►
From YouTube: IETF100-ROLL-20171115-1330
Description
ROLL meeting session at IETF100
2017/11/15 1330
https://datatracker.ietf.org/meeting/100/proceedings/
A
A
Thank
you
to
Dominique
for
a
take
minutes,
Peter
Porter's
prayer.
This
is
our
agenda,
so
how
about
the
outdoors
are
here?
Okay,
our
milestones.
We
can
see
that
we
are.
We
have
some
delays
with
some,
so
we
tried
today
to
discuss
about
that,
and
how
can
we
improve
them?
We
need
like
watering
people
that
are
willing
to
do
work.
A
A
A
So
another
post
update
to
zero
zero.
There
were
able
option
type
indicate
that
they,
when
an
IP,
ipv6,
node,
find
and
hot
by
hope,
either
and
doesn't
understand.
It
should
escape
that
optional
key
process
in
the
packet,
so
that
comes
a
flag
day
between
the
X.
When
there
is
the
network
that
handle
63,
the
previous
call
a
papal
option
type
and
the
new
one
so
I
proposed,
as
was
suggested,
the
last
meeting
us
a
total
configuration
option,
but
we
are
going
to
put
the
bone,
so
you
say
flag
on
that
cooperation
option.
B
C
The
various
deployments
I'm
aware
of
including
the
ones
that
my
company
participate
to,
they
don't
deploy
ripple.
They
deploy
something
like
Y,
Sun
or
wireless
heart,
or
something
like
that,
so
that
there
is
a
consortium
of
sorts
which
Peaks
a
number
of
features
of
documents
of
RFC's,
I,
three
respects,
etcetera
and
says
out
of
those
bikes
I'm
going
to
do
this.
C
C
Without
you
know
some
kind
of
a
certification
body
which
which
packages
something
and
say
this
is
interoperable,
so
I
don't
really
mind
about
the
flag
day
I
mind
that
a
certification
of
some
formats
at
some
point
decides
to
go
to
this
new
setting
to
go
to
six
large
and
a
number
of
other
things
which
individually,
each
each
of
them
creative,
like
that,
even
going
from
I
Triple
E
before
2015
to
2.4
a
to
do
that
form,
802
dot,
15.4
2015
creates
a
flag
day.
I
mean
not.
C
D
D
Still
there
will
be
an
upgrade
cycle,
in
which
case
this
would
be
needed.
So
what
I
I
would
really
like
to
see
this
flag.
The
flag
proposal
to
be
kept
in
I
mean,
frankly
speaking,
I
don't
have
a
problem,
keeping
this
flag
in
a
separate
document
or
in
the
same
document
as
well.
But
what
I'm
trying
to
say
is
even
if
the
certification
body
accepts
to
decide,
it's
not
like
all
the
nodes
are
going
to
get
upgraded.
D
C
If
you're
talking
about
backward
compatibility
within
one
of
those
tunnels,
yes
this
primary
arise
and
basically,
if
your
question
is,
is
there
a
backward
compatible
text,
that's
needed
to
say,
for
instance,
I'm
seeing
the
old
flag
in
the
back
yet
I
put
the
new
flag
instead
or
not
or
something
or
you
say,
the
network
cannot
have
backward
compatibility.
Yes,
you
could
spell
that
just
backward
compatibility
section
is
useful.
I
agree
with
you,
but
but.
C
A
A
F
I
still
have
a
question.
This
document
is
in
very
group
last
call.
The
very
group
has
Co
finished
about
today.
As
far
as
you
know,
I
have
not
seen
any
reactions
on
the
mailing
list
which
disappoints
me
a
bit.
I
must
say
so.
Please
have
an
review.
The
document
has
been
reviewed.
Every
passed,
I
know
that
there
has
been
happy
quite
a
lot.
Yeah
I
know
it.
Nevertheless,
you
also
have
to
go
to
six-month
or
brief
before
this
yeah.
F
D
Hello
from
hobby
and
this
work
is
about
so
it
was
that
update
last
two
last
ITF
and
in
the
previous
IETF
there
were
some
suggestions
from
Pascal
Michael
about
the
guy
about
not
changing
the
existing
bets
in
the
Dow
message
and
primarily
the
updates
are
about
in
in
that
particular
context.
So
briefly,
for
people
who
are
not
following
this
works,
the
work
is
about
optimizing
their
out
in
validation
procedure.
So
currently
in
RFC,
esperar
see
for
the
6
5
PI
0,
the
proactive
routing
validation
happens.
D
So
if
a
notes
which
is
the
parent,
it
still
sends
a
no
pad
now
from
the
previous
path
and
it's
not
optimal
by
design.
So
what
we
intend
to
do
here
is
in
place.
The
common
ancestor
node
detects
that
there
is
a
change
in
the
next
hop
and
it
decides
to
sense
a
new
new
message.
Previously
this
used
this.
We
were
calling
it
no
path
down
it's
a
downstream
message,
but
in
this
update
we
are
calling
it
a
new
message.
D
Type
and
I'll
give
rationale
on
why
we
decided
to
use
a
new
message
time
based
on
the
discussions
we
had
previous
ideas.
So
one
background:
as
of
now
all
the
Dow's
flow
upstream,
except
for
our
projection.
I
guess-
and
this
is
the
first
time-
and
maybe
this
is
the
second
time
now-
that
we
have
a
doubt
which
is
going
to
go
downstream.
So
this
is
the
primary
change
in
terms
of
like
overall
semantics
for
Ripper
okay.
D
So
this
is
the
primary
update.
So
we
have
a
new
ripple
message
code.
So
previously
we
are
setting
a
particular
bit
in
the
down
message
which
said
that
if
that
bit
is
set,
the
message
is
going
to
go
downstream.
The
problem
with
that
particular
approach
was
it:
might
it
will
impact
the
existing
implementations?
D
So,
in
the
spirit
of
having
new
implementations
incrementally
update
their
code,
we
decided
to
have
a
new
message
type
in
a
horrible
message
code
and
it
was
been
brought
to
my
notice
that
there's
no
message
codes
are
available
in
pain
20,
so
we
could
really
use
those.
These
bits
are
really
reducing.
D
G
D
D
Alright,
so
these
are
the
rationals
I've
already
talked
about
it.
Existing
implication.
Didn't
implementations
do
not
need
to
change
and
the
ripple
message
codes
are
one
plaintiff.
Also.
Well,
you
added
a
new
message:
along
comes
the
ACK
and
you
need
DCO
a
call.
So
in
this
context,
so
in
a
way
we
are
replicating
things.
D
So
I
would
really
like
to
have
some
feedback
from
the
working
group
on
whether
we
add
a
new
message,
which
I've
already
done.
I've
already
updated
the
draft
to
have
a
new
message
still
open,
but
it
comes
with
some
package.
