not
really
but
the
moment
you're
sending
the
package
down
from
the
plane.
You
have
no
guarantee
that
this.
This
is
anything
about.
You
don't
know
that
right
well,.
I
It's
it.
The
stability
of
BGP
is
going
to
be
based
on
the
fact
that
you're,
your
autonomous
system,
border
routers,
aren't
moving,
and
when
you
have
an
e
ID
prefix
comes
on
to
a
xtr
for
the
first
time
that
xtr
has
to
inject
the
prefix
into
the
BGP
routing
system.
So
until
that
b-12,
the
Eid
prefix
then
gets
propagated
to
the
core
autonomous
system
border
routers,
but
doesn't
get
propagated
out
to
the
stub
autonomous
system
border
routers.
I
K
Come
o
SPF
whenever,
whenever
I
did
the
first
time,
I
thought
this
looks
exactly
like
always
here
with
the
backbone
and
the
stuffed
routes
have
stubby
areas
that
look
exactly
the
same.
Whatever
is
trying
to
say
that
you're,
relying
on
some
other
mechanism
down
which
doesn't
have
any
connection
with
you.
So
yes,
these
connections
may
go
down
as
well
right,
no
guarantee
you're
right.
D
I
So,
with
the
arrow
proposal,
arrow
is
based
on
a
nd
ma
link
model,
which
means
you
can
use
ipv6
neighbor
discovery,
so
the
the
neighbor
discovery
messages
that
happen
are.
First,
you
get
our
redirect
and
after
the
redirect
message,
the
source
sends
a
neighbor
solicitation
then
gets
a
neighbor
advertisement
message
back
so
that
neighbor
solicitation
advertisement
tests
the
forward
path
so
that,
before
you
start
sending
data
packets
directly
from
source
to
target,
you
know
that
the
forward
path
is
working.
D
I
G
I
So
ICAO
has
this
working
group
I,
which
is
developing
a
document
called,
do
no,
it's
not
the
I'm.
Sorry,
it's
98
96
ICAO
document,
98
96
that
describes
the
ATN
IPS.
The
ipv6
application
for
the
aeronautical
telecommunications
network.
Icao
has
formed
a
sub
group
called
the
eye
mobility
sub
group
of
working
group
I,
and
we
meet
on
a
three
times
annual
basis,
but
they're
also
starting
to
hold
in
our
meetings
to
that
I
participated
in.
I
They
want
to
work
towards
a
solution
adoption
within
this
year
if
they
can,
but
so
so
we're
continuing
to
work
with
some
of
the
folks
from
the
FAA
and
also
euro
control,
which
is
the
European
equivalent
of
FAA
Airbus,
is
in
involved
in
marsan,
which
does
the
satellite
systems
is
involved.
So
you
get
a
lot
of
people
from
the
aviation
community
involved
in
this.
I
G
I
I
G
I
M
Prevents
us
because
going
some
activity
started
with
this,
the
only
showstopper
was
that
someone
has
to
provide
is
certified
and
aircraft
certified
broader
to
the
aircraft,
so
I
think
you're
in
the
business.
It's
like.
Isn't
it
just
easier
to
this
been
able
rod
sitting
in
the
aircraft?
Then
you
can
slick
of
the
different
links,
use
Wi-Fi
if
you're
and
it's
use
the
different
HF,
whatever
technologies.
If
you
have
the
reroute
on
the
aircraft,
yeah.
I
That's
what
we
are
called
client,
less
and
client
less
would
be
able
to
see
all
the
different
aviation
data
links.
So
you
might
have
a
cellular
system
you
might
have.
A
satellite
system
might
have
a
VHF
system.
There
are
reasons
why
the
community
is
going
away
from
that
they
want
to
have
as
simple
a
solution
on
the
airplane
as
possible
and
one
that
doesn't
require
software
upgrades,
and
so
they
want
this
thing
to
look
like
just
an
ipv6
end
system.
I
J
I
The
the
way
mobility
would
work
is
that
if
you
have
mobility
within
a
subnetwork,
like
I,
go
from
a
first
cell
tower
to
a
second
sub
town
cell
tower.
It's
just
like
the
way
our
AT&T
Wireless
works.
The
the
the
service
provider
takes
care
of
mobility
within
that
sub
network.
So
it's
it's
like
layer,
2
mobility
and
that's
the
endpoint,
so
the
layer
to
mobility
events
don't
get
propagated
into
the
BGP.
I
J
J
I
Andy
VPN,
they
saw
that
so
so.
We've
taken
an
example
of
this
airplane
right
here
has
to
be
GP
routers,
its
associated
with
so
there's
two
BGP
routes
that
are
in
the
routing
system
and
the
med
value
determines
which
one
gets
used
and
if
the
airplane
goes
away
from
one
of
those
ITRs
it
the
ITR
draws
the
throughout
and
bgp
will
we'll
take
take
that
route
out
of
out
of
the
rib.
I
I
Well,
so
that's
a
good
question
for
the
ground-based
lift
guys,
because
it's
the
same
thing
with
its
it
is.
It
is
because,
with
the
ground-based
list,
what
you'd
be
doing
is
updating
the
mapping
database
and
I
know
the
ground-based
Lisp
guys
have
thought
about
this.
