►
From YouTube: IETF97-ROLL-20161116-1110
Description
ROLL meeting session at IETF97
2016/11/16 1110
B
A
C
D
A
D
For
the
working
group
last
call,
we
have
had
very
little
return,
so
I
really
encourage
you
to
especially
after
the
nice
dog
by
Michael,
that's
coming
to
comment
on
the
draft
and
at
least
try
to
read
it
and
send
your
comments
like
I
tried
it
and
I
proof
of
this,
for
the
publication
of
this
rafter
will
be
very
welcome.
Indeed.
Thank
you.
A
D
Okay,
lots
of
these
drafts
are
not
just
working
group
to
house,
but
certainly
I
mentioned
in
the
the
Charter,
which
is
still
to
be
approved.
So
we
will
ask
you,
for
if
you
are
really
interested
in
the
dropships,
they
should
become
working
cockpit.
So
we
will
ask,
after
every
presentation
what
you
think
about
if
if
it
should
go
forward,
but
you
have
to
be
confirmed
on
the
mailing
list
afterwards,
of
course,
yeah.
D
A
A
E
Hi
so,
as
was
mentioned,
it's
been
in
working
group
last
call
prior
to
the
or
app
just
prior
to
the
last
ITF
arm.
Six
men
made
some
interesting
changes,
which
were
very
helpful
to
us
arm
in
particular
allows
us
to
leave
RPI
in
arm
on
the
assumption
that
they
will
not
be
processed
whatever
that
means
arm,
and
you
can
go
to
six
men
and
enjoy
that
debate
arm
that
the
so
that's
really
all
that
the
most
of
that's
changed
in
the
past
three
or
four
months
has
been
editorial
clarifying
of
some
terms.
E
E
If
you
need
help
because
they're
animated-
and
I
really
suggest
you
take
a
look
at
it
because
it
really
does
define
something-
has
been
missing
in
this
working
group-
is
exactly
what
the
data
plane
looks
like,
and
this
clarifies
that
I
think
great
deal.
That's
really
it
not
a
lot
else
to
say,
we'd
like
to
publish
it,
get
it
done.
E
B
Avocado
from
hobby,
my
question
is
this:
this
decision
of
leaving
the
rpi
yes
intact,
while
moving
it
out
of
the
network,
the
6lowpan
network
of
whatever
yeah,
so
this
decision
reduces
the
complexity.
As
I
understand
you
know,
after
reading
the
graph
so
far
from
there
is
there
any
like.
Why
was
this
decision?
I
it's
it's?
It
is
possible
for
the
the
edge
router
to
remove
the
rpi
header.
So
why
was
this
decision
taken
to
just.
E
E
Okay,
that
the
the
difficulty
was
that,
given
that
we
can't
remove
it
at
the
edge
and
given
that
the
header
has
been
marked
as
critical
or
it's
a
header
extension
at
hop-by-hop
extension
marked
critical,
given
that
it's
been
marked
critical,
then
the
next
router
that
does
not
speak
understand
it
according
to
historical
understanding
was
intended
was,
would
it
be
expected
to
drop
the
traffic?
Therefore,
any
traffic
that
unintended
ly
would
be
contained.
E
Ok,
ok,
the
downside
is
that,
of
course,
we
would
like
to
send
traffic.
The
whole
point
of
a
mesh-
that's
you
know
connected
by
IP-
is
that
you
might
like
to
send
traffic
outside
of
that
mesh
so
arm
in
order
to
be
able
to
remove
it.
What
we
discovered
was
that
we
have
to
put
I
PA,
p,
headers
in
which
has
a
significant
cost
and
which
we
then
have
figured
out
how
to
compress
such
that
we
can
remove
them.
So
now
that
that
we
we
realize
that
we
are
not
obligated
to
remove
that
header.
E
Then
a
number
of
scenarios
where
we
require
to
put
an
IP
IP
header
in
arm
become
not
so
critical,
ok,
ok
and-
and
that
was
that
that's
that's
really
important,
because
it
turns
out
there's
some
other
scenarios,
independent
of
sending
traffic
to
the
quote
internet.
If
you
have
what
are
properly
called
hosts
or
sometimes
in
this
we
can
group,
we
call
nan,
RPL
aware
leaf
nodes
right.
E
Those
may
not
necessarily
understand
that
header
and
if
we
would
like
to
send
traffic
to
them,
then
we
need
to
know
when
to
remove
that
header
and
weird.
