►
From YouTube: IETF105-LISP-20190722-1330
Description
LISP meeting session at IETF105
2019/07/22 1330
https://datatracker.ietf.org/meeting/105/proceedings/
D
E
E
So
the
usual
point
else,
I'm
Luigi
here
beside
me:
Joel
we
are
the
cultures,
but
man
didn't
make
it
this
time
she
would
join
in
remote.
She
will
anyway
guarantee
minutes
as
usual.
Thank
you
very
much
to
Patmos
and
all
the
material
is
online
and
available
okay.
So
this
light
showed
the
situation
of
the
documents
three
months
back
in
pack
and
the
progress
is
this
one?
You
catch
the
differences,
so
the
the
list
RFC
8113.
This
is
now
in
the
RFC
queue
just
waiting
for
a
missing
reference.
E
The
young
model
is
under
review
a
young
doctor,
so
this
is
Chris.
Hobbs
and
I
met
him
this
morning
and
there
is
a
mismatch
in
the
document
between
the
model
and
the
tree.
I
bet
you
are
at
one
of
the
colliders
Fabio.
Oh
you're
right
I
go
higher
in
the
air
at
key
Fabio.
No,
no!
No!
The
point
being
is
that
today
or
tomorrow,
Chris
will
contact
you
and
show
you
the
issue
so
that
this
can
be
solved.
I
mean.
E
Yeah
so
other
than
that
doesn't
mean
we
did
nothing,
but
actually
Albert
will
present
what
we
did
in
on
the
main
documents.
This
will
be
the
first
presentation
that
we
have
today
then,
on
our
update
on
the
Lisp
security
document,
and
then
we
have
a
few
presentation
that
are
not
working
group
items
so
actually
Prakash
didn't
make
it
so
we
will
not
present
lisp
site
external
connectivity,
so
we
have
more
time
for
this
discussion
at
the
end
of
all
I'm
sorry,
so
it
didn't
work.
Basically
he
make
it,
but.
E
So
we
will
have
these
overlays
Dino
will
discuss
about
I'll,
show
us
the
list,
tracer
route
tool
and
make
a
demo
on
the
mobile
node.
And
finally,
we
will
have
a
share
on
that
will
a
present
again
with
more
time
for
question,
especially
the
distributed
chef,
special
Lisp
blackboard
for
automotive,
okay.
E
F
Hello
now
it
works
so
I
think
that
we
works
in
the
last
IDF,
so
I'm
going
to
summarize
either
with
the
changes
on
the
6830
bees
and
6:33
bees
since
the
last
IDF.
So
basically,
we
posted
a
new
version
of
each
of
those
documents,
both
in
June
2019
so
1
months
ago,
and
then
hopefully
we
have
addressed
all
the
comments
made
by
the
reviewers.
It
is
up
to
them
to
the
site,
of
course,
so
we
posted
the
version
and
then
we
send
an
email
to
the
reviewers
telling
them.
Ok.
F
Here
you
have
the
new
documents
and
now
I'm
going
to
outline
a
little
bit
the
changes.
There
are
many
many
changes
you
can
check
the
diffs
and
I'm
summarizing
them
into
these
four
items,
which
is
security
rate,
limiting
MTU
and
other
and
I'm
not
going
to
divide
the
the
changes
into
that
open
or
control
plane,
but
I'm
going
just
to
mix
them
because
I
think
it's
a
little
bit
easier
to
follow.
So
the
first
change
is
that
now
gleaning
map,
versioning
and
locator
status,
bits
and
acronyms
should
not
be
used
over
the
public
internet.
F
At
the
recommendation
on
the
that
we
have
written
on
the
spec,
this
is
because
on
the
public
internet
there
are
some
issues
regarding
those
mechanism
and
then
also
locator
status
bit
should
be
used
with
my
burning,
when
my
burgeoning
is
injures.
So
that's
the
first
one.
The
second
one
is
regarding
how
we
compute
the
keys
on
my
projector,
how
we
authenticate
my
precious
term
messages.
F
Ok,
now
what
we
do
is
we
use
the
pressure
key
not
to
compute
the
H
Mac
that
authenticates
the
map
register,
but
rather
to
the
I've
another
key,
which
is
the
one
that
you
use
to
tote
integrity
map
register.
So
in
the
in
the
map
register,
you
have
this
algorithm
ID
and
this
agree
maybe
typically
was
pointing
to
the
right
algorithm.
I
did
2h
Mack
the
map
resistor,
but
now
it's
pointing
to
a
new
table
which
is
telling
you
which
Mack
algorithm
you
have
to
use
and
which
key
derivation
function
you
have
to
use.
F
F
They
are
amazing,
so
you
have
to
pick
this
field
will
tell
you
which
items
to
use,
and
then
you
have
to
to
first
of
all
take
your
pressure
key,
which
is
step
number
3,
and
then,
once
you
have
all
these
three
things,
what
you
need
to
do
is
you
need
to
compute.
You
need
to
the
develop
a
new
key
and
the
way
to
derive
the
new
key.
Is
you
apply?
F
F
F
Sorry,
this
title
is
wrong:
it
should
be
read
limiter.
So,
in
terms
of
rate
limiters,
there
is
an
RFC
which
is
80
85,
which
defines
some
guidelines
on
rate
limiting
musical
applications
over
the
internet.
So
we
are
following
those
guidelines
and
map
request
must
be
read
limited
to
one
per
second
per
ad
prefix
and
then
in
order
to
be
in
line
with
the
spec.
What
we
do
is
that
after
ten
retransmits,
you
have
to
wait
for
30
seconds.
F
So
overall,
you
are
sending
one
packet
on
average
each
three
seconds,
the
map
reply,
mass
will
read
limited
one
packet
per
three
per
few
seconds.
This
is
per
destination
or
lock.
This
also
applies
to
SMR
to
the
message
that
SMR,
centers
and
SMS
responders
sent,
and
then
regarding
the
map
register,
you
have
an
exponential
back-off
of
one
minute,
depending
on
the.
If
you
need
to
retransmit
it
then
on
regarding
MTU
and
following
the
guidelines
of
the
same
RFC.
F
So
if
you
Lisp
is
expected
to
be
deployed
over
cooperating
entities
and
if
the
deployers
are
aware
of
them
to
you,
then
they
can
set
them
to
you,
but
if
they
are
not
aware
of
the
MTU,
these
are
a
few
states
that
the
packet
size
must
be
limited
to
those
two
sizes
for
ipv4
and
ipv6.
So
we
have
specified
this
on
the
documents
and
then
other
small
changes
are
that
this
is
rather
a
clarification
instance.
Iv
is
a
24
bit
field
on
the
data
plane
and
that's
it.
This
is
what
it
is.
F
G
I
I
If
you
can
this
week
tried
to
grab
them
and
asked
if
they
had
a
chance
to
look
at
it.
Okay-
and
you
know,
and
I'll,
try
to
remind
them-
and
definitely
next
week,
we'll
get
on
or
more
because
everybody's
very
busy
right
before
the
meetings,
but
we
try
to
get
them
to
clear
them
now,
but
if
you
can
get
them
this
week,
you
know
that
would
be
good
in
case
they
don't
have
a
follow-up
right
to
talk
about
it.
G
G
So
these
are
the
updates
that
were
posted
on
June,
the
2nd,
so
the
changes,
the
main
changes
were
to
align
list
sack
with
the
changes
that
were
introduced
it
in
68
2
to
3
bees
and
to
reflect
the
same
mechanism
that
we
are
now
used,
plus
the
first
one
here
that
is
basically
allowing
an
ITR
to
securely
downgrade
to
non
this
sack.
This
was
a
request
that
came
from
ban.
We
discussed
with
him
in
Prague
and
now
the
document
reflects
that
update.
G
The
second
is,
you
will
see
very
similar
to
what
has
been
done
in
68
33,
to
secure
the
map
register
and
in
the
third
one.
There
were
a
bunch
of
comments
that
were
posted
by
made
earlier
this
year,
very
detailed
and
meant
bukata.
He
does
always
a
great
job
of
sending
comments,
so
I
want
to
thank
him
for
that
and
those
are
being
addressed
and
I
think
he
has
a
knowledge
that
they've
been
covered
so
just
to
go
a
little
bit
into
the
details
of
the
changes
that
were
done.