There's
a
company
in
Austria
called
free
cuentas.
You
might
have
heard
her
for
Qantas
and
that's
where
a
lot
of
these
ideas
have
come
from
the
ground
base
of
the
grace
list
of
ideas
will
come
from
so.
J
My
common
sense
being
it
would
be
interesting
to
see,
there's
an
opportunity
to
standardize
that
handle
so
that,
where
we
do
it
maybe
get
deep
or
lisp.
We
have
a
standard
way
of
actually
seeing
that
one
way
of
seeing
that
a
plane
is
there
or
not,
yeah,
because
otherwise
we're
gonna
have
to
create
connectors
for
each
like
a
sub,
Network
and
I.
Don't
know
it's.
I
J
I
H
A
J
D
I
Yeah,
so
that
so
just
to
get
a
little
bit
more
background
on
that.
So
what
we're
saying
is
that
the
airplanes
are
gonna,
get
night
BB,
six
prefix
that
comes
from
the
concatenation
of
a
24-bit
ICAO
prefix
and
the
24-bit
idea
of
the
airplane.
So
each
airplane
is
going
to
get
a
slash,
56
prefix,
but
you
may
never
get
any
aggregation
among
the
airplanes
that
are
in
a
sub
network.
They
may
all
have
their
own
prefixes
that
don't
aggregate
it
all
in
Ada
wines,
so
each
airplanes
prefix
is
going
to
be
injected
into
the
ibgp.
I
Now
I've
done
some
emulation
experiments
that
show
that
a
single
bgp
router
can
handle
a
million
routes
and
there's
not
going
to
be
a
million
airplanes
in
the
sky
at
any
time
in
the
near
future.
This
tens
of
thousands
of
airplanes
in
operation
any
time
so
so
million
is
already
a
pretty
good,
reasonable
scaling.
You
know
upper
level
so.
K
I
The
one
thing
I'll
say
about
that
is
that
the
number
of
these
core
ASPRS
here
is
on
the
order
of
tens.
It's
not
on
the
order
of
thousands
or
hundreds
of
thousands.
So
there's
not
there's
not
a
lot
of
convergence
time
to
get
those
things
to
converge
and
the
the
this
table
autonomous
system
border
routers.
These
guys
here
they're
there
can
be
thousands
and
thousands
of
those,
but
they
don't
keep
a
full
routing
table.
They
only
keep
default
and
the
only
maintain
to
default
that
points
to
one
of
their
core
autonomous
system.
I
K
A
I
No,
it's
it's!
It's
the
latter
but
son,
but
be
careful
that
this
is
not
necessarily
the
the
ipv6
addresses
that
will
be
given
to
the
passengers.
This
is
for
air
traffic
control,
so
this
is
for
the
pilots
to
talk
to
you
so
so
I,
don't!
This
is
answering
your
question,
but
the
airplane
is
going
to
have
potentially
an
Internet
of
Things
address.
N
I
Each
continent,
the
pink
cloud,
is
a
very
good
question:
it's
what
we're
calling
the
Internet
work
and
whether
the
internet
work
would
be
worldwide
or
somehow
subdivided
into
regions
whether
it
would
be
partition
so
that
you
can
only
reach
some
are
locks
from
it.
But
that's
that's
all
up
in
the
air
right
now.
I
think
what
the
freak
guys
have
been
thinking.
I'm
gonna,
go
check
this
with
them
when
I
go
to
the
next
ICAO
meeting.
Is
that
that
pink
cloud
would
be
one
worldwide
completely
open?
I
I
J
B
Is
also
some
work
in
the
university
aromatically
in
France
about
how
to
make
different
mapping
system
to
talk
to
each
other,
not
specifically
PPR,
generally,
in
a
generic
way,
how
to
share
information
so
that
we
can
make
different
mapping
system
interactions,
which
comes
to
the
fact
that
you
sweep
these
pink
cloud
in
several
directions
and
several
nothing.
So.
M
I'd,
like
that,
thank
you
so,
the
last
time
I
did
from
the
same
place,
they're
running
into
trouble
replacing
or
there
is
10
lines
and
the
problem
for
them
was
that
left
Iran
who
elected
Allah
in
the
financial
industry,
support
life
life
transmission.
So
if
my
ass
is
like
why
this
is
what
I'm
really
cast
about
through
large
life
feats
like
we're
doing
in
financial
industry,
because
their
biggest
concern
was
about
that
they
have
to
deliver
life
life
for
their
vacations,
which
means
you
always
have
to
have
two
streams.
M
M
So
this
was
my
proposal
because
they
asked
for
life
live
feeds
over
the
phone
past,
which
is
currently
done,
at
least
in
part
of
different
I
standings,
but
Iestyn
running
away.
So
there
probably
was
so
what
should
we
do?
Next?
They
say
hey.
Why
not
do
with
the
cost,
which
is
proper
weight,
but
they
told
me
this
will
take
10
to
10
years
to
change
the
implications.
They
are
people
right,
but
here
this
looks
like
a
unicast
Network,
but
if
their
requirement
is
will
be
cast
it
life
life.