So
now
we
have
what
DC
or
ACK-
and
one
thing
is
not
where
that
is
not
mentioned
in
this
PowerPoint
is
you'll
need
also
the
secure
version
of
this.
D
H
A
D
Yeah
yeah
yeah.
In
fact,
you
know
we
have
been
working
towards
simulating
this
at
a
larger
scale,
so
how
it
works
in
500
moves
as
metric
lines
and
all
this,
but
regardless
of
whatever
those
changes
are
or
whatever
those
numbers
would
be,
so
the
basic
criteria
or
basic
design
limitation
that
this
problem
handles
that
that's
already
there
in
place
like
you,
are
taking
a
path
which
is
suboptimal
that's
by
design,
and
we
want
to
improve
on
that.
So
I
can
give
the
numbers
actually
to
this
working
group.
Maybe
next
time
yeah
in.
I
F
Mom,
okay,
fine
excellent
someone
else,
the
more
the
better
so
to
be
nice
to
have
to
review
and
then
have
feedback
on
the
draft.
Are
you
going
to
deploy
it
anywhere?
If
do
you
have
plans,
apart
from
the
implementation?
That
also
will
be
in
an
installation
and
it
plans
about
that
I'm?
Sorry,
there
is
an
implementation,
but
will
to
be
an
deployment
of
the
in
stay
well.
Yeah.
D
C
A
C
C
C
That's
most
of
the
energy
that
was
spent
on
this
was
spent
on
the
fast
reroute
and
I
will
show
you
why
in
the
next
presentation,
so
that
that's
about
it.
The
biggest
questions
we
have
on
this
document
is
not
really
on
the
document
itself,
but
on
how
a
controller
would
get
the
full
topology
of
the
network
so
as
to
to
design
all
those
project
abroad.
So,
basically,
this
document
is
not
a
standard
document.
It
cannot
live
without
the
help
of
something
to
provide
the
topology.
C
C
But
if
we
want
a
Majin
Airy
case
of
the
projection,
then
we
would
need
another
protocol
or
some
extensions
to
repo
to
give
us
more
information
about
the
the
links
and
so
I
wonder
if
that
should
be
in
this
document
as
well
or
if
it's
another
document,
and
if
there
is
advice
in
the
room,
then
Oh
taking
it
yeah
Ravager.
The
warming
I
definitely.
G
F
So
what
have
we
seen?
We
have
a
lots
of
drafts.
Some
of
the
drafts
are
not
going
forward,
they
seem
to
be
blocked
and
there
seems
to
be
a
reason
for
that.
On
the
other
hand,
we
have
drafts,
which
are
on
the
same
subject
and
I'm
afraid
that
are
so
interference,
and
you
should
like
to
see
how
we
go
forward
and
that's
what
I
wanted
to
discuss
do
not
have
high
expectations
like
we're
going
to
solve
this
today.
F
That
makes
a
difference,
so
we
have
here
the
topics:
I've
separated
them
in
young
models,
which
has
been
has
had
lots
of
progress.
We
have
a
set
of
I,
think
independent
topics,
so
we
won't
spend
much
time
on
it.
Then
we
have,
in
my
opinion,
duct
manipulations,
which
I
think
we
should
think
what
they
mean
actually
for
the
deployment
of
the
said
of
the
ripple,
and
then
we
have
bitmaps
and
boom
filters,
which
is
a
new
subject,
but
for
which
Pascal
has
introduced
that
there
are
lots
of
solutions
to
go
forward.
F
F
You
should
do
all
along
because
about
modeling,
and
you
have
to
have
a
vision
on
how
to
do
it
and
the
working
group
RTG
area
thinks
that
is
important
to
have
these
young
models
and
I
think
they
are
important,
because
once
you
have
deployed
ripple
and
maple
etc
reasoned
in
the
field
you
like
to
manage
them
and
get
your
information
and
see
how
well
things
are
going
such
that
you
can
sit
back
and
yep
so
who
is
willing
to
collaborate
on
these?
It
is
a
bit
boring.
F
I
know,
I
tried
it
myself,
but
you
also
need
a
lot
of
knowledge
and
because
you
cannot
make
this
model
just
by
looking
at
the
draft
and
then
picking
out
all
the
attributes,
I
mean
it's
not
enough.
You
have
to
know
what
they
are.
What
they're
doing?
Actually
so
I
am
really
looking
forward
to
some
people
who
are
willing
to
either
collaborate
with
the
maple
one
or
me,
or
to
start
on
the
hippo
Vaughn
and
I'm
also
willing
to
collaborate
on
that
volunteers.
D
F
Could
include
it
yeah
I,
don't
think,
there's
any
problem
there.
It's
you
get.
You
have
to
network.
The
network
has
only
servers
where
all
these
maple
things
are
happening,
and
so
you
want
to
get
a
good
view
of
what's
happening
there.
As
such,
they
can
make
decisions
about
the
policy
changes,
switching
of
nodes
or
whatever
and
I
think
all
the
information
that
helps
you.
Even
if
you
have
high-level
information
which
is
not
actually
described
in
the
drafts,
but
you
think
are
very
useful.
It's
there
to
put
in
I
think
so.
H
F
F
F
Game
next
subject,
so
we
have
a
set
of
independent
topics.
We
have
use
of
our
people
info
a
eau-de-vie
ripple
which
is
use
them,
which
I
think
are
quite
independent,
so
they
can
be
put
into
the
in
the
stack
and
they
know
that
there's
no
work
there's
a
flag
day
necessary,
but
it
doesn't
really
affect
all
the
other
protocol
stuff
yeah.
F
We
have
to
forward
a
select
which
is
me,
I
mean
it's
also
quite
independent,
I'm
alone
I
mean
so.
If
anybody
wants
to
help
there,
then
I
will
be
very
happy
to
receive
and
co-author,
but
as
long
as
I
stay
alone,
I
don't
think
it
will
go
forward
as
it
is,
and
then
we
have
to
I
think
that's
it
so
I,
don't
think.
There's
too
much
discussion
and
necessary
about
all
those
droughts.
D
A
F
B
F
F
F
As
far
as
I
understood
and
there
are
different
results,
and
so
it
might
mean
that
there
is
no
coherent
view
from
end
user
of
the
network
or
manager
of
the
network
of
which
algorithm
should
be
used
and
what
the
selection
criteria
are,
which
you
induced
I'd
like
to
look
at
the
example,
for
example,
of
looking
at
the
connections,
and
how
should
you
optimize
the
routes
through
the
repo?
So
you
have
this
measure
essential
so
and
you
can
have
different
measures
in
it.
It's
clear
you
put
them
in,
and
the
network
works
like
that.
F
That's
not
the
case
here
so
I
saw
that
possibly
we
might
have
an
RAF
which
tells
you
how
to
modify
the
ducts.