G
The
main
changes
are
in
the
authentication,
ECM
authentication
data,
and
we
will
see
that
the
first
change
we
listed
you
know
regard
the
introduction
of
these
EBIT
and
then
the
other
changes
required.
We
regard
the
change
of
the
key
ID
and
yo
TK
up
ID.
As
you
see,
this
is
very
similar
to
what
Albert
just
described
and
we
will
look
into
it
so
for
the
secure
downgrade.
G
What
we
did
was
basically
specify
how
an
ITR
by
using
these
et
al
can
sign
bit
can
specify
that
is
willing
to
accept
known
sign
map-reply
from
from
the
map
server.
It
has
to
be
done
carefully
because
you
know
attacker
may
take
advantage
if
it's
not
done
properly
and
I.
Think
we
go
to
the
point
where
you
know
this
is
well
specified
now
and
clear
and
it
is
secure
and
then
the
other
change
is
equivalent
to
what
Alberto
just
described
rather
than
using
the
pre-shaped
key
to
protect
the
transport
of
the
one-time
key.
G
G
You
will
see
that,
basically,
when
we
identify
the
creep
sivapal
going,
we
don't
identify
only
the
wrapping
algorithm
that
we
used
to
wrap
the
actual
key,
but
also
the
key
derivation
function,
and
this
is
very
similar
to
what
Albert
just
described
for
my
producer.
Even
if
the
security
function
platform
is
different.
In
that
case,
it
was
generating
a
key
to
authenticate
the
map
register
and
this
key
is
used
to
transport,
the
one-time
key
again.
This
was
discussed
with
Ben
back
in
Prague.
G
There
were
also
a
comments
from
mad
right
after
I
posted
the
new
version
version,
19
and
I
have
incorporated
those
already
I,
even
published
version
19,
because
I
was
waiting
for
more
comments
to
come.
Maybe
it
makes
sense
to
publish
it
at
this
point
since
they
haven't
arrived,
but
thanks
again
to
make
book
a
therefore
always
the
detail
review
that
he
posts
and
yeah
and
basically
next
step
same
consideration
of
what
we
did
before.
G
To
be
honest,
I
am
not
completely
sure
on
the
status
of
the
leasetech
Draft
at
this
point,
because
we
basically
were
told
by
the
reviewer
that
these
the
review
of
the
B's
RFC's
should
should
go.
You
know,
together
with
a
review,
Allah
SEC,
and
we
have
got
a
bunch
of
comment.
We
have
incorporated
that
we
received
in
Prague
and
we
I
think
that
the
next
step
is
getting
a
feedback
back
from
from
men
and
from
the
security
Directorate,
and
that's
my
understanding
but
yeah.
We
need
to
figure
out
how
what
is
actually
the
next
step.
G
G
E
I
I
I
G
H
Okay,
so
now
let's
talk
a
bit
about
overlay,
so
so
these
are
love
that
the
bitter
submitted
couple
five
years
back,
I'm
gonna
give
an
update.
We
haven't
given
the
date
since
the
last
night
Victor
presented
so
I
think
this
was
in
Bangkok
like
quite
years
ago
that
he
but
the
draft.
Here
we
got
some
comments
from
both
it
says
and,
and
a
working
group
in
general
about
what
they
were
sent
to
to
solve.
So
we
have
revisited
RAF.
We
have
added,
sometimes
that
clarify.
H
What's
the
focus
on
the
under
thousand
in
particular,
we
we
describe
how
this
helps
interoperability
between
different
list
overlays.
We
will
have
done
minor
fixes
through
the
text
as
well
and
the
version
zero.
One
is
first
that
it
was
actually
posted
last
IDF,
but
we
would
enter
sent
it
back
then,
so
we
are
presenting
it
now.
So
what
is
this
overlay
seen?
H
So
the
basic
idea
here
is
that-
and
this
is
coming
past
different
from
operational
and
deployment
experience
that
that
we
are
gaining
on
only
is
right
that
in
some
cases
you
may
want
to
have
different
Lisp
overlays
that
use
either
different
control
planes
or
different
encapsulations
for
a
variety
of
reasons,
and
you
want
to
make
them
interoperate.
So
the
way
to
do
that
is,
but
via
what
we
call
an
overlay,
the
desired
name
Delphi
came
up
with
and
we
find
it
really
funny.
H
So
we
connect
the
different
sides
over
lesbian
on
overlay
and
and
then
you
have
border
XTS
that
connect
the
decide
overlays
with
the
good
the
early.
Why
is
that?
As
I
was
saying?
Interoperability
is
one
of
the
biggest
reason
for
this
control
plane.
So
you
can
have
different
mapping
systems
running
in
each
of
the
overlays,
and
then
they
can
exchange
my
pins
through
there
through
the
overlay,
nor
so
data
plane
that
I
think
we
didn't.
We
were
not
clear.
H
The
first
thing
we
represented
this
on
how
important
it
is
from
the
probability
to
have
the
flexibility
to
have
differ
and
the
doubling
of
relations
different
addresses
from
the
underlay
in
order
to
to
to
to
the
pre-release
in
a
physical
manner.
So
the
destructor
of
this
is
that,
as
our
same
multiple
said,
overlay
is
connected
through
a
common
transitive
early.
H
This
allows
different
local
spaces
so
think
of,
for
instance,
the
the
site
overlay
on
on
the
Left
may
have
private
address
in
on
the
underlay,
and
the
scope
of
that
popularising
can
be
constrained
within
that
site
overlay.
Also,
when,
when
you
have
an
overlay
in
the
middle,
you
will
need
to
to
be
aware
of
their
routes
to
to
to
reach
the
remote
overlay.
H
So
so
the
the
amount
of
entries
on
the
underlay
is
greatly
reduced,
which
is
a
good
thing,
so
you
have
electrodes
on
the
underlay
tool
to
deal
with
and
you
can
have
different
others,
families,
of
course,
and
then
fate
is
relation
answerability,
so
parts
of
the
network
may
may
die
and
the
rest
of
race
as
if
nothing
happens.
So
let.
E
H
G
E
H
Interconnect
between
them,
that
is
a
that,
is
an
excellent
question,
actually
in
the
in
the
list
of
next
steps.
So
in
the
drive
right
now,
we
consider
a
single
overlay
yeah.
We
have.
We
have
discuss
on
what
to
do
with
multiple
overlays
that
comes
with
a
different
set
of
requirements
and
assumptions
that
we
need
to
to
evaluate
I
mean
theoretically,
yes,
we
have
not
addressed
yet
that
yet
on
the
on
the
graph.
So
that's
one
day
on
next
step
for
us,
but
that's
a
good
question.
You.
E
E
The
simple
solution,
but
my
point
is
not
that
you
need
several
overlays.
What
can
happen
is
it
for
whatever
reason
you
have
different
overlay
sites
and
they
decide
to
interconnect
and
deployed
at
uber
laser,
so
me
and
you
and
Alberto
and
Fabio,
take
we
interconnect
pairwise
and
we
use
different
to
balance
now
in
the
future.
We
want
all
to
interconnect.
We
become
big
family.
Okay.
Now
what
you
do
you
deploy
a
new,
unique
overlay
or
you
interconnect.
J
H
I
think
I
think
that
really
is
question.
Isn't
correct
me
wrong
right.
So
you
have
two
completely
separate
like
the
picture
you
have
here,
but
two
different
ones.
So
you
have
two
overlays
connected
here
to
here
right
and
then
at
some
point
you
want
to
bring
the
four
different
site
over
list
together
and
I.
Think
that
the
answer
is
with
the
deficit
is
now.
What
you
will
have
is
the
two
overlays
that
used
to
be
completely
independent,
each
other.
J
J
You
merge
the
two
Lay's,
but
that's
what
I
wasn't
suggesting
I
was
suggesting
it's
just
one
overlay
yeah,
because
because
the
e
IDs
that
are
connecting
this
relate
to
their
own
local
mapping
system
could
register
to
a
mapping
system.
That's
global
and
it
just
looks
like
a
regular
list
overlay
as
we
know
it
today,.
K
H
E
E
G
In
that
case,
you
know
what
I
think
will
determine
what
you
do
is
how
mystic
domains
are
related
right
in
that
particular
case,
for
example,
the
requirement
is
that
there
are
some
enlisted
domain,
that
they
will
not
become
the
same,
because
maybe
there
before
provider.
So
we
distract
the
first
level
of
trying
to
put
them
together
with
a
new
ballet
is
exactly
trying
to
get
everybody
talked
to
and
depending
on
how
your
mr.
domains
have
you
can
go
to
a
single
overlay.