If
we
were
faced
with
the
difficulty
that
in
many
modes
in
particular
story
mode,
we
don't
know
what
the
node
before
the
end
header
node
is,
and
since
we
don't
know
that
node
we
wind
up
having
to
put
hop
by
hop
IP,
IP
headers,
which
means
we
have
to
remove
them
at
every
hop
and
add
them
at
every
hop.
E
F
Let's
get
you
mail
that
there's
a
few
things
we
just
in
flux,
and
there
is
now
the
influence
of
the
new
six
flourish
on
coding,
which
might
actually
change
the
deal
a
little
bit.
First
thing
is
46
man,
nothing
is
recast
in
stone.
I
think
the
the
consensus
right
now
is
that
the
update
of
2560
will
not
change
anything
dramatically
because
you
just
you,
have
to
go
to
interest
energy.
F
You
don't
change
the
way
things
behave,
so
that
I
think
the
consensus
about
going,
producing
new
specs
about
what
you
can
do
and
what
you
cannot
do
and
the
arguments
against
inserting
others
don't
necessarily
apply
to
removing
others.
So
what
I'm
getting
at
is
if
you've
got
this
packet
coming
from
the
ll
n
going
outside
the
other
end
through
the
root,
I
think
it
still
if
the
root
can
the
best
practice
for
us
should
be
to
remove
that
header
now
doing
that
in
the
uncompressed
form
is
probably
cumbersome.
F
The
first
thing
you
find
if
you
have
a
packet
which
does
not
have
a
pin
ID
right,
that's
what
we
are
getting
at.
We
don't
have
IP
in
IP,
but
we've
gotten
our
PR.
Well,
if
you
look
at
the
compressed
form,
the
FBI
comes
first,
which
we
tells
you
that
at
the
moment
where
you
expand
the
64
edge
into
an
ipv6
packet
to
go
to
the
Internet,
you
could
actually
cut
it
up.
You
could
cut
it
out,
yeah
and
just
just
ignore
it.
F
E
E
That
could
happen,
I'm
not
screwing
with
you
and
that
the
documented
says
right
now
essentially
says:
don't
do
for
upwards
traffic.
Don't
do
IP
and
IP
just
do
RPI,
that's
true
for
storing
and
non
storing
mode
and
that's
nice.
It's
a
single
transmit
path
for
upwards
traffic
and
the
decision
as
to
remove
it
or
not,
is
a
decision
at
the
root
which
is
probably
more
capable,
more
configurable
and
more
upgradable.
So
that's
that's
on
some
sense.
A
decision
by
the
operator
to
do
that.
E
As
for
the
arguments
for
removing
the
headers
are
probably
not
as
strong
as
inserting.
Certainly
one
of
the
arguments
is
that
it
breaks
pathum
to
you
is
not
true
that
it
breaks.
Authentication
header
is
true
on
that
where
this
may
not
be
relevant
because
it's
not
clear
we're
using
a
H
headers
ever,
but
there
may
be
other
things
that
matter
so
I'm
comfortable
with
it
is
and
I'm
uncomfortable
the
fact
that
we
can
send
them
out
to
the
internet
and
not
break
the
architecture
as
it
is
I
think
that
that's
a
good
scenario.
E
E
You
know
you
decide
you're
doing
it
it's
on
and
that's
it
there's
no
there's
no
flag
days,
but
but
essentially,
if,
if,
if
a
node
needs
to
know
whether
or
not
it
can
follow
the
use
of
rpi
rules
or
that
should
follow
some
historical
process,
then
I
would
say
well,
if
it's
doing
60
rh
header
compression
as
we've
described,
then
it
should.
It
should
follow
these
rules
and
that's
the
reality.
Whatever
is
doing
happen
before
is
some
other
scenario
right.
F
F
And
so
that's
been
your
debate
as
well,
and
what
past
is
G
now,
because
it's
in
the
RFC
editor
queue
with
the
six
flower
age
is
that
we
actually
snip
those
addresses
so
as
to
save
energy,
etc,
etc,
and
there's
a
discussion
in
the
future
RFC
about
the
fact
that
it
breaks
aah
with
the
fact
that
we
don't
you,
we
don't
have
a
specification
for
IH
in
in
those
networks
anyway.
So
we
don't,
there
is
no
plan
to
use
it
and
we,
the
only
value
of
using
aah
with
rotting
either.
F
E
E
I
agree
with
effectively
the
rules
for
removing
headers
are
as
weak
as
for
inserting
headers
for
our
point
of
view
and
and
I
think
we
could
I
think
we
could
do
that.