So
we
have
a
uniform
way
of
handling
the
dark
so
that
it
can
be
handled,
and
then
we
have
all
the
day
current
ways
to
handle
them
and
eventually
decide
how
they
can
interoperate
in
an
interoperable
way
such
that
the
devant
effect
is
reached.
Any
suggestions
about
that
for
the
moment.
For
me
they
are
standalone,
possibly
contradictory
drafts
I.
C
Adopt
something
later,
it
has
to
be
non
conflicting
with
existing
work,
obviously
or
state
that
is
either
oh,
but
but
we
very
clear
on
that
yeah.
The
way
you
see
it
is
that,
ultimately,
you
know
we
have
done
6550,
which
was
complex
enough
like
that,
and
we
want
it
to
ship
it
so,
but
we
knew
there
were
things
missing
and
the
content
P
Dow
find
since
is
something
that
we
could
figure
out
the
time
or
just
being
able
to
pull
some
dowels
from
selected
targets.
C
So
we
don't
have
to
have
all
the
states
of
the
network
when
you're
on
story
mode.
We
knew
all
those
things
and
we
didn't
act
on
them
yet
because
it
will
be
now
a
number
of
small
additions
which
could
be
options,
for
instance
in
the
final
document.
What
I
see
is
that
in
the
future,
we'll
probably
do
a
new
repo
will
do
will
take
6550
will
take
all
those
amendments
and
we'll
make
the
internet
standard
for
repo
right.
C
You
see
we'd
have
to
take
it
from
normal,
it's
the
track
to
internet
standard
and
when
we
do
that,
just
like
a
pv6
did
with
8200
will
package
all
those
documents
which
which
became
smaller
of
C's,
and
we
will
write
the
current
version
of
the
whole
thing
right
at
the
moment.
All
we
want
to
do
is
do
them
one
by
one
make
sure
they
fit
well
within
repo,
explain
what
they
do
and
when
we
do
this
internet
standard
we'll
do
the
packaging.
Rightly
that's
how
I
see
it.
C
F
Let
me
summarize
first
understand
correctly,
you
say
it's:
we
have
two
drafts,
they
are
on
stand
alone,
they
are
correct
and
they
have
described
correctly.
It's
only
during
the
deployment
that
might
be
interaction
which
are
not
possible
to
be
foreseen,
and
if
you
want
to
solve
them,
we
wait.
We
go
to
the
understand
that
internet
standards.
C
What
I
said
is,
as
we
take
them,
and
we
took
to
only
the
other
ones,
I
still
personal
submission.
We
have
no
garage
and
we
expect
that
they
interact
without
an
issue
right,
it's
down
design
for
that.
Unless
we
were
very
wrong,
they
okay
and
we
just
pick
them
one
by
one,
make
them
arrive,
cease
and
when
we
happy
will
do
the
reps
eat.
C
I
am
65
fifty
and
we
gather
some
experience
from
the
field
about
how
the
internal
work
and
everything
then
we
write
the
internet
standard
which
wraps
them
together
and
and
if
we
found
some
small
issues
in
the
field,
then
we
in
the
Internet's.
Now
that
would
say
by
the
way.
I
should
do
this
with
this
one.
Don't
do
that
with
that
one,
but
we
will
take
all
the
text
and
and
put
it
together.
Ok,.
F
D
D
F
F
That
was
the
next
stage,
so
they
had
to
I
had
to
variation
this
and
the
one
is
the
collaboration
between
the
different
house.
I
mean
I,
understood,
Fusco
Nazim,
everybody
seems
to
concede.
We
first
have
to
gain
more
experience
and
then
we,
if
we
go
to
internet
standard,
that's
when
we
write
it
up.
That's
what
we
understood:
fine
yeah
and
the
second
one
is.
F
D
C
So
basically,
what's
happening.
Pascale
again
is
that
we
are
experiencing
situations
in
the
field
or
getting
requirements
from
some
environments
and
with
proposed
solution
for
those
as
amendments
to
repo,
we
are
not
incapable
to
say
everything,
that's
wrong
or
could
be
updated
in
repo
and
write
the
problem
statement
on
that
I
just
don't
know
what
I
know
is,
prime
by
prime,
what
we
are
facing.
We
are
facing
crimes
like
big
geo
tags
and
no,
it's
wanting
to
join
it
out
do
decade.
We
are
talking
about
that
at
six.
Now,
we'll
bring
it
here.
C
We
are
experiment,
we're
experimenting
packet
loss
when
the
Doug
reorganizes.
We
want
to
avoid
that
that
the
first
we're
not
saying
so,
each
time
we
experience
something
we
try
to
come
up
with
a
solution
which
is
compatible
with
what
already
exists
and
propose
that
addition,
as
an
amendment
to
repo,
but
it's
not
like
I,
have
an
O
encompassing
view
of
everything
that
people
needs
right.
So
I,
starting
with
a
prime
statement.
My
view
is
no
go.
We
are
fixing
prime,
which
is
diversity.
Okay,
so.
D
So
I
was
not
suggesting
actually
a
major
major
major
change
to
the
ripple
or
anything.
So
I
was
suggesting
specific
amendments,
given
the
specific
section,
an
example
would
be
the
use
of
DTS
and
the
change
in
DTS
in
number,
and
it
has
to
be
burned
out
into
the
device,
so
it
has
to
be
stored
in
the
persistent
memory.
Well,
there
isn't.
That
is
not
practical.
D
F
F
F
C
Number
of
things
actually
which
relate
to
the
same
world
and
correctly
packaging
them
on
this
slide
and
right
today
they
are
packaged
as
two
documents,
plus
an
encoding
one
and
III,
because
we
are
this
right.
I
propose
a
number,
a
number
of
solution,
but
I
just
I
didn't
say
we
have.
We
need
to
have
two
documents.
I
said
we
could
do
this.
We
could
do
this.
We
could
do
this.
Let's
propose
that
to
the
group
and
see
what
the
group
thinks.
Okay
would
put
a
slide
with
this.
Yes,.
F
That's
what
I
thought,
but
as
I
won't,
do
not
to
have
already
the
decisions
about
which
documents
are
going
to
be
down
here,
but
poorly
discussed
later
on
the
work
on
the
mailing
list.
I
sorta
shouldn't
put
the
forward
here,
yeah
that
was
the
ID.
Is
there
anyone
who
is
interested
in
this
in
this
topic?
I
thought
it
might
be
interesting
if
you
wanted
to
reduce
actually
the
the
tables
at
set
and
the
headers
for
the
during
the
routing.
So
having
that
and
having
salts
awesome,
good
subject.
D
So
I'm
definitely
a
I,
don't
know
the
details
of
how
be
are
actually
woods,
but
I
have
seen
the
problem
statement
and
I've
seen.
The
initial
I
have
not
gone
through
the
details
of
this.