You
can
go
to
you
know
a
single
who.
E
H
H
Well,
you
will
see
this
yeah,
so
the
idea
is
that
each
site
overlay
has
its
own
mapping
system,
and
then
you
have
the
borders
that
subscribe
to
everything
into
that
mapping
in
the
mapping
system
in
the
local
mapping
system,
so
take,
for
example,
the
borders
from
day
on
the
top
of
the
picture.
Those
guys
are
gonna
subscribe
to
0/0
on
the
on
the
mapping
system
on
the
on
the
top,
so
they're
gonna
get
all
the
mappings,
but
they
are
gonna.
H
Do
then
is
they
are
gonna
potentially
aggregate
those
mappings
and
register
the
map
ends
into
the
into
the
and
the
overlay?
So
then,
with
that,
what
happens?
Is
that,
within
the
same
site
overlay,
the
mapping
resolution
happens
a
syllable
in
in
any
list
pay
overlay
for
communication
across
different
site
overlaid?
H
What
happens
is
that
the
the
the
exertion
aside
try
to
resolve
the
the
mapping
is
not
local
is
not
registered
into
the
into
the
local
site,
so
they
go
to
the
border
and
the
border
he's
able
to
query
the
new
map,
insist
on
the
overlay
and
find
which
site
overlay
has
that
map
in
and
and
send
the
traffic
there
yeah.
This
is
the
flow
with
the
deplane,
as
I
was
saying
right,
so
so
any
any
hidden
day
on
the
Cedars
last
year.
H
So
next
steps
and
and
I
think
we
already
have
this
discussion
right
now.
First,
we
want
to
have
some
discussion,
this
Monday
on
the
mailing
me.
So
after
the
meeting
Wilson
I
made
the
middle
is
asking
for
discussion,
then
we
would
like
to
check
with
with
you
guys
and
with
the
chess
team.
Do
you
think
that
this
is
a
topic
that
the
working
groups
will
take?
So
we
are
not
saying
that
the
the
draft
is
final
on
his
car
and
foreigners.
H
As
we
point
out,
we
we
need,
maybe
to
others
day
the
challenges
have
multiple
overlays
will.
We
would
like
to
know
if
the
working
groups
will
take
upon
this
draft
and
make
it
a
work.
Group
item
and
yeah
with
the
the
questions
that
you
have
get
in
there
at
the
bottom
is
precisely
especially
we
were
discussing.
H
So
how
do
you
play
with
the
with
the
overlays
in
terms
of
which
mapping
system
you
want
to
use
that,
and
you
want
to
have
multiple
overlays,
which
are
different
ways
you
can
use
to
reduce
the
state
on
the
overlay.
So
these
are
the
questions
that
we
have
from
the
table
or
another.
We
will
like
to
discuss
and
address
in
in
further
edition
of
the
draft.
So
that's
that's
why
we
we
want
to
to
get
some
feedback
from
the
working
group
on
this.
H
For
instance,
think
that
you
have
multiple
ways
to
register
the.
Let
me
go
one
step
back:
these
guys
kept
the
borders,
they
need
to
register
the
mappings
from
the
local
site
overlays
into
the
overlay,
and
you
can
take
different
approaches
for
that.
So
you
can
register
the
mappings
as
you
get
them
from
the
local
site
or
you
can
try
to
aggregate
as
much
as
possible,
but
then
the
the
bottom
is.
They
said
how
to
aggregate
doors,
and
then,
if
there
are,
if
there
are
gaps,
you
need
to
register
the
gaps
and
so
on.
H
E
E
E
H
E
J
Well,
the
analogy
I
was
going
to
get
us
look.
If
you
have
a
single
bgp
process,
that's
running
with
multiple
VPNs.
You
can
keep
the
VPNs
separate,
but
they're
shared
by
one
process
and
one
routing
protocol.
But
if
you
ran
two
bgp
routing
process
that
are
completely
separate,
it
may
have
a
different
period
topology
altogether.
That
would
be
the
same
as
using
you
realize.
H
Okay,
so
yeah
the
centralized,
so
this
is
partially
coming
from
from
the
discussion,
the
the
victor
of
how
we
are
having
with
the
iCloud
guys
right.
So
in
the
India
case-
and
this
is-
that
is
not
final
yet,
but
the
latest
discussion
is
about
how
they
can
deploy
it's
it's
organization
deploying
its
own
site,
Overland
and
then
the
overlay
is
federated.
So
an
organization
has
control
over
the
over
the
overlay
right.
H
So
they
were
it's
not
clear
yet
how
that's
gonna
end
up
looking
like,
but
they
were
discussing,
for
instance,
okay,
each
of
us
it
takes
I'm
up
server
in
the
in
the
overlay
and
then
we
we
fed
the
rate.
So
there
is
no
single
initiative
domain
for
the
overlay,
so
those
are
the
kind
of
discussions.
The
that
we
would
like
to
do
have
infidelity
seems.
E
H
E
E
H
Yeah,
that's
fine
yeah
yeah!
So
that's
why
the
first
thing
here
is
discussing
on
the
mailing
list
and
I
think
that
you're
saying
that
the
Ennis
we
need
to
agree,
though,
on
the
scope
of
this
drive
like
what
the
graph
is
gonna
address,
but
I
was
gonna
cover
and
then
state
clearly
what
the
RUF
is
not
address
in
your
quadrant
right
yeah.
So
that's
that's.
E
E
J
L
J
What's
secret,
which
we
should
I
look,
so
basically
what
I
wanted
to
try
to
accomplish
his
trace
the
encapsulation
path,
the
round-trip
path,
from
an
ITR
to
an
e
TR
or
from
an
r
TR
to
an
e
TR
or
RT
r,
to
r
TR
or
depending
on
the
number
of
things.
So,
basically,
at
a
client,
that's
running
that's
running
as
an
encapsulator.
You
would
launch
this
list
traceroute
tool
and
what
I
would
show
you
is
the
underlying
hops
between
the
XPC.
J
So
it
would
show
the
path
from
ITR
to
RTR,
DTR
or
through
multiple,
our
TRS,
and,
of
course
it
works
for
an
ipv4
ipv6
overlay
when
an
ipv6
and
ipv4
underlays
is
there
as
well,
so
the
command
line.
It's
a
Python
program,
it's
pretty
simple!
You
can
specify
the
LTR
command
with
an
optional
source
address.
If
you
don't
supply
a
source
address,
it
picks
a
source,
C
I
D
from
a
configuration
file
that
in
the
lispers
done
that
implementation,
the
destination
could
be
an
e
ID
or
a
dns
name.
J
J
This
is
what
the
output
would
look
like.
This
is
basically
showing
a
path
where
the
map
caches
are
populated
in
all
the
Lisp
routers,
so
it
would
show
you
example,
and
then
I
put
the
names
out
here,
so
it's
basically
caps
later
the
capsule
later
the
time
it
happened
in
the
name
of
the
encapsulator,
so
the
rest
of
the
presentation
just
describes
the
fields.
This
is
an
example
where
there's
a
map
cache
miss.
So
let's
say
the
the
ITR
is
wants
to
where
basically
Lisp
trace.
J
J
That
this
has
nothing
to
do
with
the
original
trace
rod,
so
doesn't
use
TTL
mechanisms
at
all.
Yeah
I'll
explain
that
was
like
that
was
Luigi's
question.
He
wanted
to
know
how
not
what
but
and
since
I
since
it's
not
written
up
in
a
draft
there's
their
source
that
you
can
look
at
so
anyways.
Let
me
describe
so
I
guess:
I
should
have
said
the
beginning
of
presentation.
This
is
a
presentation
of
the
functionality
of
LTR,
not
how
it
does
it,
but
I'd
be
willing
to
answer
her
questions.
Of
course.
J
J
J
This
was
the
path
from
the
IPRC
ID
to
the
destination,
and
this
is
where
it
was
turned
around.
So
you
see
the
both
the
asymmetric
paths
now.
This
is
pretty
in
sting
because
you
can
see
if
the
encapsulation
path
takes
the
same
or
an
asymmetric
path.
Remember
if
you're
going
if
you're
returning
the
packet
back
to
1.1.1,
it
may
not
go
through
the
same
RTR,
because
the
pads
are
unidirectional
in
this
case.