I
think
that
we
should
write
that
down
as
a
separate
document
and
have
a
separate
debate
over
what
the
what
the
6
lb
are
should
do
on
outgoing
packets.
E
F
Has
to
say
something
somewhere
all
right,
I
mean
I
agree
with
the
way
document
is
welded
so
far
that
we
don't
do
I,
pinay,
p,
I,
think
it's
safe,
but
now
what
do
we
do
with
the
rpi
I?
Think
it's
always
safer,
not
to
keep
it
in
the
IP
packet
mean.
What's
the
point
in
putting
it
I
mean
at
least
you
know.
If
you'd
get
the
cisco
reg
formulation,
it's
very
easy
to
snip
it
up,
it's
actually
more
complex
to
keep
it
because
you
have
to
reincarnate
into
the
compressible
uncompressed
form.
F
If
you
have
a
non
six
large
packet,
I
would
say:
okay,
pressing
the
packets
always
a
risk.
So
why
don't
you
pass
it
down
so
so
keeping
the
capability
to
either
pass
it
on
or
not
pass
it
down,
but
I
would
say
basically
the
recommendation
would
be
600,
hu
sniff
it
out
plenty.
I
catch!
You
keep
it
something
like
that.
Can.
E
I
feel
I
I
feel
that
that
will
be
that
will
that
will
drag
this
document
into
the
six
men,
morass
and
and
so
so
what
I
would
that's
why
I
would
propose
that
we
we
do.
This
is
a
separate
document
and
we
wait
three
to
five
months
for
2460
bits
to
settle
and
and
then
we
will
either
point
to
the
resulting
thing
and
say
it
says
we
can
do
it
so
we're
doing
it
or
we
won't,
and
that
gets
us
out
of
otherwise
I.
F
E
You
said
you
can
keep
the
rpi
in
the
packet
as
it
goes
out.
It's
not
it's.
You
don't
have
to
remove
it
by
putting
IP
IP
header,
we
didn't
tell
you
to
remove
it.
We
said
you
can
leave
it
in
right,
so
there
so
L
aveiro,
like
do
you
agree
with
me
that
that
I
think
we're
stepping
into
a
disaster
if
we
go
somewhere
else
or
do
you
think
God's
I'm
wrong,
it's
over.
H
A
A
E
I'd
like
to
pop
I
could
be
done
with
that's
what
I
would
prefer.
That's
why
I
don't
want
to
open
a
new
issue,
but
I
mean
tell
me
how
strong
you
feel
about
it.
I
I
think
that
we
could
easily
write
this
as
a
later
document,
because
I
think
that
it's
operate
a
balut
and
it's
easily.
The
route
is
relatively
easy,
upgradeable
versus
the
firmware.
This
decision
has
no
effect
on
the
small
devices
out
in
the
out
on
the
edge
of
the
network.
D
J
G
Okay,
well
I'm
Charlie,
Perkins
and
I'm
here,
to
tell
you
a
little
bit
about
the
updates
to
our
document,
a
0
DB
ripple
and
it's
a
sort
of
a
new
take
on
the
p2p,
our
peer
to
peer
use
of
ripple
and
there's
an
experimental
document.
Rfc
69
97,
but
able
to
V
extends
the
functionality
from
that
document.
G
Things
change
fast
around
here,
so
we
have
a
sub
support
for
symmetrical
and
asymmetrical,
but
still
bi-directional
are
links,
and
this
approach
avoids
the
requirement
for
having
an
address
vector
as
in
the
product
from
the
previous
document,
p2p
RPL.
So
the
way
it
is,
we
have
route
request,
message
and
replied.
This
sub
is
more
or
less
adapted
for
use
with
ripple
from
the
work
that
was
done:
unnao
TV
and
a
0
DB
version
2
and
another
working
group
and
oops.
Now
what
I
do
holy
smokes.
A
G
Must
have
a
big
thumb,
so
the
changes
that
were
made
as
we
actually
wanted
to
reduce
the
amount
of
space
that
is
required
in
a
rare
reply,
and
we
could
do
that
by
actually
having
us
repaired,
dode
eggs.
And
so
we
wanted
to
have
an
easy
association
between
the
route
reply
and
the
request
and
the
way
we
did
that
not
fairly
straightforward
way,
which
I'll
mention
in
just
a
minute.
G
The
route
request
actually
is
sent
from
the
requesting
node
or
the
originating
node
to
finally
received
by
the
target
and
but
the
when,
when
the
target
received
at
what
has
happened
is
that
the
target
now
has
a
way
to
transmit
package
from
from
itself
to
the
originating
node
and
the
route
reply
then,
is
transmitted
back
towards
the
originating
node
according
to
some
path,
possibly
a
symmetric
path
and
when
it
arrives,
then
the
originating
node
has
a
way
to
transmit.
