►
From YouTube: IETF99-ROLL-20170720-1330
Description
ROLL meeting session at IETF99
2017/07/20 1330
https://datatracker.ietf.org/meeting/99/proceedings/
A
A
Thank
you
to
my
notice
and
describe
again.
Okay.
This
is
our
agenda.
We
are
going
to
have
people
information,
news
cases,
then
music
as
PR,
no
part
ow,
everyday
ripple
and
first
election
and
the
prediction,
so
any
variety
comments:
okay,
okay,
these
are,
these
are
our
milestones.
Okay,
we
have
a
lot
of
red.
A
B
You
could
come
over
here.
Well,
that
might
be
a
part.
We
have
space
free
here
lots
of
space,
so
I'm
just
curious.
Are
the
group
in
the
room
how
many
have
actually
slog
through
the
entire
document,
all
24
cases?
Dominique?
Oh
yeah?
Okay!
So
alright,
that's
good!
Thank
you,
yeah!
Actually,
we
we
have
actually
literally
changed
all
the
bits
in
the
wire
I.
B
Turned
this
thing
yet
so
so
that's
what
I'm
here
to
talk
about
about
that
that
so
it's
it's
quite
incredible
to
me
that
we're
actually
still
talking
about
this
document
and
that
well
I,
don't
know
how
your
vision!
So
we
need
a
bunch
of
revisions
before
got
adopted
to
I
thought.
It
was
gonna,
be
pretty
quick
and
simple.
Once
we
figured
out
what
was
going
to
be
there,
does
it
work.
B
Okay,
so
Mike
heard
had
a
number
of
comments
over
quite
a
number
of
months
and,
as
it
came
back
from
various
reviews,
we
decided
to
take
him
a
little
to
decide
that
while
we
agreed
it
with
his
comments
when
he
first
made
them,
we
weren't
sure
that
we
could
do
something
so
as
radical
as
he
suggested.
And
so
is
he
here.
Mike
heard
I,
don't
think
so
he's
not
a
frequent
person,
contributor
and
so
based
upon
pardon.
D
B
B
B
B
So
we
would
essentially
be
building
a
layer
to
transport
between
IP
and
IP
and-
and
that
was
just
so
far
away
from
what
most
people
had
imagined.
We
would
do
that
that
just
seemed
ridiculous
all
right.
So
then,
there
were
changes
to
24
60
bits,
which
is
now
standard
86,
which
suggested
in
the
initial
discussions
up
to
I,
would
say
last
last
year
in
Seoul
that
essentially,
these
bits
that
you
see,
which
are
part
of
the
hop
by
hop
type
definition,
would
be
essentially
obsoleted.
B
There
was
a
suggestion
even
that
we
would
just
merge
it
all
into
one
big
8-bit
wide
field,
and
that
none
of
this,
how
would
have
any
meaning
at
which
point
we
could
say,
hey,
that's
really
the
behavior
we
wanted
fundamentally
250
65-53
had
picked
a
1
discard
the
packet
when
maybe
this
would
have
been
a
better
choice.
The
reason
it
did
that
was
because
what
we
didn't
want
is
we
did
not
want
the
possibility
that
ripple
packets
would
escape
the
mesh
somehow
and
wander
somewhere
else
and
be
accepted
and
transmitted
right.
B
So
it
was
a
kind
of
a
firewalling
thing,
which
is
great
until
you
realized
that
part
of
the
whole
point
of
having
a
sensor
network
is
to
tell
someone
about
it
right,
and
so
it
would
be
nice
that
the
packets
could
actually
get
places
where
they
needed
to
be
told
about
right.
So
that's
the
kind
of
kind
of
constraint.
In
the
one
hand,
you
want
to
build
a
closed
network
that
you
know
only.
You
only
works
where
you
want
it
to.
B
On
the
other
hand,
you
don't
want
to
have
to
screw
around
with
equipment
to
make
it
work.
So
basically,
what
we've
said
is
hey.
We've
made
a
mistake:
let's,
let's
simplify
the
light.the,
the
processing
by
making
it
clear
that
we're
gonna
use
a
type
code
which
says:
if
you
do
not
understand
this
hop-by-hop
header,
you
shall
ignore
it
and
keep
and
just
keep
processing.
Okay.
Now
there
are
some
text
in
standards
86.