The
problem
statements
that
it
tries
to
solve
are
pretty
interesting
and
it
would
reduce
control
overhead
by
a
big
factor
is
what
I'm
trying
to
say,
but
given
the
fact
that
there
are
a
lot
of
things
that
are
there
already
in
that
context,
I
don't
know
how
to
so.
D
C
Okay,
so
first
you
got
it
right.
There
is
a
there's,
a
non-coding
which
takes
a
lot
less
space.
They
are
full
appear
to
us
for
unicast
and
mostly
multicast
operation,
using
bits
which
point
at
each
individual
destination.
That's
beer,
so
that
you
can
say
I
want
to
set
it
to
do
specific
five
guys,
just
by
putting
five
bits
in
a
bitmap
and
sending
to
the
bitmaps.
That's
yeah,
it's
as
simple
as
that.
Now
the
those
two
trucks
combined
in
different
fashion,
three
dimensions.
C
C
So
you
have
a
fixed
number
of
nodes
in
your
network
because
that
the
size
of
your
bitmap
or
do
you
express
that
as
a
bloom
filter,
which
ICS
a
optimization
which
gives
you
some
more
elasticity
on
how
many
devices
you
can
have
in
a
network
at
the
price
of
more
complexity,
because
you
have
to
do
all
those
ashing
and
encoding
and
stuff
like
that?
Distribute
the
action
stuff
so
first
dimension,
storing
non-story
in
six
second
dimension,
encoding,
a
bitmap
and
the
vs.
C
bloom
and
the
third
one
is-
is
ears
in
mind
that
now
it's
just
scripts
but
anyway,
so
I
must
have
it
in
my
in
my
mail
point
being
that
you
could
decide
now
how
you
structure
the
work.
If
we
decide
to
go
to
it,
for
instance,
do
we
want
sub
to
chest
out
because
it's
three
dimension,
like
nine
documents,
you
know
I'm
doing
storing
with
bloom?
C
You
know
that
could
be
many
actions,
although
you
say
Oh
bloom
is,
for
instance,
and
optimization
that
that
goes
on
top
of
that
it's
compatible
with
everything
on
the
of
possible
modes,
in
which
case
I
could
say,
there
is
a
what
I
do
without
bloom
and
there
is
how
I
blue
eyes
it'd
be
optional,
so
this
is
kind
of
the
discussion
I
want
it
to
have.
Do
we
want
to
do
storing
non-story
beer,
bloom
source
route
destination,
all
those
things
it
is
it.
F
Okay,
so
I
think
we
have
had
those
presentations
from
she
cast
and
your
presentation
on
the
last
ITF
I
don't
see
the
that
at
this
moment
is
enough
knowledge
here
to
to
see.
How
do
you
have
to
go
forward?
May
I
invite
you
for
the
next
in
London
that
you
give
in
20
minute
presentation
about
all
these
yep.
F
A
A
J
Hi
everybody
mama
cousin
from
Edinburgh
University.
So
this
is
the
third
version
for
the
draft
and
just
a
quick
hint
here.
We
combined
in
this
draft
to
draft
the
current
draft.
Besides
the
who,
from
how
are
we
dropped?
Okay,
so
we
already
presented
this
in
the
previous
in
98
Chicago
meeting.
So
let's
skip
this
and
just
lift
these
slides.
J
Besides
those
parameters
and
the
content
of
the
metric
of
of
the
new
metric,
as
you
can
see,
those
parameters,
the
flag
and
the
be
flag
which
is
in
the
case
0
and
1,
whether
it
contains
the
parent
address
or
not
depends
and
I
will
explain
this
in
the
next
slide
and
the
number
of
direct
children
for
the
CNC.
At
the
moment
it
can
be
optimized
and
the
future
to
consider
the
accumulated
number
of
children,
not
only
the
direct
number
of
children.
J
If
the
bones
will
see
that
and
the
max
number
of
children
that
can
hold
can
be
whole
and
the
parent
address
optional,
it
depends
optional
or
not,
based
on
the
mode,
as
you
can
see
here,
if
it
is
in
the
storing
mode,
then
the
number
of
children
can
be
counted
from
the
DAO,
and
this
is
for
the
upstream
and
the
DI.
All
just
inform
the
number
of
the
data
streams
and
the
proposed
options,
as
you
can
see
at
the
bottom
of
the
slide.
J
J
Yeah,
ok,
so,
as
you
can
see
here,
so
this
d
io
will
back
to
this
parent,
ok,
so
by
injecting
the
parent.
Okay,
the
the
parent
can
count
how
many
children
okay
has,
and
in
this
way
we
can
utilize
the
di.
Oh
okay,
in
efficient
way
to
count
the
number
of
children
and
the
proposed
rank
calculation
will
be,
as
you
can
see,
from
this
formula.
Okay
and
the
255.
It
indicates
that
this
is
the
maximum
number
of
children
can
be.
The
hold
cannot
hold
any
more
number
of
children.
Yes,.
B
E
B
J
For
the
sake
of
stability
issue,
actually
here
we
can
use
another
metric
along
with
the
number
of
children.
So
this
is
not
the
sole
okay
metric,
okay
to
optimize
the
load.
Balancing
and
here
I
want
just
to
highlight
very
important
point.
We
are
not
here
measuring
the
accurate
100%
traffic.
We
are
here
just
to
optimize
the
traffic,
because
this
is
random
deployment
and
we
are
not
counting.
It
is
indirect,
it
is
indirect
and
direct
just
we
can
optimize
the
traffic
as
much
as
possible
in
this
way.
Yes,.
G
C
G
C
So
so
let
me
give
some
some
background
information
here,
which
is
coming
from
six
dish,
so
for
the
last
2-3
ietf,
six
dish
has
been
considering
some
primes.
We
have
in
the
field
again
which
have
to
do.
For
instance,
if
I
have
too
big
do
tags
or
one
smaller
than
the
other,
and
that
is
not
that
needs
to
join,
which
of
those
to
do
tags
if
he
sees
both,
which
one
should
he
join,
and
so
it's
a
problem
we
have
in
the
field.
C
C
C
So
you
see
there
are
a
number
of
metrics
which
affect
the
willingness
for
a
particular
node
to
be
joined.
For
instance,
of
an
India
storage
I
have
to
store
the
towel
in
story
mode
right
stuff
like
that.
So
so
there
are
many
metrics.
They
are
not
useful
routing.
There
are
different
metrics,
but
we
could
propagate
them
through
metric
containers.
C
That
seems
to
make
a
lot
of
sense,
and
we
would
like
to
advertise
that
to
note
which
one
to
join
the
this,
these
parents,
which
advertises
this
so
it
sits
next
to
the
ranch,
but
it's
used
just
for
the
desire
of
joining.
So
we
are
working
on
that.
We
are
trying
to
express
the
metrics
we'd
like
to
see
there.
We
are
trying
to
express
how
that
could
be
going
down.
You
know
from
the
upper
layers
the
rotting
layer.