Packets
are
transmitted
data
to
the
target
node.
G
So
don't
breeder
is
that
there's
a
adesh
a3
document
that's
available
and
actually
this
squat.
This
slide
describes
the
changes
between
our
previous
one
document
and
the
end
of
three.
So
the
draft
update
specifies
this
way
of
pairing
the
request
and
reply
instances,
and
it
allows
us
to
eliminate
a
large
field
in
the
route
reply
message
and
we
have
implementations
in
process
there's
a
work
that
you
can
read
it.
G
There
works
on
kontiki
with
the
kuja
simulator
which,
by
the
way,
I
did
not
do
this
work,
but
one
that
co-authors
and
SV
Anand
has
done
this.
So
it
does
and
implementation
supports
about
symmetric
and
asymmetric
modes
and
if
making
comparisons
against
a
ripple
storage
mode,
so
we
will
make
a
more
full
report
on
this
at
the
next
ITF.
This
has
worked
it's
in
process
and
the
intention
is
to
actually
make
a
demo
at
the
next
time
ITF.
G
So
the
pairing
works
so
that
you
can
have
the
upstream
is
considered
to
be
transmission
from
the
destination
or
target
node
back
to
the
originating
node
and
downstream
the
other
direction.
So
the
upstream
sense
instanceid
for
the
route
request
instance
is
selected
to
be
an
odd
number
and
then
the
intermediate
routers
stored,
that
instance
ID
and
then,
when
the
rat
reply
comes
back,
it's
the
route
request,
instance
plus
1,
and
so
that
makes
it
straightforward
for
the
intermediate
routers
to
know
what
is
the
originating
node
towards
which
they
are
establishing
the
reverse
path.
G
So
that's
the
main
technical
difference
of
nuh
is
now
present
in
the
updated
draft
the
I'm
not
going
to
go
into
the
gratuitous
robbery
right.
We
play
it's
it's
the
same
as
it
was
before,
but
basically
on
now
we
allow
it
to
work
with
this
some
way
of
doing
a
paired
pairing
between
the
route,
request
and
route
reply.
Instances.
G
G
B
Giada
from
wally,
so
one
of
the
claim
of
the
previous
draft
I
mean
have
not
gone
through
the
latest
dr.
povitch's,
for
that
the
previous
draft
said
that
this
enables
asymmetric
operations,
a
symmetric
link,
so
basically,
but
the
way
previous
draft
of
operated
was
by
sending
message
from
A
to
B
and
establishing
route
in
the
reverse
direction.
Oh
yes,.
G
G
You
might
want
to
have
the
best
path
going
from
source
to
destination
by
a
much
different
mechanism
than
from
destination
to
source,
but
we
do
expect
at
least
that
the
control
messages
will
be
transmitted
in
the
metrics
evaluated
in
that
way.
So
if
you
want
to
enable
fully
the
more
general
case
for
there
are
unidirectional
links,
then
I
think
additional
work
is
needed.
Ok,
ok,.
B
One
more
question
regarding
the
simulation
that
was
done
in
the
kontiki
so
which
model
was
selected
was:
was
it?
Was
it
you
GD
amor
de
grm,
so
I
mean
basically
I
want
to
understand
if
the
default
mode
was
selected,
which
I
mean
if
you
use
the
default
mode?
Definitely
the
previous
doctor
at
least
would
work
yeah.
G
A
F
So
I'm
past
caring
with
Cisco
presenting
on
the
dial
projection.
So
basically,
we've
kept
this
yeah.
We've
got
this
well
the
repo
protocol,
which
actually
enables
a
number
of
different
ways
of
doing
things,
which
is
so
that
we
could
actually
turn
repo
and
we
knew
it
from
from
the
experimental
RFC
before
into
a
reactive
protocol.
F
We
can
also
easily
turn
repo
into
a
centralized
routing
approach
and
we've
not
done
that
before,
but
there
are
a
number
of
applications
for
which
this
could
be
useful.
An
example
application
is
industrial
control
loops.
So
what
you
build?
This
ripple
based
cluster
of
sensors
and
actuators
and
programmable
logic
controllers,
and
with
this
you
can
enable
control
loops
inside
your
factory
network.
But
then
at
some
point
you
want
to
establish
special
path
between
this
particular
sensor
and
this
particular
actuator
with
guarnteed
capabilities.