B
B
Well,
our
curt
instructions.
Is,
you
add
an
IP
IP
header,
you
add
the
our
h3
header,
but
if
you're
going
to
do
that,
you
know
two
things:
okay,
one!
You
know
the
exact
path
to
the
last
ripple
router,
so
you
can
actually
address
the
packet
to
that
router.
That
router
will
remove
that
header,
and
so,
even
if
the
packet
turns
out
to
be
address
to
a
leaf
node
which
doesn't
speak
ripple,
then
that's
okay.
You
know
explicitly
what's
going
on.
B
We
actually
could,
at
this
point,
specify
in
a
standard
track
action
that
we're
going
to
insert
headers
and
we
might
be
able
to
get
away
with
it,
but
we're
not
proposing
that
here
right
so
that
that's
the
compromise
that
I
heard
from
certain
area
directors
that
a
standards
action
could
actually
do
this,
and
and
maybe
that
that
would
be
a
better
idea,
but
not
proposing
that
at
this
point,
okay
go
ahead.
Pascal.
E
B
That's
why
I'm
not
really
suggesting
that
we
do
that,
because
we
did
all
that
work.
That's
what
initiated
this
whole
six
six
lower
H
work.
We
started
that
in
2013
in
Toronto.
Okay,
four
years
ago,
for
three
years
ago,
2013
in
Toronto,
2013
I
know
because
I
drove
my
van
I've
had
my
van
three
years,
those
three
years
ago,
okay
14
in
Toronto.
That's
when
we
started
we've
completed
that
work
right.
We
now
have
even
better
compression
than
we
had
before.
B
As
a
result
of
this
challenge,
we
have
an
improvement
to
the
situation
that
we
might
not
otherwise
have
done.
So
that's
kind
of
a
good
thing
right,
that's
right
exactly
so
one
of
the
things
that
we
had
discussed
six
months
ago,
we
had
discussed
making
this
change
and
we
discussed
making
this
change
specifically
doing
it
prior
to
publishing
six
low
RH
such
that
six
out
lower
H
when
it
decompressed.
B
Essentially,
if
you're
using
the
new
compression
scheme,
then
you
should
be
using
the
new,
the
new
value
and
since
the
compression
scheme
compresses
this
value,
you
know
we're
gonna
have
to
figure
out
whether
we're
updating
I
think
we
have
to
write
an
update
to
six
lower
H
in
this
document.
That
says
that
when
you
decompress
your
decompressing
with
byte
23,
not
63,
and
it's
a
change
to
the
protocol.
B
Okay,
but
if
you're
going
from
the
stock
compression
to
the
to
the
new
compression,
that's
a
that's
a
flag
day
on
the
wire
to
begin
with,
and
so
what
I
hoping
is
that
if
we're
gonna
make
this
type
change
that
we
could
roll,
this
is
the
same
flag
day.
So,
basically
you
can't
do
this.
You
can't
change
this
thing
and
interoperate
with
old
nodes.
That's
the
first
first
assumption.
How
do
people
feel
about
that
before
I
continue.
B
Right,
I'm
gonna
deploy
new
code.
You
go
to
upgrade
every
node
in
their
network
or
not
right.
Okay,
so
that's
a
possibility,
but
right
we
could
do
something
about
the
flag
day,
all
right.
So
what
we
could
do
is
we
could
keep
using
the
old
code
until
we're
sure
that
every
device
in
the
network
speaks
the
new
code
right
and
so
essentially
what
we
could
do
is
we
could
observe
what
who
supports
what
and
if
everyone
supports
the
new,
the
new
code,
then
we
could
announce.
B
B
So
essentially
that
allows
you
to
incrementally
deploy
new
firmware,
knowing
they're
going
to
continue
using
6:3
and
they'll
be
compatible
with
everything
else
until
you're
sure
that
you've
deployed
everything
at
which
point
you
toggle
a
bit
in
the
in
the
dough
tag
route.
It
sends
out
this
this
new
day,
I/o
option
and
everyone
switches
to
the
new
code
and
everyone's
happy
and
then
well.
At
that
point
you
know
you
can
now
hook
up
leaf
nodes
that
don't
speak
ripple
well,.
E
B
A
good
suggestion:
yes,
that's
a
good
suggestion.