C
Well,
this
is
understood
from
the
matrix
to
the
lower
layer,
all
the
way
down
to
the
beacons
so
that
it
can
be
expressed
in
the
beacons
that
you
hear
when
you
decide
to
join
the
network.
So
we're
trying
to
look
at
that
and
when
we're
ready
and
when
we
understand
what
we
really
want
to
do,
then
we
probably
will
come
to
roll
and
say
a
in
this
other
whole
design
that
we
see
there
is.
We
need
those
metric
containers.
Why
don't
we
piggyback
them
into
repo?
C
C
We
have
interesting
talks
on
the
mailing
list,
and,
and
so
we
are
working
on
that,
the
number
of
children
is
not
by
experience
the
most
interesting
thing,
but
it's
useful
in
story
mode
decide
know
how
much
if,
for
instance,
there
are
two
parents
which
are
equal
for
already
everything
else.
I
would
pick
the
one
with
the
least
children,
but
first
I
would
pick
the
geotag
with
the
less
traffic
I
would
pick
the
geotag,
which
is
smaller
globally,
not
just
the
children
of
this
before
at
and
then
you
know,
all
things
being
a
qua.
C
Maybe
I
would
look
at
the
number
of
children.
That's
one
of
those
things
which
I
look
at
so
so
this
is
work
which
we
are
doing
at
6.
If
you
want
to
look
at
what
we
are
doing
and
participate
to
the
60
spanning
list
and
then
at
some
point.
Yes,
we
expect
to
be
coming
to
you
going
to
I
Triple
E
as
well
to
do
that
in
an
ie
of
things
like
that,
but
it
really
has
to
do
with
join
nothing.
To
do
with
repo
should
not
be
that's
right.
K
C
What
I
just
said?
It's
just
same
prime
as
the
arrays.
At
some
point.
We
want
to
abstract
what
we
see
and
understand
be
with
complex
matrix
at
the
barrios
into
a
no
pack
that
we
can
carry
in
the
lower
lower
layers,
because
what
no
dlcs
is
a
charm
is
a
beacon.
So
at
some
point
we
want
to
be
able
to
express
some
information
from
the
rata
rotting,
a
router
advertisement,
and
some
information
like
this
from
the
DAO
into
in
a
compressed,
concise
form
into
a.
C
We
want
to
be
able
to
encode
some
of
this
information
down
into
the
the
beacon
and
that's
what
these
drug
free
chats
on
join
thing
does
john
priority
or
whether
it's
coffee,
because
that's
a
big
under,
did
you
I
see,
but
how
we
build.
It
is
not
an
eye
Tripoli
crime,
but
when
we,
when
we
encode
it,
it
becomes
a
discussion
with
the
at
record
because
we
want
to
fit
it
into
the
directly
effect.
Also,
no
back.
L
Thank
You
Pascal
for
introducing
the
work
in
60,
but
I
have
a
question
because
that
the
scenarios
that
can
be
used
in
60
is
narrow
as
far
as
I
not
can
only
be
used
in
the
industry
monitoring.
So
if
you're
saying
that
there
is
a
work
down
to
balance,
they
they
load
in
60
and
I,
think
we
can
say
the
60s
can't
it's
just
like.
You
are
solving
a
problem
in
an
arid
scenario,
but
now
we're
talking
about
and
more
right,
I
mean.
F
C
This
dish
is
the
working
group
where
the
joint
is
being
done.
Ryuuji,
it's
not
because
of
times
that
each
and
they're
happening
it
has
to
do
with
the
fact
that
some
group
had
to
do
the
joint
process.
So
we
are
looking
at
the
joint
process.
What
is
the
security
of
the
joint
process?
What
are
the
flows
for
the
joint
process?
We
get
two
working
group
documents
which,
what
one
is
we
basically
enema
for
IOT
and
the
other
ones?
Basically,
PSK
variety
right.
So
in
that
context,
basically
six
is
doing.
C
It
is
doing
it
for
IOT
at
large
and,
as
we
are
doing,
it
realized
that
part
of
the
drying
process
is
to
select
a
node
that
we're
going
to
join
and
in
that
context,
which
again
has
nothing
to
do
with
the
seh
and
nothing
much
to
do
with
15-4
in
general.
Just
the
fact
that
we
want
to
encode
it
in
an
ie.
C
J
You
very
much
okay,
so,
as
I
said
here,
okay,
so
by
using
a
stresses
threshold,
we
can
just
control
and
the
number
of
children
that
we
can
hold
for
each
node
based
on
the
application
requirements.
So
you
can
K
set
up
the
threshold
according
to
the
application
requirements
and,
as
I
just
want
to
highlight
a
very
important
point,
caressed
by
Dominique
regarding
the
energy
metric
container
for
this
stuff.
J
Actually,
yes,
we
could
use
the
metric
container
for
that,
and
this
is
the
expected
balancing
after
the
after
using
this
metric,
but
for
the
energy
okay,
we
can
see
here,
there's
a
congestion,
okay,
four,
not
three
and,
as
you
can
see
here
from
this
slide,
so
all
these
nodes,
okay
and
I-
done
this
by
Kentucky
simulator
right.
So
when
we
have
this
scenario,
all
these
nodes
goes
to
three
and
left
to
only
with
one
child.
So
now
three,
it
is
congested
with
all
these
nodes.
J
Okay,
now
three,
if
we
use
the
energy
metric
between
three
and
two
to
optimize
the
load
balancing.
Yes,
we
could
do
that,
but
when,
when
the
energy
of
three
dropped
or
a
drain
or
almost
very
close
to
die,
then
those
node
will
be
handed
over
to
two.
We
are
trying
in
this
draft
to
avoid
this
situation.
J
It's
not
necessary
to
reach
the
threshold
of
energy
for
three
in
order
to
hand
over
these
nodes
or
the
sharing
nodes
to
to
okay
from
the
beginning,
once
the
dude
egg,
okay
established
or
constructed
okay,
then
we
have
to
consider
from
the
beginning
this
issue
to
avoid
this
case.
So
this
is
the
main
point
or
the
key
point
from
this
draft
happy
to
take
any
feedback.
How
many.
M
K
B
B
L
C
Basically,
what
mahmoud
said
with
which
I
agree
is
that
at
the
beginning
of
times,
notice
to
join
a
parent
and
it's
a
joint
time
problem
and
that's
why
we
are
working
on
the
join
time
metric
at
six
days
and
like
I,
said.
There
are
many
reasons
why
you
would
want
to
compute
this
in
the
fashion
on
other,
that's
how
it
resembles
a
rank
depending
on
what
you
want
to
achieve.
C
You
have
objective
functions
to
compute
ranks
and
you
need
to
have
some
joint
functions
to
compute
a
join
willingness,
which
will
depend
on
which
matrix
your
care
for
I.
Take
this
example.