F
For
instance,
you
had
a
six
digit
ACH
Mac
layer
and
with
this
Mac
layer
you
can
program
time
slots
so
as
to
to
grantee
deterministically
a
path
between
the
sensor,
this
plc,
programmable
logic,
controller
and
then
back
to
an
actuator.
If
you
want
to
do
that,
you
need
a
a
way
to
go
and
program
the
path,
the
rods
into
the
devices.
So
at
some
point
you
need
to
program
the
time
slot
themselves
and
that's
not
the
objective
at
all
for
this
work.
F
But
on
the
other
hand,
you
also
need
to
program
the
rats
into
the
devices
and,
if
you're
well,
it
depends
on
which
one
are
you
peek.
But
if
you
want
to
use
writing,
you
need
to
go
and
program
the
rod,
and
so
this
is
one
way
of
actually
using
repo
as
the
protocol
to
install
efficiently
a
route
between
the
point,
a
and
point
B
inside
this
disla
policy
Network.
F
If
you
get
a
deep
network
and
you're
using
non
story
mode
at
some
point,
you'd
say:
oh,
if
I've
got
this
particular
long
line
here
on
the
right
and
they
know
the
capabilities,
the
memory
space
that
you
get
in
those
devices
I
could
actually
save
all
these
sort
information
in
all
those
packets
by
installing
pinpointing
some
rods
in
those
particular
nodes,
and
since
it's
done
centrally,
it
can
be
done
with
some
awareness
of
how
much
memory
there
is
in
those
devices.
So
you
know
you're
not
overloading
those
devices
by
this
additional
rod.
F
B
F
So
the
path
is
actually
in
non
story.
Mode
is
actually
already
decided
by
the
dowel
process,
which
creates
the
non
store
across.
So
so
the
cost
it's
done.
Its
work
is
done.
The
topology
is
built.
What
you're
doing
for
the
non
story
mode
is.
You
are
just
compressing
the
expression
of
that
path,
but
you
are
not
changing
the
way
you
compute
it.
So
the
cost
is
past
its.
It's
done.
Ok,.
I
F
F
Ok,
ok,
right
but
you're,
not
it's
not
a
routing
protocol,
it's
just
a
way
of
enabling
a
little
bit
of
storing
mode
in
a
non-story
network
so
that
you
can
transform
streets
of
rotting
in
toulouse,
so
Swati
right
right
and
this
is
doable
because
the
route
is
expected
to
know
the
capability
of
the
mode.
So
it
goes
on
this
node.
I
can
install
5
rods.
F
Let
me
optimize
which
not
get
which
information
now
that
the
version
2
of
this
draft
introduced
a
story
mode
variation
of
that
which
cannot
be
fully
achieved
until
we
send
more
information
into
the
route
about
the
topology,
because
the
non
story
mode
only
work
works
on
the
information
that
the
route
already
has,
which
is
enough
information.
You
just
follow
the
same
time
if
you
want
to
use
the
second
technology
which
is
enabled
by
this
draft,
which
works
in
storing
mode
as
well
towards
in
both
modes.
F
You
need
some
additional
protocol
element
which
this
draft
is
not
providing,
and
we
have
to
decide
if
one
to
take
this
path
and
provide
those
elements
within
the
draft
or
with
another
draft,
and
with
these
additional
topological
elements,
you
can
be
able
to
transfer
soul
path,
divus
tration,
so
let's
go
through
those
slots.
Okay,
so
there
are.
Those
two
modes
will
have
to
decide.
F
If
we
pursue
this
draft,
whether
we
do
or
need
a
non-story,
in
which
case
we
don't
need
new
protocol
elements
or
if
we
also
want
to
do
the
transversal
path,
which
is
really
interesting
piece
in
in
some
scenarios
like
industrial
and
in
that
case,
we'll
need
to
talk
about
topological
elements,
how
we
express
that
to
the
root
or
to
the
PC.
Like
tease,
you
know
this
type
of
information
that
net.
F
So
it's
no
nothing
much
I
I
mostly
did
editorial
work
fixed
the
abstract,
which
was
deal
from
version
1,
so
we
talked
about
so
throughout
and
so
clarified.
What
you
can
do
with
this
transversal
rods,
which
were
introduced
in
version
2
so
I
am,
is
we
could
work
on
the
draft
a
lot?
But
at
some
point
we
really
need
to
see
where
the
working
group
wants
to
do
about
those
two
technologies.
If
you
want
to
pursue
them
and
then
we
can
work
together,
also
type
1.