We
could
do
that
instead,
I,
don't
think
I'm
too
picky
about
how
we
do
it,
but
you
know
we
just
we
had.
We
had
redesigned
it.
You're
right
we
should.
We
should
have
done
that.
That
would
have
been
a
smarter
thing
to
do
as
mice
thought,
that's
the
new
ripple
DAO
option,
but
we
could
put
a
configuration
option.
I'll
take
your
point.
So
basically
you
have.
We
have
a
small
one,
byte
DAO
option
flag
and
that's
a
we
defined
right.
B
B
If
you're
building
a
new
network
where
you
expect
to
have
leaf
nodes
or
expect
to
speak
to
the
Internet,
well,
then
you'd
better
use
the
new
code
because
it
will
be
ignored
at
which
point
you're,
building
a
new
network
and
you
probably
don't
have
a
flag
day
anyway.
So
maybe
you
don't
need
a
configuration
option
right
so
I
don't
know
any
opinions
about
what
what
should
we
bother?
It
seems
like
a
lot
of
effort
for
a
fairly
small
use
case
to
me.
B
E
Get
again
the
one
thing
that
troubles
me:
yes,
it
seems
to
me
that
my
device
reboots
it
will
come
back
to
the
63
mode
unless
it
sees
the
option
again,
which
tells
me
that
we
have
to
sound
deception
all
the
time,
which
is
not
what
I
want
to
achieve
so
hah.
That's
a
piece
I,
don't
like
right!
If,
if
the
default
was,
if
I
don't
see
the
option
that
I'm
23,
that
would
make
me
happy,
but
that
doesn't
serve
what
you
want
now,
that's
what
the
toilet
so.
E
B
B
F
Other
from
worried
with
regards
to
the
restart
problem,
you
know
there
is
that
other
other
state
information
like
like,
for
example,
the
DTS
M,
which
is
lost
when
the
node
is
restarted.
So
it
is
expected
that
a
persistent
storage
takes
care
of.
It
is
implicitly
expector
I.
So
if
a
DTS
L
is
lost,
essentially
none
of
the
new
dows
from
that
node
after
restarted
is
going
to
take
into
effect.
So
a
persistent
storage
is
expected.
Anyways
right
so
I
mean
this
can
go
along
as
well.
G
B
Yeah,
so
so
yeah
you
would
have
to
as
I
said.
That's
aha!
If
you
rebooted
cold,
you
would
have
to
listen
for
a
whole
bunch
of
stuff.
Before
you
could
you
could
operate
and
or
you'd
have
to
store
it.
So
I
think
that
would
be
reasonable,
but
so
can
can
I
get
a.
How
many
are
in
favor
of
this
option?
Maybe
you
could
just
hum
if
you
like.
Happy
would
like
to
have
this
anti-flag
day
option.
B
B
B
B
I
I
B
Create
routing
pathologies
within
the
network
I'll,
let
you
answer
to
in
a
moment.
So
what
it
does
is,
if
you
had
some
situation
where,
for
some
reason,
the
the
machines
in
the
the
nodes
in
the
DOE
dag
were
also
connected
for
other
ways.
They
had
an
Ethernet
or
something,
and
they
suddenly
started
spraying
packets
at
the
wrong
interface
that
those
packets
would
get
discarded
because
they're
not
they
contain
a
hop-by-hop
better,
the
better
that
that
other
nodes
are
not
supposed
to
understand.
Okay,
so
that
was
a
sort
of
it
was
a
firewall.
B
B
I
E
Need
to
read
this
paper,
but
I
won't
still
answer
your
question
because
some
table
so
everything
that
Michael
said
was
correct.
Obviously
this
is
Pascal.
The
piece
that
you
didn't
say
is
what
happens
inside
the
ripple
Network.
If
somebody
does
not
understand
how
my
other
shoe
right
right,
which
is
basically
that
he
may
inject
the
e
start
multiple
instances,
you
may
inject
the
the
packet
in
the
wrong
topology
and
then
the
next
hop
may
inject
it
back
in
the
right
topology.
E
J
B
They
didn't
put
the
rpi
header
in
at
all,
but
it
kind
of
worked,
okay
and
and
they
deployed
sensors
and
those
sensors
talked
to
the
internet
because
there
was
no
header
to
remove
and
everyone
was
happy
and
then
I
also
believe
that
some
of
the
original
authors
of
65-53
had
the
notion
that
you
could
delete
headers
in
the
middle
of
the
packet,
and
they
never
quite
expressed
that
because
they,
no
one
had
told
them.