The
encapsulation
paths
are
symmetric,
the
underlying
pass-
you
never
know,
but
you
can
find
ok.
J
J
Now,
in
the
notice
that
the
source
address
is
not
the
tenth
address,
it's
the
104
address,
because
the
NAT
translated
it
and
notice
that
the
packet
was
addressed
to
10.24
T,
because
the
RTR
was
it
which
was
in
Google.
Cloud
was
also
behind
a
NAT.
This
is
really
important
information
when
you're
debugging.
So
now
when
the
RTI
wants
to
encapsulate
going
to
the
ETI.
J
Of
course,
you
see
his
global
address
and
now
he's
using
35
to
to
144
92,
which
is
the
translated
address
of
the
ETR,
the
port
and
when
the
ETR
gets
in,
he
gets
on
the
address
or
that's
the
address
of
the
RTO
source
and
that's
his
own
address,
which
is
also
translated.
So
we're
doubling,
adding
like
crazy
through
this
path,
but
we
need
to
find
out
what
these
paths
are.
J
What's
also
reported
from.
The
ITRs
point
of
view
is
what
his
recent
arty
teaser,
so
the
RTR
is
sending
our
local
probes
to
all.
As
our
looks-
and
this
here
is
keeping
the
last
three
round-trip
times
so
you
can
see
that
there
was
some
variability
here-
141
milliseconds
in
55
milliseconds,
since
75
milliseconds
and
in
my
implementation,
I
know
the
number
of
hops.
So
the
number
of
forward
hops
from
the
ITR
to
the
RTR
is
5
and
the
number
of
reverse
hops
coming
back
would
be
6.
J
J
The
way
it
works
in
a
high
level
is
it's
part
of
the
encapsulation
path
and
I
implemented
it
I
added
a
new
message
called
a
trace
message
and
what
it
does
is
that
when
they,
when
a
trace
message
comes,
it's
actually
processed,
and
you
put
JSON
of
information
in
here
and
in
this
case
right
here,
the
JSON
that's
being
put
in
this
trace
packet.
These
aren't
data
packets.
These
are
control
packets,
the
JSON
for
this
entire
round
trip
was
about
800
bytes,
which
is
kind
of
large,
but
death
that
could
be
optimized
as
well.
J
I
did
JSON
because
it
was
easier
to
debug
reading
it
on
a
sniffer
trace.
So
the
idea
is:
there's
a
trace
that
starts
off
and
the
ITR
that
basically,
the
information
that
gets
out
of
its
map
cache
it
puts
it
in
JSON
and
puts
it
in
that
packet
and
that
encapsulates
that
to
the
RTR
the
RTR
gets
this
packet
D
capsulate
s--.
It
sees
it's
a
trace
message
and
takes
the
D.
Capsulation
information
puts
it
in
the
trace
message:
it
appends
it
and
then
now
it's
ready
to
encapsulate
it.
J
It
puts
the
encapsulation
information
in
that
trace
message
and
then
encapsulate
that
and
sends
it
on.
So
that's
basically
how
it
works.
It's
not
it's!
So
it's
exactly
one
message:
that's
going
out!
That's
one
message
that's
being
sent
from
here
that
goes
to
each
of
these
hops
and
being
returned.
So
it's
really
not
a
lot
of
overhead
on
the
network,
so
I
wanted
to
trade
off
the
size
of
the
with
the
number
of
packets
being
sent.
J
F
J
Right
it
would
not
it's
absolutely
true
that
the
the
forwarding
implementation
has
to
change
to
understand
the
trace
and
then,
when
I
sent
the
slides
of
Luigi
I
said
you
know,
I
obviously
haven't
written
this
down.
I
did
the
code
first,
but
we
can
certainly
write
a
draft
if
the
working
groups
interested
in
it.
So
yes,
I,
don't
know.
Do
you
think
it's
useful?
J
It
it
knows
the
number
of
hops
and
the
RPGs
I
have
another
tool
called
locator
LOC
8tr
that
actually
will
go,
go
to
an
ITR
map
cache
and
actually
go
through
the
map,
cache
trace
route
to
each
of
the
arlok's
and
give
you
the
actual
physical
hops.
So
that
can
be
used
in
conjunction
with
this.
The
big
question
is:
it:
should
I
use
a
detailed
version
and
launch
that
tool?
J
J
No,
it's
a
different
program,
I'll!
It's
it's
a
different
tool,
all
together
that
just
queries
the
map
cache
in
the
ITR
and
for
all
the
arlok's
that
are
in
the
map,
cache
that
a
trace
routes
each
one
and
at
all,
and
it
also
pulls
out
this
information
from
the
map
cache.
So
it's
trying
to
marry
the
information.
That's
in
the
map
cache
that
our
log
probing
is
doing
with
the
actual
physical
path
using
tracer,
yeah,
yeah,
yeah
so
merging
it
is.
J
J
J
Using
the
same
outlook,
should
we
switch
over
dynamically
it'd
be
nice
if
we
could
put
work
into
that
as
well.
To
show
that
you
know
these
these
paths,
because
now
that
you
can
do
the
switching
and
get
better
network
performance
on
the
overlay,
because
you
have
the
arlok's,
you
can
switch
back
and
forth
and
you
could
do
it
as
often
as
you
want.
You
don't
affect
the
underlay
at
all.
So
that
would
be
interesting
for
somebody
to
look
at
possibly.
J
E
There
is
a
missing
part
right,
because,
in
order
for
this
to
work,
you
need
support
on
all
the
RTR
sides.
With
your
specific
message.
Yeah
so
I
need
some
some
code
that
is
running
on
the
RTR,
for
example,
during
the
server
part
listening
to
your
specific
man's
messaging
cursing
the
information.
This
is
also
on
the
Nita
repository.
J
J
It
turns
out
that
this
information
would
just
not
be
part
of
the
output,
so
you
could
skip
over
our
TRS
that
didn't
support
it,
but
if
the
art,
if
the
RTR
doesn't
support,
it's
never
going
to
turn
the
packet
around
so
though
it
would
just
be
a
lost
trace.
So
the
if
the
if
the
question
is,
can
there
be
incremental
deployability
of
this?
Yes,
but
not
useful?
You
really.
You
know.
J
Ok,
so
this
is,
this
was
really
fun.
We
did
a
list
mobile
node
demo.
The
guys
at
UBC
did
a
wonderful
job
of
taking
the
arc
code
and
actually
make
a
run
on
iOS
and
since
I
have
an
iPhone
and
I'm
kind
of
on
it.
24/7
I
was
able
to
play
a
lot
of
lists
of
stuff
with
it.
So
you
know,
a
lot
of
credit
goes
to
Albert
Lopez,
who
is
was
very
fixed
fast
in
fixing
bugs
and,
of
course,
Albert
Cabela's.
Who
was
our
Oversight
Committee?
J
J
Mobile
node
and
the
draft
has
been
around
for
quite
a
long
time,
I
think
2009
timeframe
or
or
right
after
the
working
group
a
long
time,
and
so
it
was
an
individual
submission
for
a
long
time,
but
we
made
it
a
working
group
document
and
I
thought
it
was
time
to
actually
do
some
implementation,
so
we
could
maybe
move
this
strap
along
and
see
good
this
ideas.
What's
really
nice
about
lists
mobile
node
is
that
a
phone
can
be
multihomed
and
used
any
service
provider
anywhere.
J
J
So
this
is
what
we're
gonna.
What
we're
demoing
we're
gonna
show
list
mobile
node
on
an
iPhone
and
I
have
some
our
TRS
deployed
in
Google
compute
platform
or
Google
cloud,
and
in
AWS
and
the
list
mobile
node
is
going
to
talk
to
so
CN
in
in
the
mobility
world.
See
n
stands
for
correspondent
node,
which
means
it's
a
node
on
the
internet.
That's
not
moving!
So
we're
gonna
show
that
the
list
mobile
node
is
going
to
talk
to
another
list.
J
Node,
that's
on
the
internet
and
we're
going
to
show
that
a
list
mobile
nodes,
gonna
talk
to
a
non-lisp
mobile
node
so
that
we're
doing
interworking
as
well.
To
show
that
list
can
talk
to
anybody
to
make
my
iphone
useful
and
to
keep
list
running
all
day.
Long
I
want
to
do
I'll
use
all
my
applications
on
it,
so
I
want
to
be
able
to
talk
to
non-lisp
sites
and
we're
using.