B
F
We
we
could
have
some
intelligence
so
as
to
refuse
a
down
from
a
child.
If
the
table
is
already
saturated-
and
you
could
do
that,
but
then
you're
pushing
the
Prime
on
the
neighbor
and
maybe
the
neighbor
at
the
same
Prime
and
needs
to
be
pushed
and
some
point.
There
is
no
way
right,
so
so
that
that
kind
of
pushes
the
Prime
a
little
bit
and
enables
a
bit
more
well
a
number
of
new
situations,
but
it
doesn't
solve
the
problem.
It
just
pushes
the
prime
for
further
away
the
approach.
F
Can
it's
very
hard
to
make
the
distributed
approach
work,
and
so
we
can
try,
but
it's
very
hard.
The
centralized
approach
is
doable
because
the
route
knows
which
routes
it's
injecting
into
the
network
right.
So
it
does
that
because
it
has
information
about
the
capabilities
of
the
node.
That's
part
of
the
expectation
you
know
which
notes
you're
talking
about
you
know
how
much
stage
you
can
install
those
nodes.
You
know
the
lifetime
of
this
state,
so
it's
very
controlled.
F
F
F
D
D
I
D
Good
morning,
lots
of
jet
lag
well
I.
Try
to
see
if
there's
any
interest
in
the
work
we
did.
It
is
about
nipple
and
especially
about
the
number
of
four
virus.
You
have
any
metal
in
an
internetwork.
If
you
have
lots
of
network
which
is
very
dense
and
there
are
lots
of
four
virus
in
it
maple
for
routers,
then,
when
you
want
to
multicast
the
message,
you
get
an
enormous
storm
of
broadcast
messages
and
that's
not
really
helping
the
performance
of
the
network.
D
So
for
those
networks
it
is
important
that
you
can
reduce
the
number
of
maples
so
that
the
Equality,
the
density
of
maple
for
virus
in
the
network
is
not
too
not
too
high,
and
you
can
have
a
reasonable
number
of
messages.
So
that's
what
we
try
to
do
here.
I
have
an
example:
robot
you
might
like
yo.
You
have
this
network.
If
they're
all
forwarders
and
one
will
start
the
forwarding,
you
will
get
all
the
other
blue
ones
will
forward
it.
D
D
How
many
forward
a
suit
you
select?
Well,
if
you
have
some
network,
where
you
have
only
one
hope
network
and
then
you
think
more
forward
a
sufficient,
and
you
would
need
one
for
order
because
suppose
that
you
have
a
broadcast
coming
from
the
right
hand,
green
note,
it
was
sent
it
to
everybody.
They
receive
the
message,
but
then
bounced
the
link
is
between
the
dead
link.
D
Has
done
then
you're
very
happy
that
they're
still
and
passed,
which
goes
for
you,
the
forwarder
to
the
other
nodes,
which
has
this
mission
link
so
having
at
least
one
when
you
have
maple,
and
you
wanted
more
reliability
having
at
least
one
for
Weiser.
Also
in
the
single
hope
network
is
sync
encouraged.
D
The
other
case
is
that
you
have
them
too
hot
network,
and
then
you
want
at
least
24
hours.
I
think,
because,
if
the
link
to
some
of
the
forerunners
and
the
destinations
breaks
down,
then
you
have
the
second
forward
to
continue.
So
the
reliable
ability
is
riddled
increased.
We
did
some
measurements
in
an
offices
and
we
sold
that
actually,
the
even
stable
links
a
better,
as
is
very
good,
etc
that
if
you
will
say
20
minutes
per
week,
you
have
links
which
actually
crash,
and
then
you
have
these
two
routers.
D
You
just
continue
working
during
during
the
year
yeah
doctor.
So
what
does
the
note
maintain?
We
might
try
to
keep
to
it.
Another
simple
and
what
you
do
is
that
we
have
count.
The
number
of
neighbors,
which
have
the
state
that
there
for
barter
state
is
FF
means
I'm
a
for
barter.
So
every
note
maintains
the
number
of
neighbors
that
he
has
that
he
has
a
forward
of
is
and
the
other
one
is
that
he
looks
at
the
number
of
neighbors
there.
D
The
number
of
for
virus
is
slow
lagos
by
smaller
than
the
wanted
number
of
forwarders
and
duplicate,
which
you
should
well
in
this
stuff.
Work
isn't
course
to
be
too,
and
we
have
also
an
hour
both
there's
the
number
of
neighbors,
which
is
larger
than
n
duplicate,
because
value
constructing
the
num
is
selecting
the
number
of
for
virus.