It
wasn't
true,
so
so
that
never
occurred
them
to
ask,
because
we
did
it
all
the
time
in
v4.
B
So
they
resigned
this
thing
and
said:
oh
well,
I
just
assumed
well,
we'll
just
delete
it.
If
it's
a
problem,
then,
if
the
route
we'll
just
delete
it-
and
everyone
will
be
happy
and
of
course,
that
never
properly
got
vetted-
that
now
was
never
expressed
as
a.
This
is
what
you
do,
okay,
and
actually
that's.
Why
we're
here
in
the
first
place,
because
this
document
was
created
because
of
realization
that
people
didn't
know
when
to
put
our
pies
and
when
to
put
arch
threes
in
and
when?
B
Specifically,
you
needed
to
put
an
IP,
IP
header
in
to
add
or
remove
them,
and
so
that's
what
this
doc.
And
why
is
this
document
as
24
cases
of
flows
between
nodes
that
are
either
aware
or
unaware
of
of
the
extensions
and
in
order
to
accommodate
them,
you
have
to
put
IP
header
IP,
IP
headers
in
appropriately
addressed,
so
that
they
can
be
removed
at
the
right
place.
Okay,.
E
E
And
part
of
the
consequences
that
we
found
in
the
earlier
contiki
of
not
having
it
is
also
that
somewhat
breakages,
we
are
not
discovered
of
after
a
very
very
long
time.
There's
also
two
side
effects
you're
implementing
on
your
portion
of
the
protocol.
If
you
don't
implement
it
and
that
there's
all
sorts
of
consequences,
humourist
implemented.
B
So
so,
for
instance,
you
know
my
my
linux.
Implementation,
which
does
not
speak,
does
not
have
an
RPI
processor,
which
is
a
bug,
has
inter
operated
with
kontiki
for
years,
just
fine
until
they
put
the
RPI
header
in
with
a
63
and
then
and
then
it
breaks
because
we
dropped
the
packet
right
and
it
was
very
mysterious,
so
I
said:
wait,
wait,
you're
running
the
new
version
of
kontiki,
which
has
RPI
in
it.
Oh,
yes,
yeah,
I'm!
Sorry,
it
won't
work
right
so
important.
B
So,
okay,
we
have
some.
We
have
some
tables
here.
If
you
really
want
to
talk
about
the
different
cases.
So
one
of
the
suggestions
was
to
make
our
document
shorter
by
removing,
by
by
leading
a
lot
of
the
cases
and
I'm
not
crazy
about
that.
I
think,
partly
because
there's
a
lot
of
work
to
write
them
all
out,
but
but
but
because,
because
some
of
them
are
a
little
bit
subtle
and
and
the
the
cases
differ
by
a
very
small
amount.
B
D
A
K
So
we
presented
this
work
in
ATF
91
and
in
96
in
a
little
more
detail
and
and
I'm
going
to
run
through
this
relatively
quickly.
But
of
course
you
have
the
original.
Did
you
ask
questions
so
please
use
that.
So,
basically,
multicast
doesn't
work
in
the
networks
that
we
are
talking
about.
So
we
are
done.
We
can
go
to
the
next
slot.
Well,
sometimes
you
want
to
make
them
work
anyway,
and
there
are
situations
where
you
actually
want
to
use
multicast
and
so
far
we
have
had
a
number
of
ways
of
doing
that.
K
One
is
in
ripple.
One
is
the
separate
protocol
name
which
is
not
really
about
multicast.
If
you
look
closely,
but
we
can
ignore
that
for
the
moment
and
the
existing
version
of
multicast
for
ripple
is
for
storing
mode
which
doesn't
exist,
even
though
Pascal
sometimes
seems
to
believe
it
does
right
exactly
so.
The
question
was:
how
do
we
make
it
work
for
non
storing
mode,
so
the
the
River
multicast
uses
multicast
daos
to
announce
the
listeners.
K
K
It's
I
think
it's
pretty
much
still
untested.