So
the
interworking
draft
allows
you
to
do
route
injection
with
a
p
ITR
or
do
something
called
list
nat.
J
J
So
some
magic
sauce
there's
we're
trying
to
do
multiple
efforts
at
one
time.
So
the
list
mobile
node,
is
not
running
a
control
plane.
This
is
really
important.
It's
not
sending
one
Lisp
control
message.
All
it's
doing
is
the
data
plane
and
I'll
explain
what
the
motivation
was
there
lists.
Mobile
node
basically
is
just
configured
0,
which
means
any
IP
packet.
That's
sent
it's
going
to
be
sent
Peter's
proxy
ETRS
or
our
TRS,
because
in
some
cases
they're
read
encapsulating.
J
If
it's
going
lista
less
the
RT
R's
are
configured
to
glean
the
XT,
our
mappings,
so
as
packets
are
coming
in
that
are
encapsulated
by
the
phone.
We
look
at
the
source,
port
and
source
IP
address,
and
we
put
that
in
the
map
cache
so
return.
Packets
can
go
there,
so
that
means
an
iPhone
could
be
behind
a
NAT
and
it
just
works.
There's
no
NAT
traversal
logic
that
occurs.
We
have
a
net
traversal
spec
that
has
a
lot
of
control,
plane
machinery,
but
we're
not
using
that
for
this
demo.
J
Now
this
isn't
an
effort
to
implement
a
lighter-weight
xtr
make
lisp
even
simpler,
less
resource,
constraining
and
one
that
possibly
can
run
on
dash
cams,
perhaps
and
you're
gonna
hear
sharona's
presentations.
So
this
was
an
effort
to
try
to
understand
how
lightweight
we
could
put
lisp
on
small
memory,
challenged
or
wanna
reduce
the
cost
of
of
it.
May
a.
E
E
I
was
sitting
in
Montreal.
It
worked
on
the
cellular
network.
How
do
you
change
it
because,
basically
I
otherwise
you
have,
you
may
have
a
long
stretch
in
the
in
the
passage.
It's
always
the
same.
Peter
I
mean
if
it's
your
own,
then
your
traffic
is
going
back
to
California.
You
know,
even
if
you
are
connecting
to
something
that
is
here.
Yes,.
E
E
J
E
E
F
J
F
Maybe
it
was
it's
a
more
general
comment,
but
we
have
been
working
with
mobile
implementation
for
a
long
time
and
we
had
some
deployments
I'm
speaking
from
a
university,
so
the
Bremen
for
me
is
different
than
for
other
people.
Okay,
so
it's
much
smaller,
but
still
we've
had
some
experience
and
90%
of
the
code
of
mobility
is
not
reverse.
F
J
That's
a
really
good
point,
because
the
list
mobile
node
document
doesn't
say
anything
about
Nats.
It
just
assumes
they're
in
this
pristine
environment,
which
is
not
the
capital
I
internet,
where
the
arlok's
that
the
mobile
nodes
registering
are
all
in
global
space,
so
I
mean
a
lot
of
these
documents,
don't
reflect
reality.
You
know.
J
I
have
a
docker
container
that
runs
on
my
laptop,
the
mobile
phones,
I
D
is
13
and
the
container
is
14
and
what
the
plan
is
is
for
13
and
14
to
talk
to
each
other,
and
that
should
be
able
to
work
as
the
mobile
phone
switches,
but
both
of
them
should
also
be
able
to
talk
to
these
non
Lisp
destinations,
which
is
the
DNS
server
at
Google
and
my
web
site
WWE
at
which
is
in
Wix
comm.
So
it's
just
a
you
know:
it's
just
a
regular
server
host
out
there.
J
These
are
the
RTR
that
are
deploying
once
deployed
in
Google
cloud
and
through
your
deployed
in
AWS,
okay,
so
the
mobile
node
is
running
in
the
container
in
the
RT
hours
of
running
my
lispers
done
that
implementation.
Okay.
So
this
this
is
the
live.
Demo.
I
decided
to
record
it
just
because
I
didn't
know
what
Canada
was
going
to
give
me
in
terms
of
connectivity.
So
what
I'm
going
to
do
is
there's
some
audio
I.
Think
I
can
make
it
work
by
holding
this
here,
but
we'll
see.
C
J
D
J
So
so
what
we're
going
to
show
first
is
14
pinging
at
the
way,
so
up
there
in
that
window
right
there
14
is
pinging
13
okay,
so
that
means
the
client
ping
on
14
is
is
pinging
the
phone
the
phone
is
configured
with
O
R,
there's
the
Eid.
That's
configured
and
I
bring
up
a
ping
program
here,
and
what
it's
going
to
do
is
showing
you
that
it's
connected
to
Wi-Fi
see
up
there
where
my
finger
was
like
yet
I
actually
could
stop.
If
I
need
to,
let's
go
back
a
little
bit.
J
Yeah
I
can
stop
okay,
so
you
see
up
there.
It
shows
that
it's
connected
on
Wi-Fi
at
my
house
and
now
I'm
about
to
hit
that
ping
right
there.
That's
pinging,
WW,
lispers
net,
okay
and
we're
gonna
see
the
pings
flowing
from
the
phone
to
listeners
dotnet,
which
is
a
non-lisp
site
from
13
2ww
lispers
net.
J
Okay,
now
I'm
gonna
about
to
go
back
to
the
phone
and
switch
it
to
LTE
and
that's
what
I'm
about
to
do
now:
I
bring
it
down,
I
turn
off
Wi-Fi
and
then,
if
you
look
at
the
very
top,
let's
see
if
we
can
catch
the
top,
the
pings
are
still
running.
We
lost
exactly
one
packet.
Let
me
go
up
a
little
bit
there
stop.
Can
you
see
it
stop
see
up
here?
It
says
LTE.
So
it's
connected
on
LTE.
The
idea
here
is
we're
doing
sustenance.
J
Survivability
you
connect
the
pings
or
a
TCP
connection
would
stay
up
as
we're
switching
back
and
forth
from
Wi-Fi
to
LTE.
This
here
is
showing
that
packets
are
going
from
13
to
14.
Here's
a
ID
13,
here's
Eid
14,
so
the
ping
that
the
container
was
doing
to
the
phone
was
happening
at
the
same
time,
the
dark
green
counters
mean
that
the
packets
were
were
encapsulated
or
forwarded
within
the
last
second
and
the
light
green
means
they
were
encapsulated
in
the
last
minute
and
the
185
is
WW
lispers
net.
J
So
you
see,
13
packets
are
going
to
13
from
two
different
places
and
it's
sending
packets
to
do
different
places
as
well.
So
that's
what
that's
showing
there
and
I
don't
know
what
I'm
saying
on
the
audio.
Oh
I'm
refreshing
the
screen
now,
so
you
should
have
seen
the
counters
changed
there,
because
I
was
refreshing
now.
J
What
I'm,
showing
is
that
166
is
the
outlook
that
AT&T
on
my
cell
phone
is
giving
me
and
now
I'm,
switching
back
to
Wi-Fi
and
then
I'm
going
to
refresh
the
screen
on
the
RT
R
to
show
that
that
address
is
changing
and
you
see
a
change
to
98.
Not
only
did
it
change
the
are
low,
but
it
also
changed
the
encapsulation
port
because
it's
going
through
two
different
Nats.
If
we
go
back
and
see
the
pings
are
still
running
there
and
I
think
it's
just
stopping.
J
So
there
is
basically
when
this
switch
was
happening
on
the
local
iPhone.
Ios
has
gotten
really
good,
where
it
can
still
keep
the
old
connection
up.
Well,
it's
associating
with
the
other
connection.
So
really
in
most
cases
when
I
do
one-second,
pings
I
lose
zero
packets
when
I
do
about
a
hundred
hundred
millisecond
pings
I'm
getting
two
or
three
packets
that
are
being
lost
just
for
some
switching
that's
going
on,
but
basically
what's
happening
is
as
long
as
packets
are
coming
from
this
guy
who's.
J
Switching
the
RT
R
will
then
automatically
latch
to
a
new
set
of
our
looks.
This
was
this
was
kind
of
the
point
that
Albert
was
making
on
how
this
is
super
simple,
to
not
only
implement
but
easy
to
deploy.