If
the
guy
on
the
left
not
for
is
sending
one
packet
per
minute
and
each
of
the
guys
on
the
right
are
sending
one
packet
per
year,
I
would
say
that
the
loaded
guy
is
the
guy
on
the
left
right
and
and
it's
so
it
really
depends
on
your
network.
You
have
to
decide
how
your
objective
function
run
works
to
compute
around.
C
B
L
B
N
B
Sfx
well
and
mr.
X
says
I'm
going
to
use
ATX
and
I.
Don't
need
a
dmg,
so
please
do
writing
one
that
does
use
them.
Gee
I'd
love
that
as
a
co-op
see
all
right.
Thank
you
very
much,
but
no
half
half
talking
in
your
draft.
You
say
comparing
to
previous
situations
comparing
to
EDX
excetera.
We
are
doing
better
but
again,
I
think
we
haven't
exploited
the
full
potential
of
the
dmz
yet
because
nobody.
A
J
G
I
A
It
would
be
nice
if
you
can
put
in
the
appendix
the
performance
that
you
have
with
your
proposal
and
how
it
will
be
with
these
kind
of
other,
with
the
methods
that
already
exist.
You
know
if
your
proposal
doesn't
exist.
What
how
can
we
handle
with
what
whatever
we
we
have
already
pronounced,
compare
and
see?
How
is
the
performance
with
the
comparison?
Okay,
thank
you.
O
K
K
The
context
we
are
working
here
in
the
area
of
the
very
music
networks,
where
the
networks,
in
addition
to
be
reliable,
we
need
to
be
to
reduce
the
idiot
at
minimum.
To
show
one
of
the
potential
solutions
is
the
packet
application,
elimination
and
the
concept
of
theory
is
about
each
node
for
2008.
It's
it's
single
data
packet
and
resume
the
ones
to
his
default,
what
we
call
reporting
and
to
his
alternative
parent.
K
Information
of
the
opponent,
a
maker
here
like
this,
we
increase
the
probability
of
this
information
to
manage
to
destination
so
in
in
60s,
working
with
Wii
Remote.
What's
called
requirements
set
of
requirements
to
talk
of
appearing
to
be
implemented
and
to
work
correcting,
and
some
of
them
are
the
hot
select
the
default
selection,
and
what
are
the
requirements
to
do
for
this
about
the
permissible
hearing
sales
without
acknowledgement
since
you're
transmitting
and
one
will
acknowledge
the
other
overhearing
node
will
not
acknowledge
which
things
to
be
discussed,
the
elimination
issues.
K
K
K
K
K
So
ask
us:
it's
the
four
opponent,
a
from
the
from
the
ripple
acres,
D
and
because
II
as
a
associative
opponent,
then
what
happens
is
that
a
may
have
an
addition
to
a
which
is
default.
Appearances
be
as
a
potential
is
not
in
his
parents,
said
no
rights
or
potential,
and
similarly,
because
he's
default
word
of
the
day,
which
is
here
underlined
and
c-e-o
he's,
including
his
part
and
said.
Finally,
we
have
note
B,
which
works
in
the
same
principles.
Cuz
he's
es
default
button
and
a
is
what
member
of
its
parent
said.
K
What
we
do
here
is
by
default
message
transmitted
everything
this
operation,
then
every
time
when
aim
is
sent
out
within
the
area
with
the
container
and
say
what
we
can
define
here,
this
part
of
mod
set.
So
every
see
is
look
at
here.
My
button
sets
or
CD
in
and
the
same
way
they
will
say
message
is
in
his
neighborhood.
K
This
is
my
policy
now
when
it
comes
time
to
this
to
pick
up
his
alternative,
parenchyma
chick
okay.
This
is
my
the
for
pointing
from
google
right
and
there
is
the
full
equation.
This
is
more
comfortable
to
call
it
since
I
receive
information
from
beans
way
and
I
can
check
that
they
is
very
opaque
okay.
So
what
we
trying
to
construct
here
is
the
letter
based
or
postcode
calls
topology.
K
So
if
the
end
of
the
day
we're
having
one
common
ancestor
between
R
and
B,
which
is
the
organ
for
pests,
so
then
s
we
pick
up,
know
dennis
boren,
be
in
the
same
way
we
have
for
the
other
guy,
so
a
week
of
his
own
day
as
a
default,
and
so
whatever
will
be
see
in
the
same
principles
and
main
work
of
again
people
and
d
through
our
audio
message.
What
when
constructed
here?
K
So
this
is
to
summarize
that
we're
covering
the
four
part
and
the
alternative.
Thanks
to
this
da-jung
message,
which
we
implemented,
this
planet
mode
set
in
V
everything
these
NSA
object
and
we
sell
via.
We
are
able
to
pick
up
our
alternative
parents
in
order
to
employ
packet,
replication
elimination
procedure.
So
again,
not
that
this
is
small
port
within
the
larger
scope
of
pachyderm
education,
elimination.
There
are
more
works
to
be
done
and,
yes,
we
complementation
Contiki,
OS
and
yeah
some
questions
further.
There
are
like
to
review
and.
D
D
Ola,
just
it
has
to
so.
If
there
is
no
common
subset,
then
this
one
this
would
then
there
is
no
way
to
identify.
Then
there
is
no
way
to
do
packet
about
application
you're
right,
alright,
so
one
more
thing
is
in
case
of
high
densities,
network
yep
yep.
So
there
is
a
limit
on
how
many
parents
said
how
much
of
how
big
parents
at
you
can
send.
Never.
C
D
C
It's
it's
a
lot
worse
than
that
the
scale
again.
What
you
see
I
mean
churches
did
not
have
time
to
tell
you
the
whole
story.
There's
a
big
story.
Part
of
the
story,
is
you
don't
want
any
virus?
What
you
see
is
correct,
but
we
are
doing
Wireless.
We
have
a
lot
lot
lot
lot
more
lost
on
each
hub,
but
yes,
Wendy
sends
to
e.
You
see
this
green
arrow.
C
Because
I
was
applied
unicast,
but
using
multiple
things
at
the
same
time
to
increase
reliability
for
the
exact
same
reasons
that
you
are
advocating
yesterday.
So
so
that's
what
we
are
doing
actually
with
this
thing,
if
it's
centralized
computation,
we
also
practice
that
it
works
really
well,
and
we
have
the
video
ratio
with
in
due
time,
which
is
critical
for
determinism
becomes
good
I
need
to
mention
her
chair.
There
is
IPR
behind
this.
There
is
IPR
behind
us.
K
G
C
So
so
ishyama
is
telling
us
that
network
coding
works.
They
are
and
yes,
I
mean.
What
we
are
doing
now
is
topology.
We
are
building
a
topology
which
enables
replication
and
elimination,
which
is
one
way
of
doing
redundancy
coding
is
another,
basically
see
this
as
a
form
of
flooding.