So
how
do
we
do
with
this
phone
known,
storing,
mouch
and
the
problem
really
is
to
do
the
the
classical
way
of
doing
Martha
cost
every
node.
That
is
forwarding
needs
to
know
that
there
are
members
in
the
tree
below
it,
but
we
don't
want
to
have
all
that
stage
in
the
non
story
mode.
So
how
do
we
get
the
route
to
actually
Express
all
those
decisions
in
the
packet
that
it
sends
down?
K
So
essentially,
we
have
to
then
two
source
routing
format
and
the
obvious
solution
is
to
list
all
outgoing
interface,
let's
say
every
packet,
so
when
something
is
sent
down
here.
These
two
forwarders
here
may
have
no
idea
about
the
multicast
group,
or
there
may
be
members
whatever.
They
need
to
know
that.
There's
there's
a
listener
down
here
that
actually
needs
that
back.
So
how
do
we
make
it
find
out?
So
just
embedding
an
outgoing
interface
list
like
we
do
for
unicast
doesn't
really
work
very
well,
I
mean
even
with
the
compression
that's
non-starter.
K
Beer
has
been
around
for
a
while
and
that's
why
we
presented
our
web
address
that
hadn't
previously
at
ITF
91,
because
we
had
a
buff
at
the
time
is
based
on
lumbering
routers
and
then
based
on
that
numbering,
you
can
put
a
bitmap
into
the
package
that
tells
you
which
interfaces
should
be
forwarding
and
what
shunt
I
started.
I.
E
Hoped
you
would
say
the
last
world
before,
but
now
I'm
here
so
Pascal?
Well,
just
the
point
is
there
is
a
another
beer
which
is
called
beer
traffic
engineering,
VRT
I,
don't
think
it's
workgroup
document
at
beer,
but
it's
presented
there
and
it's
designed
to
traffic
engineering
to
lay
out
a
path
exactly
like
you're
doing
so.
Actually
your
clothes
are,
they
are
reality
and
then
you
seem
to
imply.
There
is
really
this
concept
that
vo
it's.
It's
really
a
good
concept
and
people
appreciate
it.
Yeah.
K
I'm
not
trying
to
belittle
B
it's
just
not
exactly
the
problem
that
the
solution
that
that
we
might
want
to
use-
or
maybe
we
do
so
most
people
who
actually
have
worked
in
the
medical
space
know
that
that
bloom
filters
have
been
given
as
a
solution
to
a
number
of
problems
in
this
space.
So
it
seems
kind
of
natural
for
us
to
use
bloom
filters
and
and
see
how
far
we
get
so
bloom
filters
is
a
really
old
technology.
K
So,
by
putting
a
bloom
filter
in
the
in
at
the
root,
each
of
the
potential
forwarding
nodes
can
check
their
outgoing
interface
address
against
this
bloom
filter,
and
only
if
there
are
bits
for
in
this
case,
both
hash
functions.
Hashing,
the
outgoing
interface
number,
the
packet
is
actually
forward
so
problem
solved
and
the
very
problem.
Of
course,
this
bloom
filters
are
probabilistic,
so
we
do
get
false
positives
and
yeah.
We
have
to
find
out
whether
that
is
acceptable.
It's
not
breaking
anything,
but
it
requires
more
energy.
It
requires
more
spectrum
usage.
K
Then,
if
the
forwarding
decision
is
exact
and
Pascal
will
tell
us
how
to
do
that,
so
yeah
this
guy
here
much
get
a
false
match.
For
instance,
if
by
accident
the
hash
functions
just
hit
these
two
bits
here
and
yeah,
then
it
starts
forwarding
stuff
and
getting
hot,
even
though
it
don't
have
to
okay,
how
bad
are
those
they
waste
energy?
We
did
some
numbers
on
this
and
and
in
general,
it's
not
too
bad.
It's
in
particular,
it's
very
bearable.
If
you
have
a
relatively
large
network
and
I
have
small
multicast
groups
within
that
network.
K
I
Excuse
me
again
a
stupid
question:
if
that's
okay,
can
you
come
back
to
the
previous
slide?
Please
you're
saying
it
doesn't
cause
any
semantic
damage.
So
what
happens
if,
due
to
a
bloom
filter
collision,
you
send
out
a
package
through
the
interface
that
came
in
with
you've
got
a
routing
loop
right.
You.