D
You
will
see
that
well,
it's
not
fantastic
every
time
so
that
when
you
have
too
many
men
neighbors,
which
has
for
viruses
selected
within
a
certain
area,
you
can
reduce
it
and
also
then
the
network
configuration
changes,
so
you
can
select
other
ones
and
they
move
on
so
to
keep
this
reasonable.
Now
I
have
a
very
complex
picture.
Oh.
D
Yeah
this
is
the
next
one.
So
what
do
the
notes?
They
do?
They
sent
a
link
locomotive
cast
and
they
send
their
adverse.
There
says
I.
The
number
of
four
for
bar
the
neighbors
to
have
the
end
or
under
and
n,
are
both
deaf
and
their
own
state.
Of
course
she
were
format
is
used,
who
doesn't
and
only
messages
from
her
yeah
even
draft
so
I.
D
The
only
messages
from
fell
out,
neighbors
taken
into
account
and
only
message
message
reception.
You
updated
note
and
you
execute
the
selection
algorithm
expect
complex.
Oh
here's,
the
commissary,
here's,
the
complex
picture
I
wanted
to
show
that
this
NR
under
actually
gives
you
information
about
the
to
to
the
to
the
to
help
to
hop
neighbors
yeah.
So
I
have
here
to
sit.
Bones
were
selected
to
be
a
forwarder.
You
will
see
that
the
Wonder
in
the
middle.
D
He
sees
that
this
number
of
for
viruses
to
you
see
only
others
that
they
have
the
number
for
viruses
to,
and
then
you
have
all
these
months,
which
are
the
neighbors
which
have
to
enter
under
they
can
tell,
and
by
selecting
the
one
which
has
the
highest
and
are
under
for
all
your
neighbors.
You
may
consider
that
this
is
the
highest
number
of
notes.
You
are
helping
with
this
for
Ryder.
D
So
what
we
do
is
we
look
at
the
number:
try
to
select
death
note
of
all
the
neighbors
which
has
the
higher
numbers
of
Anna
rounder,
and
that
number
may
be
the
same
for
several
notes
into
town
break
ties.
We
use
the
network
address
so
choose
the
highest
one,
for
example,
to
have
only
one
note
of
all
the
neighbors
which
are
actually
going
to
be
for
murder.
Yeah,
you
don't
want
all
them
of
them
decide
simultaneously.
D
So
this
is
what
it
on
selection
algorithm.
So
every
note
orders
the
neighbors,
including
itself
on
the
N
or
unknown
number
value
and
the
address
so
that
you
have
this
and
what
do
you
do
is
select
at
the
first
one
and
if
that
you
and
become
a
forwarder
simple,
as
that,
you
also
look
at
NR
above
this
first
star
ignore
and
if
it's
equal
to
the
number
of
valid
neighbors.
D
You
know
that
all
neighbors
see
too
many
months,
and
you
know
that
if
you
are
and
forward
your
Islam
too
many
in
this
company
and
you
remove
yourself
yeah.
So
that's
how
first
there's
also
an
you
want
this
to
happen
in
a
stable
network.
So
you
don't
want
to
make
decisions
about
networks
which
are
constantly
changing
aspects.
D
So
what
the
cotillion
for
that
is
to
look
at
the
looking
at
all
the
neighbor
messages
and
if
the
NR
undervalue
is
changing,
if
the
end
are
under
value
changes,
you
don't
do
anything
once
that
you
see
the
deer
in
cycle.
All
the
message
are
the
same.
It's
the
same
and
a
runner
for
oldham.
You
say:
look
here:
I'm
stable
I
can
make
my
decision
yep
some
examples.
D
So
if
here
and
that
did
some
simulations
of
it,
the
distance
between
the
notes
are
about
10
meters,
it
isn't
too
hot
network,
so
the
ones
down
below
don't
see
the
ones
up
above,
but
the
month
in
the
middle
apparently
see
all
of
them.
You
want
to
have
in
the
past,
between
all
the
four
murders.
So
what
to
do
I
first
make
sure
that
there
is
long
for
water
and
a
forwarder
can
only
be
selected
when
he
sees
another
forward
so
which
issues
that
there
is
a
path
between
all
four
murders.
D
D
Each
note
is
marked
with
the
number
of
forward
as
it
sees
and
they
all
have
at
least
two,
where
you
see
when
the
number
of
for
where
this
is
five
or
six,
you
may
wonder
if
it
should
not
be
removed,
while
giving
the
algorithm
where
we
had.
You
see
that
actually
some
of
them
remove
themselves
yeah.