If
we
go
back
and
we
show
what
has
to
be
done
on
the
let's
see,
if
you
just
see
what
has
to
be
I
wanted
to
show
you
the
settings,
let's
see
if
I
can
find
that,
there's
a
settings
stop,
but
basically
all
that
it
has
to
be
configured
is
the
e
ID
and
the
instance
ID?
J
Okay.
So
what
we
try
to
do
is,
let's
see
if
we
could
make
to
scale
this
out
a
little
bit
and
let's
have
the
container
sent
to
wer
balusters
net
and
let's
have
it
be
configured
with
for
Peters
or
for
our
TRS
and
let's
load
split
the
traffic
there
and
that's
what
we
did.
The
traffic
was
loads,
but
I'm
just
going
to
show
you
this
output
that
it
actually
works.
J
Thank
you
Joe,
so,
what's
going
on
here
is
the
packets
that
are
being
sent
to
w
SS
personnel.
This
is
the
iPhone.
This
is
the
containers
map
cache
and,
of
course
he
has
a
0
/
0
through
4
locators,
one
in
GCP
and
3
and
AWS,
and
you
can
see
that
it's
load
splitting
packets
across
all
of
them
now
what's
interesting.
Is
this
implementation
is
sending
our
load
probes
to
the
RTR,
so
we
actually
know
the
performance
of
the
our
TRS.
J
We
can
see
that
the
RTR
it's
700
milliseconds
to
that
one
I
think
that
one
was
Jesus
yeah,
that's
the
Google
one,
and
here
are
the
ones
in
AWS
227,
335
218.
So
these
are
TT
delays
or
these
delays,
or
these
hair
painting
of
packets,
like
Luigi's
Point
brought
up,
is
something
that's
really
important,
because
obviously
the
path,
if
you
turn
oor
off
the
path
directly
W
of
lists
Rashaad
net,
is
on
the
shortest
path
on
the
internet.
J
Okay,
so
we
kept
playing
unless
said
now,
it's
time
to
do
large,
packets,
and
so
what
I
did
for
my
iPhone
was
I
decided
to
drive
from
my
coffee
shop
to
my
house
and
what
I
did
was
I
wanted
to
test.
Listen
to
this
new
video,
so
I
was
streaming
the
audio
there.
You
can
see
it's
going
on
and
then
I
liked
it
so
much
I
decided
to
start
downloading
some
of
the
music.
J
Well
I
did
that
all
over
Lisp
as
well,
so
streaming
audio
and
downloaded
music
while
driving
and
during
that
time,
when
I
was
doing
that
I
switched
between
all
these
different
connections.
I
was
you
know:
Xfinity
just
has
open
Wi-Fi
connections,
so
I
was
connected
to
that
and
then
I
was
switching
back
and
forth.
While
I
was
sitting
at
Pete's
coffee
shop
right
and
everything
was
fine,
and
course
I
was
switching
from
AT&T,
LTE
I,
say
5g
laugh
out
loud
because
the
upper
right
hand
corner
says
5ge,
so
it
must
be
5g.
J
I
didn't
know,
I
can
get
5g
without
paying
any
more
money.
Not
so
that
didn't
that
wasn't
the
case,
but
I
was
able
to
be
able
to
go
to
all
three
of
these.
As
I
was
driving
to
my
house.
They
have
some
weak
loss
or
some
dead
ranges
where
my
connection
downgrades
to
4g,
so
I
saw
it
actually
downgrade
to
4G
as
well.
I
didn't
check
to
see
if
the
outlook
changed,
because
I
was
driving
and
I
couldn't
look
at
the
RTR,
but
I
basically
went
back
and
forth
all
these
four
things.
J
J
This
mobile
node
must
send
before
it
can
receive,
because
the
it's
our
looks
are
being
discovered
from
data
packets.
If
it
never
sends
data
packets,
then
there's
no
way
the
RTR
can
encapsulate
packets
back.
So
if
you
wanted
to
deploy
deploy
a
server
application
that
was
doing
a
socket,
listen
and
expect
people
to
connect
to
it.
First,
it
wouldn't
work.
There's
lots
of
ways
around
this
by
putting
extra
tools
on
the
implementation
to
send
packets
or
a
background
ping,
but
that's
just
a
caveat.
J
So
two
mobile
nodes
can
talk
to
each
other
as
long
as
they
talk
to
another
list
mode
or
not
in
list
mode
first,
and
then
they
get
up.
That
was
your
point
yeah
and,
of
course
this
was
the
Luigi
point.
You
know
Luigi's
always
one
step
ahead
of
me.
Right
latency
exists
to
learn
this
mobile
node
when
it
is
discovered,
but
less
than
doing
a
mapping
system
lookup.
If
we
would've
did
a
mapping
system
lookup
or
registered,
it
would
have
taken
even
longer.
So
you
know
it's
still
not.
Is
it
good
enough?
Is
the
question?
J
That's
subjective,
I
don't
know,
and
then
we
have
this
asymmetry
problem
where
if
this
mobile,
no
one
uses
one
RTR
and
this
mobile
two
uses
a
different
RTR
and
they
haven't
thought
they
can't
talk
to
each
other
right
because
they
don't
have
each
other's
our
look.
So
so
this
mobile
node
would
have
to
talk
to
the
RTR
that
somebody
else
is
using
so
basically
building
a
symmetric
cluster
of
our
TRS.
That
you
talk
to
all
of
them.
All
the
time
makes
you
discoverable
for
other
things
that
want
to
talk
to
you.
J
J
If
you
had
one
and
one
and
two
and
two,
we
just
load
split
across
the
two
and
eventually
when
we
get
our
look
Pro
being
working
and
you
find
out
that
the
ones
the
priority
ones
have
gone
down.
It
just
automatically
switches
to
priority
two.
So
there's
a
lots
of
robust
is
that
you
can
that
the
phone
can
do
by
itself
to
find
better
paths.
It's
really
kind
of
nice
to
do
lists.
J
But
if
we
do
our
look
probe,
you
could
do
the
sort
of
things
I
just
mentioned
and
then
the
question
is
is:
do
we
need
crypto
if
we
can
do
list
crypto
on
the
phone,
everything
that
leaves
the
phone
regardless
of
what
service
provided
you
run
over,
will
be
encrypted
and
it
will
be
encrypted
all
the
way
to
the
RTR
and
can
be
encrypted
from
RTR
to
another
list
mode.
If
it
goes
to
anomalous
note,
it
obviously
has
to
be
unencrypted
and
then
sent
as
it
is
in
the
thing.
J
So
there
could
be
some
security,
that's
added,
so
we
can
maybe
do
some
experiments
with
that
in
the
future
and
do
we
need
multiple
ee,
ID
support
and
instanceid
support.
Should
the
phone
be
part
of
two
different
instance?
Ids
one
may
be
a
work
instance,
a
work
VPN
or
a
private
VPN
that
might
be
interesting
to
use
and
do
they
use
different
addresses.
Are
the
EEI
cds
addressed
it
from
a
global
registry
versus
an
enterprise
IT
department,
or
something
like
that?
And
of
course
you
can.
J
No
job
is
finished
until
you
do
multicast
and
I
have
been
able
to
do
multicast
from
the
phone.
So
if
I
ping
to
twenty
four
one
one
one
from
the
phone
the
packets
go
to
the
Artie
ours,
that's
wonderful!
So
the
single
0/0
map
cache
entry
even
handles
multicast,
so
I
think
it's
going
to
be
easier
to
do
a
phone,
oh,
are
being
a
multicast
source
versus
a
receiver,
a
receiver.
We
would
have
to
make
sure
that
the
app
supported
IGMP
and
that
IGMP
stuff
uses
the
signal
free
stuff.
J
M
M
J
J
Should
we
be
smart
enough
to
know
that
the
Peters
aren't
doing
well
without
our
look
probing,
which
may
be
your
point
and
maybe
turn
it
off
and
then
try
to
go
native,
and
then
you
lose
a
session
survivability
multihoming,
because
the
network
connectivity
isn't
good.
I
found
that
here,
a
little
bit
too
by
roaming
around
was
stuff,
wasn't
working
I'm
going
what's
going
on
and
it's
the
ietf
hotel
Wi-Fi
network,
that's
kind
of
bad.
So
this
is
a
great
environment.
I'm
gonna
be
doing
testing
in
fact,
I'm
going
to
be
testing
all
this
week.
J
If
anybody
sees
me
walking
around
and
they
want
a
quick
demo
right
there,
I
just
have
to
pull
out
my
wallet
and
boom
or
pull
out.