We
build
something
which
looks
like
a
ladder
you
see,
and
now
you
flood
it
and
it's
not
more
flooding.
When
you
flood
something,
you
see
a
second
copy
of
what
you
already
flood.
You
discard
it.
C
There
are
tons
of
history,
and
now
you
do
flooding
and
how
you
limit
eliminate
the
duplicate
copies
of
flooding.
You
can
do
basic
flooding
with
the
original
frames.
You
can
do
Network
coding,
you
can
do
anything
you
like
the
problem
we
have
here
is
just
to
build
a
structure,
there's
a
limited
number
of
links
on
which
you
will
apply
those
techniques,
and
so
it's
it's
sort
of
an
ode
to
the
fact
that
you
decide
to
the
network
coding,
one
plus
one
redundancy
or
redundancy
at
every
house,
et
cetera.
That
comes
later.
F
F
F
C
Thank
you.
We
discussed
about
Sdn
for
ripple
with
the
Dow
projection.
We
discussed
bf
re-poll.
Now
we
just
discussed
that
net
for
repo.
So
maybe
maybe
it's
time
to
discuss
the
spirit
variable
just
to
make
sure
we've
got
complete
picture
okay.
So
let's
talk
about
far
out
for
a
minute.
So
what
the
prom
is.
So
you
get
this
dude
I
hear
that
you
are
lucky
you
can
see,
but
I
can't
and
so
I
have
the
isolated.
C
So
basically
you
have
to
isolate
a
number
of
nodes
which
are
impacted
by
the
fact
that
you
lost
your
parents.
Basically,
all
the
nodes
which
didn't
have
much
tuition
with
you
to
go
to
the
destination,
and
you
have
to
freeze
them
until
you
have
repented
of
one
of
them-
are
apparently
that
you
can
find
a
new
exit
pass
through
them
and
the
prime
is
with
ripple.
C
So
we
are
looking
at
doing
fast
reroute,
which
is
this
technique
by
which
you
you
go
around
the
problem
to
a
point
where
the
problem
is
no
more
so
so
you
have
to
find
a
way
to
go
to
go
back,
maybe
or
tunnel
somewhere
to
a
point
which
is
not
affected
by
the
disruption.
That's
a
generic
thing
that
first
we
wrote
does
so
that's
that's
why
we
have
this
other
insertions
that
are
being
discussed
at
6-man.
You
know
the
fault
which
does
so
starting
to
to
that
point
where
the
prom
is
avoided.
Instance:
okay,.
J
C
So
there
are
basically
two
approaches
that
we
have
been
discussing
at
six
and
so
I'm
presenting
them
to
you
and
I
was,
and
we
will
see
what
could
be
done
within
roll
if
we
decide
to
take
this
bus.
So
the
first
thing
that
we
looked
at
is
whether
we
well
there
is
a
zero
thing
actually,
but
I
Simon
is
not
there
with
us.
Simon
is
claiming
that
in
most
times,
I
mean
the
the
loops
that
ripple
creates
by
attaching
lower
the
packet.
C
Can
stain
that
look
till
the
new
parent
is
found,
and
sometimes
the
brought
the
packets
are
not
yet
destroyed
and
new
parent
is
found.
So
he
said
that
sometimes
that
saves
the
day.
We
can't
claim
it
safe
that
every
time,
but
it
may
happen
right,
so
we
don't
lose
every
packet
in
the
situation.
Some
are
caught
in
a
loop
till
the
loop
is
resolved,
but
that's
when
it's
very
slow
when
there
are
very
few
packets
right.
So
apart
from
that,
the
sort
of
thing
we've
been
looking
at,
we
have
looking
at
two
approaches.
C
One
years,
the
current
stinking
of
Freeport
is
to
do
things
in
the
data
path.
We
already
have
this
ripple
packet
information
in
ripple
which
allows
you
to
detect
that
there
are
some
writing
errors.
We've
got
like
three
flags
in
the
RPI
that
allows
you
to
know
if
there
is
doubt
or
not
or
if
your
packet
is
going
down
when
should
be
going
up,
and
so
you
have
an
inconsistency
and
based
on
this,
it
can
see
it
consists
in
consistency.
C
You
can
repair,
so
we
already
have
some
data
path,
information
in
ripple
which
triggers
the
routing
but
which
detects
the
error.
So
we
thought
hey.
Why
don't
we
use
the
data
path
to
see
whether,
by
exploring
my
children
I
effectively
find
a
path
to
the
route
which
which
avoids
me
to
wait
for
the
freezing,
with
the
risk
that
I
choose
wrong
the
child,
but
then
I
detected
I,
eliminate
it
and
I.
Take
the
next
chart
see
basically,
rather
than
waiting
to
be
sure
that
this
child
is
usable.
C
Let's
try
to
use
the
shot
children
we
have
and
see.
If
that
creates
a
loop,
let's
pretty
much
the
idea.
If
it
does
not
create
a
loop,
then
I
can
repair
onto
this
guy
and
I'm
all
set.
So
the
first-
and
there
are
two
sub
approaches
to
that.
The
first
is
on
this
screen
here
and
it
basically
says:
let
me
put
a
new
bits
and
we
have
one
bit
left
in
the
API.
So
it's
okay,
let's,
let's
use
the
bit
to
indicate
that
we
are
sending
the
packet
the
other
way
that
we
are.
C
We
are
sending
the
packet
down
so
when,
if
we
see
it
again,
it
means
we
sent
it
to
the
wrong
child.
Let's
not
use
that
one!
So
you,
basically
you
pick
a
child.
Whichever
and
you
say,
let
me
see
if
that
child
does,
that
that
guy
is
Montessori,
a
child,
that
guy
is
my
child.
If
he's
my
child's
gonna
loop,
if
it's
my
grandchild,
etc,
it's
gonna
loop,
but
if
it's
not
my
grandchild,
if
it's
just
somebody
lower
than
me,
but
not
my
grandchild,
then
it's
not
gonna
loop
is
a
feasible
successor.
C
I
can
refer,
unto
him
rapidly
right.
So
you
want
to
discover
that
situation.
So
what
you
do
is
you
effectively
tag
the
backyard
say?
Oh,
that
packet
went
through
me
I
and
you
you.
You
select
one
of
those
guys
and
if
you
see
the
packet
coming
back,
then
this
guy
cannot
be
used
at
speak,
another
one
and
it's
just
just
a
little
trace
that
you
place
in
the
packet
okay.
So
that's
that's!
Basically
you
see
this
year.
The
approach
works,
Rahul.
C
It
does
not,
it
will
just
drop
the
packets
when
they
come
back.
He
may
actually.
Basically,
the
idea
is
you
expect
only
one
breakage
right,
so
you're
the
one
in
front
of
the
breakage
to
expect
there
is
not
another
breakage
which
which
is
being
fixed
at
the
same
time,
if
there
is
too
bad
for
the
guy
below.