So
one
of
those
there,
the
life
blue,
one,
has
been
a
forward,
but
you
saw
that
the
density
was
too
high.
D
If
the
others
do
not
go
down,
it
means
that,
on
the
edge
at
least,
there
is
one
note
which
only
sees
two
4runners
and
you
can
do
not
remove
himself
yeah.
So
probably
when
you
look
at
the
evolution
of
the
Anne,
Arundel
etcetera
can
do
much
more
with.
So
all
the
data
you
get
in
the
algorithm
and
do
some
refinements
I
didn't
do
them
I've
here
another
example
also,
so
it
is
a
very
dense
network,
it's
not
too
bad.
D
G
I'm
I'm,
Charlie,
perkinson
I
think
this
is
a
great
idea
was
involved
a
lot
with
the
forwarding
specification
in
manet
and
there's
been
a
huge
amount
of
work
on
this
area.
I
mean
I.
One
of
the
terms
that's
used
is
connected
dominating
set
when
you're
trying
to
construct
one
so
I
but
I
guess
to
have
a
couple
of
questions.
I
mean.
First
of
all,
do
you
really
think
that
it's
typically
important
to
have
more
than
one
forwarder.
D
Yes,
I
mean
that's
if
you
want
at
least
one
hundred
percent
availability
over
a
year
in
this
room
and
not
when
you
switch
it
off,
you
say
all
right,
good
quarter
or
for
now
and
try
again.
Yes,
you
need
at
least
in
one
hop
networks,
at
least
one
and
in
larger
network.
We
have
multiple
hops.
You
need
as
many
forward
ssink
yeah
you.
G
D
D
Quite
agree:
if
there
are
other
misleading
sings,
I
mean
may
choose
your
network
addresses
such
that
the
algorithm
is
improved.
I
mean
there
are
lots
of
things
you
can
do,
but
yeah.
What
I
also
wanted
to
do
is
look
if
this,
how
far
is
from
the
optimal,
so
how
many
forwarders
you
look
for
the
optimal
case?
I
didn't
do
all
these
things.
I
mean
I.
Just
was
happy
with
this
for
the
moment,
and
so
it's.
G
D
A
very
good
question:
what
is
stable
I
mean
this
table
is
with
respect
to
the
number
of
times
you
sent
your
you
sent
your
message
out.
So
if
you
sense
out
your
maple,
let's
just
like
they
did
every
five
minutes
and
you
have
about
twenty
neighbors,
then
stable
is
longer
than
20
times.
Five
minutes.
Yeah.
E
E
Okay,
that's
that's
great
to
understand
and
I.
Think
you
said,
I
was
going
to
ask
something
about
about
about
your
definition
of
stable
as
well,
and
it
sounds
like
you
couldn't,
so
you
can
basically,
if
the
if
the
instability
of
the
of
the
network
or
if
it's
not
fixed,
orbits
moving.
If
it's
slow
enough
compared
to
your
your
transmission
time,
then
it
effectively
stable
yeah.
So
you.
E
Well,
so
my
other
I
haven't
read
your
documented.
Also,
I
really
have
no
idea
so
is,
is
your
does
your?
Does
your
protocol
live
on
top
of
trickle
on
trick
on
trickle?
So
it's
a
it's
a
it's
a
independent
of
the
the
nipple
forwarding
itself,
okay,
so
that
what
I
was
curious
about
is
that
it
that
I
I
know
little
about
this,
but
they're
the
people
that
are
trying
to
do
beer
to
rip
to
do
multicast.
E
D
F
Piscataway
I
love
this.
So
if
you
you're
looking
for
voices
to
say
yes
to
continue,
please
continue
this
work.
Okay,
it
looks
to
me
that
it's
an
interesting
variation
around
and
something
very
similar
acronym,
which
is
NPR.
The
multi
punch,
realize
you
know,
and
I
saw-
and
I
was
just
wondering
it
seems
to
me
that
you're
looking
for
an
optimum
for
certain
radio
condition,
you
know
the
radio
conditions
change
a
lot
there's.
F
D
Idea,
actually,
it's
fits
with
my
mom
of
my
other
sorts,
because
when
we,
when
I
see
the
stable
solutions,
if
you
move
around
the
forward
a
little
bit
to
another
location,
the
whole
pattern
may
look
differently
and
you
may
remove
of
them.
So
I
was
thinking
of
actually
an
implementation
you
may
improve
by
moving
them
around
and
so
moving
them
around
would
be
a
good
idea,
I
think,
actually
in
general,
yeah.