J
E
Path
yet
you
bail,
so
my
customary
told
us
two
things.
He
told
us
that
you
could
basically
use
bits,
two
to
two
indexed
to
shore
path,
to
trace
a
path,
and
he
showed
us
that
you
could
compress
that
information
using
bloom
filters
and
those
are
two
things
which
are
very,
very
good,
they're,
important
and
also
very
often
at
all.
So
keep
that
in
mind
when
I'd
be
presenting
the
other
stuff,
because
bloom
filters
could
be
used
as
well.she
in
my
solution,
Orbitz
could
be
used
as
well.
Here
the
solution
comes
later.
E
E
You
don't
take
things
for
your
Deponia
straight
when
there
is
a
good
news
and
bad
news.
The
good
news
is
there
are
all
those
pages
on
the
web
where
you,
just
until
the
size
of
your
network
default
network,
but
what
you're
looking
for
and
it
will
tell
you
I-
mean
Yash
functions
with
its
mapping.
So
there
is
it's
all
well
known
and
optimized.
You
can
decide
exactly
what
is
the
optimal
bitmap
come
to
match?
E
You
know
daily
energy
you're
ready
to
waste,
to
the
false
positive
rights
versus
the
size
of
the
number
of
bits
that
you're
willing
to
to
to
use.
So
that's
the
good
news.
The
bad
news
is
sure
you
want
to
optimize
it.
You
need
to
know
what
your
network
is,
and
that
will
be
true
for
the
bits
as
well,
because
the
optimum
depends
on
the
number
of
things
that
you
have
to
address.
Yeah.
K
So
we're
done
with
that
and
of
course
not.
Nobody
should
be
implementing
this
uncompressed
thing,
so
we
have
a
pretty
efficient
way
to
represent
this.
Now,
where
are
we
with
implementations?
We
did
this
work
originally
a
couple
of
years
ago,
with
Kentucky
at
the
time,
Kentucky
didn't
have
an
on
storing
mode.
So
we
just
hacked
one
in
and
one
of
the
nice
observations
is
that
the
forwarders
can
compute
those
hash
functions
once
because
they're
their
outgoing
interface
address
is
reasonably
stable.
So
they
do.
We
have
to
do
relatively
little
computation.
K
It's
really
just
a
bit
matching
operation
that
you
do
there
now,
of
course,
if
you
have
multi-page
functions
and
maybe
multiple
bitmap
sizes,
then
it's
a
little
bit
more
work
and
a
little
bit
more.
You
have
to
stall,
but
it's
it's
only
a
very
limited
number
of
its
interface,
and
the
other
thing
we
noticed
is
that
it's
really
easy
to
add
a
member
to
a
group
in
the
route,
but
you
actually
have
to
recompute
the
bloom
filter
for
every
leaf.
K
That's
not
a
disaster,
but
it's
a
little
bit
more
work,
because
it's
linear
to
the
size
I
mean
unless
you
start
building
binary
trees.
For
this
you
can
do
that,
but
it's
probably
not
in
the
organization.
You
don't
need,
ok,
so
this
is
the
implementation
we
did
just
to
get
the
numbers.
It's
not
an
implementation
that
you
would
want
to
work
with,
and
the
news
is.
There
is
an
industrial-strength
implementation.
A
E
Okay,
so
this
is
Pascal
Chima,
so
keep
in
mind
like
I,
said
that
very
to
angle,
whether
you
want
to
do
sauce
watching,
which
is
pretty
much
what
custom
has
been
showing
us,
whether
you
want
to
do
storing
mode
state,
which
is
what
I
will
be
showing
you
now.
That's
one
thing,
and
the
other
thing
is
whether
you
want
to
represent
things
as
individual
bits,
which
is
what
I
will
show
you
here,
just
to
show
you
that
it
can
be
done
and
it
would
work
in
the
sauce
ratings
world
as
well.
E
That
would
be
beauty,
but
what
I'm
showing
here
could
be
easily
translated
in
bloom
filters,
as
well
for
the
exact
same
reasons,
and
you
would
have
the
exempt
pros
cons
that
that
we
can
see
with
blooms
using
vs.
bits
right,
I
mean
basically
bits
will
have
a
fixed
size.