My
phone
I
could
show
you
so
there's
a
lot
of
different
access
points
that
have
these
bad
characteristics.
But
your
point
is
taken.
Should
you
just
our
look
probe
a
few
of
them
versus
all
of
them,
because
there's
all
those
phones
are
going
to
be
our
look
probing
the
same
ones,
but
I
have
a
feeling
we're
going
to
have
location-aware
our
TR
deployment
to
solve
the
Luigi
problem.
J
J
J
E
E
J
H
H
J
J
L
Present
mobility
network
based
on
Lisp,
it's
a
specialized
network,
but
actually
for
me,
the
last
more
than
10
years
specialized
network
is
where
a
list
really
shines
really
nicely
more
than
just
generic
networks.
This
is
one
of
them.
It
combines
Lisp
with
a
grid
of
the
earth
called
h3
which
divides
the
earth
into
hexagons
and
we
are
using
lists
for
mobile
nodes
and
for
addressable
geographical
locations.
So
I
can
talk
to
Gio
States.
There
is
actually
an
additional
standard
which
is
combined
in
here.
It's
called
BDD.
L
It's
consortium,
the
Berkeley
Deep
Drive,
which
is
studying
what
is
worth
noting
about
the
road
about
the
Geo
States,
which
is
worth
communicating.
So
mobility
networks
communicate
information
about
the
road,
so
I'll
repeat
some
of
the
motivation
quickly.
I
did
it
yesterday
in
the
author
of
C,
if
somebody
heard,
and
then
we
quickly
gonna
dive
into
the
to
the
draft
itself,
there's
been
a
lot
of
work
done
in
the
last
three
months.
L
So,
as
you
are
all
aware,
there's
a
lot
of
cameras
in
the
streets
in
the
in
the
roads,
specifically
my
commodities
with
dash
cams.
We
distribute
a
lot
of
them.
There's
hundreds
of
thousands
tens
of
thousand
active
in
each
city,
but
there's
others
that
we've
been
working
with.
There's
cameras
on
junctions,
part
of
the
infrastructure,
Road
side,
there's
other
sensors
and
they're
seeing
the
world,
so
the
drop
of
prices
of
cameras
is
not
only
is
not
only
price
reduction
but
also
being
able
to
understand
what
the
camera
sees.
L
Okay.
So
we
understand
what
we
see.
The
question
is:
how
do
we
communicate-
and
this
is
the
extent
that
there
are
two
networks
that
were
available
to
do
that?
One
is
an
offline
network
which
is
kind
of
like
builds
a
map
which
can
tell
you
the
state
of
a
traffic
sign,
but
not
the
traffic
light
timing
or
if
the
traffic
light
is
out,
it
is
always
a
delay,
a
gap
of
at
least
15
minutes,
if
not
more.
There
is
also
a
real-time
Network,
which
day
is
by
a
standard
called
DSRC.
L
It's
a
layer
1
through
7
peer-to-peer
network
for
cars
to
tell
each
other
what,
if
the
conditions
share
information
about
the
road,
so
when
I
mean
share
is
I'm
not
sharing
with
you
that
you're
gonna
hit
me
in
half
a
second,
that's
something
for
the
sensor
to
pick
up.
But,
for
example,
if
I'm
going
to
turn
a
Junction
into
a
problem
and
be
stressed
out
in
negotiating
my
way
out,
let's
say
a
double
parking
and
many
cars
see
the
problem.
L
They
can
communicate
it
to
me
beforehand
in
real-time
and
that's
where
peer-to-peer
falls
short
and
lease
player
3
can
really
help
us.
Why?
Because
I
may
have
timing
issues
when
who
saw
the
problem?
Maybe
somebody
saw
the
problem
I'm
turning
into
bi
drove
away,
maybe
I'm
rushing
into
a
highway
slowdown,
but
there's
nobody
to
tell
me
that
I'm
rushing
to
a
highway
slow
down.
So
now
information
is
up
in
the
air
and
we've
seen
a
lot
of
pile.
L
Ups
and
I
have
a
lot
of
footage
of
pile
ups
from
next
our
cameras
from
the
last
winter,
and
so
the
idea
instead
is
to
wrap
up
all
these
annotations
and
put
them
in
tires,
which
represents
the
physical
world.
We
tiled
all
the
roads,
based
on
hundreds
of
millions
of
miles
and
based
on
a
maps,
and
now,
when
I
see
something
I'm
going
to
report
to
the
tile
I'm
going
to
talk
to
the
tile,
no
car
will
talk
to
any
other
car.
L
There's
no
privacy
problem,
there's
no
timing
problem,
because
the
tile
is
always
there
and
I'm
going
to
use
lispy
IDs
both
for
the
publisher
and
the
tiles
and
to
subscribe
to
the
information
about
the
roads.
So
all
this
is
going
to
be
combined
use
of
lists
and
based
on
h3
IDs.
So
we
did
a
lot
of
work.
I
mean
this
group
here,
especially
Deena
was
the
whip
to
get
some
detail
into
into
the
draft,
and
I
want
to
go
over
some
of
the
progress
made
with
your
permission,
okay,
all
right.
L
So
what
we
have
is
a
partition
of
the
timing
of
the
roads
using
a
ch3
r15.
These
are
one
square
metal,
meter,
hexagon
tiles
I
think
we
talked
about
before.
Why
hexagons
it's
almost
a
circle,
but
it
tiles
videos
and
elegantly
you
can
propagate
neighbors
if
there
is
a
stopped
vehicle
and
who
is
impacted
and
it
is
very
elegantly
hierarchical.
There
is
a
very
nice
structure.
Every
seven
hexagons
soro
make
up
a
bigger
hexagon,
make
up
a
bigger
hexagon,
so
these
cars,
as
you
can
see,
not
only
they
see
stuff,
but
they
can.
L
There
is
a
technology
to
localize
it,
which
is
where
is
it?
How
far
is
it
from
me
from
the
camera
and
what
is
it
exactly
so
all
that
translates
into
ID,
which
tile
and
an
enum,
which
is
what
do
I
see.
Okay,
I
see
something
in
this
style,
but
I
also
some
to
see
something
in
this
style.
I
have
the
technology
to
do
that.
L
If
we're
going
to
misuse
cases
because
of
latency,
then
we
will
not
be
able
to
put
to
push
that
use
through.
It's
always
gonna,
be
impaired
by
this
bar
and
I'm
gonna
do
describe
what
we
did
using
controlling
latency.
What
kind
of
timing
we
were
able
to
get
to
with
some
AT&T
test
bit
all
right.
So
how
does
it
work
just
like
in
a
list
mobile
node,
the
mobility
clients?
Okay,
they
have
a
lisp
endpoint,
which
is
non
control
plane
when
you
start
a
ride.
You
get
an
eID
this
through
the
generic
network.
L
You
triple
a
authenticate
blah
blah
blah.
You
start.
Alright,
you
get
an
e
ID
for
yourself
and
an
R
rock
for
your
edge
RTR,
okay
and
then
you
begin
to
write.
We
load
balance
the
RT
RS
in
the
city
so
that
all
the
cars
will
be
able
to
engage.
The
RTR
is
in
a
given
metro
area.
The
latency
here
is
fixed,
it's
the
RF
plus
the
mobile
Metro
area
network.
L
Okay.
So
now
we
want
to
exchange
information.
Why?
Because
this
car
sees
this
cars
future
and
vice
versa.
So
we
want
to
share
the
information
about
the
road,
the
state
of
the
road
using
the
list
network,
the
client
once
he
identify
something
exactly
where
it
is
and
what
it
is.
What
is
the
ena?
He
sends
that
information
to
the
tile.
What
does
that
mean?
They
are
15.
1
square
meter
is
algorithmically
translated
to
the
r9,
which
is
a
hundred
thousands
of
those
okay
and
that's
another
hid
algorithmically
by
bit.
Operation
and
the
r9.
L
Okay
is
translate
algorithmically
to
a
list
PID,
okay,
so
now
I'm
selling
I'm
sending
a
packet
from
an
ephemeral
car
Eid
to
a
state
CID.
The
r9
I'm
telling
you
here
is
something
I
saw
here.
Okay,
you
may
not
believe
me,
you
may
require
a
few
of
these,
but
I
can
talk
to
the
tile
instead
of
talking
to
other
cars,
that
packet
reaches
the
edge
RTR,
which
then
looks
up.