But
the
point
is
you
want
to
solve?
It's
just.
You
know
something
to
try
to
do
something
faster,
I'm,
not
saying
it
solves
multiple
breakage
at
the
same
time,
etc.
C
You
expect
one
localized
localized
one
breakage,
so
you
just
pick
one
child
and
you
try
it
so
you
send
frames
and
if
you
see
one
of
those
frames
coming
back,
you
say
wrong
child.
Otherwise
you
keep
sending
your
traffic
there
and
if
it's,
if
it's
an
okay
child,
your
traffic
will
effectively
take
it
go
to
its
destination
with
a
bit
down.
C
C
So
if
there
is,
as
you
said,
because
your
eyes
are
eating
the
prime
correctly,
if
there
is
another
breakage
later,
the
second
guy
who
sees
this
packet
may
think
is
the
one
who
said
the
bit,
and
in
that
case
you
will
discount
the
packet
and
say
I
picked
the
wrong
child,
so
that
might
be.
Confusion
is
out
breakage
that
time
and
and
the
next
slide
will
solve
it.
So
I
don't
want
to
spend
too
much
time.
C
C
Till
you,
the
freezing
source,
so
there's
a
certain
time.
No,
it's
just
that
repo
will
solve
this.
Eventually,
it's
just
that
there
is
a
short
window
of
time,
till
you're
apparent
where,
where
you
don't
have
a
parent
you're
in
this
situation,
where
you're
looking
for
advantage,
it's
like
looking
for
a
job
and
during
that
sort
of
window
you
should
drop
the
packets,
but
rather
than
dropping
them,
you
say
hey.
Let
me
hope
and
pick
one
guy
who
is
lower
than
me
because
I
see
nobody
I
oughta
well
higher
than
me.
C
C
It
does
number
of
drawbacks.
I
just
wanted
to
mention
it
right,
it's
very
hard.
For
instance
it
when
you
switch,
and
you
see
a
packet
coming
back,
it
might
be
one
which
went
through
the
old
path
so
that
there
are
a
number
of
little
windows,
race
conditions
which
are
not
too
keen,
but
is
just
to
mention
that
gives
you
an
idea
of
this
sort
of
thing.
I
want
to
do
in
this
type
of
case,
just
good
just.
C
The
solution
could
be
pick,
the
lowest
Frank,
because
that's
the
guy,
which
you
know
if
first
the
lowest
Frank,
could
be
the.
If
it's
a
child,
then
it's
probably
a
direct
child.
If
it's
direct
child
that
it
will
just
send
you
the
packet
back,
you
can
eliminate
it.
That's
what
I
drew
it's
only
when
you
you
have
to
dig
deeper,
then
then
you,
you
may
have
a
grandchild
and
it's
harder
right.
So
but
the
normal
case
is
you
pick
somebody
with
yeah?
Yes,
Frank
possible!
C
C
The
alternate
is
to
generate
traffic
to
send
a
probe,
but
you
generate
yourself
so
now
you
know
it's
yours,
because
you're
so
service
is
you
right
and
basically
what
you
do?
Is
you
ping
the
route?
We
are
a
number
of
selected
other
guys.
So
you
pick
like
those
three
four
guys
with
the
ideas
Frank
that
you
can
still
see
and
you
send
a
packet
to
the
route
and
you
see
if
it
goes
back
or
if
the
route
responds.
If
the
route
responds
you're
very
sure
that
this
part,
this
guy
is
a
feasible
successor.
C
You
can
repair
it
if
the
packet
comes
back.
Well,
you
eliminate
this
guy
from
your
list
and
you
pick
others
see
so
that
solves
number
of
the
problem
that
we
have
just
seen
in
the
initial
case,
but
you
have
to
generate
new
trafficking,
and
so
it's
basically
an
explorer
of
all
the
guys
behind
you
to
see
if
one
of
them
is
actually
not
parented
via
you.
If
one
is
found,
then
you
just
switch
to
him
and
then
ripple
will
so
you
went
down.
C
Maybe
you
went
down
too
low,
but
now
ripple
will
allow
you
to
go
back
up
right
and
well.
That's
it
and
then
ripple
as
you
go
down,
so
some
other
node
will
decide
to
stay
with
you
or
move
away
etcetera.
That's
no
more
ribbon
operation!
Okay,
that's!
Basically
what
we
could
do
in
Charlie.
Do
you
still
have
a
question
or
I
see
you
well.
E
C
N
N
C
You
could
go
into
ping
multicast,
a
big
world.
This
is
a
series
of
unicast
really
so
you
just
say:
I
have
fought
four
guys,
which
I
think
I
hope.
One
of
them
is
Sasebo
I,
just
ping.
The
routes
throughout
them-
bang,
bang,
bang,
bang
and
the
ping
that
comes
back
to
me
from
the
route
tells
me
that
this
guy
is
useful.
The
I
can
use
any
of
them
till
I,
don't
know
better.
C
N
C
N
C
No
solution
would
be
an
e
phone
on
story
mode.
Mine
works
in
any
case
by
pinging
the
route
I'm
sure
that
I'm
going
higher
than
I
am.
If
you
knew
by
by
heart,
somebody
was
higher
than
you
are,
then
then
you
could
okay,
you
could
ping
your
previous
parent,
and
that
would
work
but
I
don't
know
if
it's
much
better,
maybe
smokers
Li.
So
if
we
free
optimization
space,
we
can
discuss
on
the
meaningless.
Like
Peter
said.
N
I
C
Just
gives
you
that
just
gives
you
ideas
on
how
that
can
be
done
in
the
data
plane.
I'm,
not
saying
it's
a
fine
assertion,
I'm
just
throwing
ideas
at
you
right,
yeah,
the
other
one
which
is
more
of
impact
to
us,
is
basically
to
apply,
centralized,
computation
and
probably
arcs
to
compute
Plan
B's,
in
which
case,
in
which
case
the
doubt,
projection
draft
is
impacted
a
little
bit.
Thank
you
very
much.
I
C
Was
good
so
please
go
through
the
slides.
We
probably
will
do
them
next
time
and
I'll
be
glad
to
come
back.
If
you
invite
me
so
we
discussed
the
data
plane
and
there
is
a
forwarding.
Sorry,
the
writing,
plane
cultural
plane
solution
which,
which
can
do
the
same
thing
and
we'll
discuss
it,
but
it
impacts
the
whole
projection.
That's
why
we
good.
Thank
you
very
much.
A
F
Think
you'll
continue
and
we
certainly
have
to
maybe
organizing
and
discussion
session
next
to
the
the
normal
world
meeting.
So
we
have
to
roll
meeting
may
represent
the
current
drafts
and
we
may
have
a
discussion
meeting
that
we
discuss
all
these
subjects
so
that
you
can
come
to
an
common
conclusion.