Where
is
that
r9?
L
So
we're
going
to
deploy
in
carrier
conditions
and
by
placing
the
r9
in
in
locations
are
locks
which
guarantee
latency
as
an
example,
we
were
able
to
do.
The
following:
a
car
was
rushing
through
a
junction
about
to
breach
a
red
light,
100
kilometers
an
hour
before
it
bridges
the
red
light.
The
camera
understood
is
going
to
breach
a
red
light.
The
junction
camera
is
propagated
information
to
the
tile
from
there
to
the
cars
which
are
involved
in
that
junction.
L
L
Where
is
the
Eric
of
that
state
or
which
other
HRT
are
it's
connected
to?
And
then
it's
going
to
route
the
traffic
just
like
normal,
is
to
the
edge
RT
R
and
from
there
to
the
control
plane
less
encapsulator
in
the
a
ar9
state
container.
Okay.
Is
that
clear,
okay,
I'm,
seeing
something
it's
in
this
bigger
area?
Okay
and
I'm
telling
that
area?
L
That's
what
I
saw
okay,
so
the
goal
here
is
to
have
efficiency,
meaning
is
bigger,
are
as
bigger
aggregation
as
possible
for
efficiency,
so
I
don't
have
too
many
addresses,
so
our
nines
in
a
metro
area,
there's
about
20,000,
okay,
a
but
not
too
big
such
that
when
I'm
now
I'm
going
to
update
other
cars
about
what's
going
on,
I
won't
have
too
much
information
to
tell
them
which
is
not
relevant
to
them.
Okay,.
G
L
L
83-78
a
signal
free,
okay,
it's
a
multicast
which
does
not
propagate
it
in
a
unknown
way
across
the
network,
just
like
in
traditional
IGP,
but
it's
based
also
on
the
mapping
system,
meaning
that
our
nine
address
is
now
a
public
safety
Channel.
Okay,
so
I
know
it
is
a
is
registered
here
as
the
public
safety
channel,
but
also
whoever
drives
into
that
online
knows
from
his
GPS
that
he
needs
to
tune
in
to
this
public
safety
Channel.
Okay,
so
he
sends
an
MSD
to
his
HRT
are
from
is
data
playing
XD
are
to
the
HRT?
L
Are
that
I
am
tuning
into
this
Public
Safety
Channel
mean
somebody
is
breaching
a
red
light.
I
want
to
know
so
that
HRT
are
then
registers
in
the
mapping
system
registers
in
the
mapping
system
too,
that
he
has.
He
has
a
customer
for
that
Public
Safety
channel,
which
is
this
block,
or
the
other
block
okay
and
I
mean
from
now
on,
I
will
know,
there's
a
emergency
vehicle
if
it
was
an
accident
if
they
stole
vehicle.
It's
a
tree
branch
of
this
deer.
L
If
there
is
flooding,
I
will
know,
because
somebody
will
see
will
update
the
r9
and
the
r9
will
propagate
using
signal
free,
multicast,
so
signal
free,
multicast,
I'm,
not
sure,
is
good
for
everything,
but
for
this
special
Network,
it's
very
very
good,
because
there's
many
channels
with
not
so
many
subscribers
to
them.
Okay,
so
basically,
this
is
how
it
works.
L
The
aggression
is
going
on
with
multiple
companies.
We
think
this
stack
for
mobility,
which
adds
to
the
generic
stack,
can
help.
Multiple
companies
engage
in
using
lists
as
a
mobility
network,
specifically
Intel
Toyota
and
specific
carriers
like
Verizon,
AT&T
and
LG
you,
and
we
are
also
integrating
tuning
into
the
public
safety
channels
with
navigation
apps
that
are
now
going
to
see
as
an
overlay
as
a
graphical
overlay.
L
The
condition
ahead,
heads
up
for
this
stress,
driving
and
not
to
get
into
how
to
negotiate
road
situations
always
right
there
on
your
navigation,
app
is
kind
of
a
strip
so
by
by
factoring
out
state
and
making
it
addressable.
We
opened
up
the
system.
We
are
making
interoperability
easier.
There
is
interest
to
host
these
states
to
interact
with
these
states
and
I.
L
Think
we
brought
it
to
a
maturity
level
which
is
pretty
reasonable
and
hopefully
become
part
of
the
workgroup
items
going
forward,
because
mobility
is
quite
important
these
days,
but
we
kept
the
draft
such
that
it
does
not
change
the
list
protocol.
So
changing
from
HIDs
to
e
ids
is
algorithmic.
There's
no
need
for
special
consideration.
We
use
RFC's
six,
eight
three
zero
and
a
signal
free
in
a
very
specific
way
which
we
have
documented.
L
We
are
that
in
terms
of
the
BDD,
in
order
to
do
that,
and
we
are
able
to
sort
of
give
development
kit
or
just
make
people
use
list
map
stacks,
so
they
can
integrate
mobility,
clients,
first
responder
applications
command
and
control
by
Highway,
Patrol's
and
other
such
applications
in
entertainment
center
in
cars.
So
the
potential
for
interoperability
here
is
really
cool.
F
L
G
So
it's
a
combination
of
use
cases
that
are
basically
bringing
together
things
that
in
the
last
few
years
we
thought
were
possible
with
the
least,
but
what
I
think
is
very
helpful.
This
use
case
here
is
bringing
all
together
in
one
you
still
with
a
very
clean
practical
application,
also
I
agree
which
around
that
this
is
specific
to
this
particular
abuse
case.
But
I
think
is
not
very
hard
to
think
of
other
use
cases
that
are
not
related
to
vehicle
mobility
and
traffic
road
conditions
that
may
have
similar
attributes
right.
G
You
want
to
have
basic
the
mapping
service
that
is
highly
distributed
available
at
the
edge
of
the
network
that
offers
you
quasi
real-time
information.
So
I
think
this
is
very
general
in
many
application
that
we
see
coming.
So
that's
why
I
think
there
is
a
lot
of
value
in
you
know,
working
on
this
use
case
and
basically
trying
to
see
how
all
the
things
that
we
have
done
in
the
least
working
group
fit
together,
and
it
would
be
interesting
to
see
if
you
know
there
are
new
things
that
are
needed.
G
L
Yeah,
thank
you
for
that.
Fabio
like
that.
There's
English
those
many
things
we
selected.
What
is
very
attuned
to
this
use
case
and
we
use
it
in
a
very
specific
way,
but
it's
also
important
for
us
to
stay
in
contact
with
this
revolution.
For
example,
if
I
have
a
deployment
of
Verizon
and
AT&T
I
would
need
an
overlay,
so
it's
important
to
stay
with
with
the
evolution
of
list,
even
though
there
is
a
twist
here
which
is
we're
talking
to
tiles,
and
that's
the
twist
here,
the
indirect,
the
broker.
D
J
D
D
We
could
Publishing
the
informational
document,
because
this
deals
with
geolocation
topics,
I'm
trying
to
find
out
from
the
geolocation
experts
here
at
the
IETF
how
they
want
to
make
sure
that
appropriate
review
from
the
teams
who
have
dealt
with
these
problems
before,
because
this
is
not
a
new
topic
to
the
IETF,
then
the
usage
is
new,
but
the
general
topic
of
communicating
information
about
geo
locations
and
such
like
his
one
that
they've
worked
on
a
lot.
So
I
want
I'm
going
this
week
to
talk
to
some
of
the
people.
D
D
Historically,
there's
been
sensitivity
when
we
deal
with
geographic
information,
so
I
want
to
make
sure
not
that
I
want
to
make
sure
we
get.
The
review
and
involvement
for
people
have
been
in
that
space
and
I,
don't
care
which
working
group
ends
up
owning
the
problem.
It
may
be
that
what
we'll
do
is
ask
you
to
present
this
material
at
the
ops
area
group
next
time
or
something
I've
got
to
talk
to
folks.
I've
talked
to
one
of
the
ops
ADEs
I've
talked
to
some
of
the
others.
I've
talked
to
some
of
the
geolocation
people.
D
D
D
Making
it
informational
or
just
any
type
of
work,
just
whatever
type
I
mean
it
looks
to
me
like
it.
It's
mostly
informational.
It
needs
just
to
find
a
new
al-kahf,
but
that's
all,
and
the
al
khair
for
registration
procedures
would
allow
that,
but
that's
a
that's
an
it
in
the
picture.
As
far
as
I'm
concerned.