Blooms
gives
you
elasticity,
but
then,
if
you
renew
the
Asch
function,
though,
it
must
be
a
protocol
for
doing
that,
synchronizing
that
there
are
all
sorts
of
pros
and
cons
at
some
point.
E
We
need
to
put
all
that
in
the
table
and
and
in
the
one
hand,
which
may
decide
to
work
on
sorting
or
storing
both
the
one
hand.
On
the
other
hand,
we
may
decide
to
use
bits
problem
features
right.
It's
so
I
will
show
you
with
the
bits,
but
I'm
sure
that
you
will
figure
how
it
would
work
with
blooms
as
well.
The
cool
thing
with
known,
for
instance,
is
you
can
also
configure
your
bits.
E
If
you
know
the
Asch
functions
with
the
bits
you
have
to
distribute
them,
that's
sort
of
the
things
that
you
have
to
ponder.
So
there
is
a
topology,
so
I
spent
some
time
drawing
them
all,
and
here
is
like
a
set
of
geo
tags
that
will
form
out
of
my
topology.
That's
what
people
will
be
for
you,
and
out
of
that,
you
can
build
a
preferred
tea,
for
instance,
for
for
multicast.
That's
what
Reba
also
does
for
you.
So
the
way
ripple
would
today
distribute
multicast.
E
That's
pretty
much
very
close
to
what
we
do
for
unicast,
but
you
can
do
multiple
copies
and
with
the
casco
addressing
if
we
are
using
storing
mode,
then,
basically,
there
is
the
offset
of
address
that
are
installed
on
those
nodes
and
when
you
do
the
the
non
storing
Dao,
basically
F
would
they'll
be
I'm.
The
child
I
have
put
the
root
and
I'm
a
child
of
B
and,
if
you're,
storing
mode
well,
as
you
know,
the
order
that
will
be
reflect
in
the
routing
table
in
the
individual
nodes,
as
you
follow
this
this
graph,
okay.
E
So
in
the
case
of
non
storing
which
I'm
taking
as
a
base
here,
it's
basically
what
you
end
up
with
in
the
root
you
end
up
with
this
table,
which
is
based
on
those
those
down
the
sponsoring
Dao,
which
reflect
this
child.
To
parent
yaki-
and
you
end
up
with
this
table,
that's
pretty
much
what
the
Taoist
I've
told
you
and
the
the
goal.
E
E
I
don't
come
back
to
the
operation
because
justin
has
said
it,
but
it
boils
down
to
every
destination
must
have
a
bit
somewhere
in
a
bitmap
and
by
putting
a
bitmap
in
the
back
edge
with
a
number
of
number
of
bits
set,
I'm
saying,
which
particular
devices
I
want
this
packet
to
be
sent
to
supposed
to
classical
multicast.
That
gives
you
a
good
control
of
who
exactly?
Is
your
destination
real
much
gas
route?
Classical?
E
You
don't
know
if
there
is
one
guy,
nobody
or
hundreds
of
people
in
there
with
B,
and
you
have
a
very
good
control
of
how
many
targets
in
which
they
are
so
that's,
basically
yeah.
So
first
allocation
policy
can
be
static.
You
can't
figure
the
bit
in
every
device,
that's
what
the
beer
guys
do.
Second,
is,
since
we
have
this,
this
big
do
diagonal
single
prefix.
E
You
can
carry
a
current
bitmap
of
allocated
bit
or
you
can
carry
a
blue
filter
of
the
current
allocated,
better
whatever
to
compress
it,
and
based
on
that,
you
can
have
a
relative
good
chance
for
a
newcomer
to
take
a
bit
which
is
not
already
located
and
then,
since
we
have
died
anyway
for
the
IP
address
of
this
new
device,
we
can
also
put
the
bit
that
you're
selected
and
the
route
can
say
yes,
no
option
to
the
6lb
are
upon
the
that
process
can
return.
Here.
E
Is
the
bit
I'm
giving
you
I
accept
your
address
in
my
network
that
the
attack
process
in
six
level
nd
on
the
way
back
here,
is
the
bit
that
you
get
for
this.
So
you
used
a
6
lb
anatomy
as
the
white
board
of
all
the
ipv6
addresses
in
your
network,
but
also
as
the
the
bit
on
deliverer
of
the
bit.
There
is
a
variation
to
that,
which
is
that,
with
the
60
security,
the
the
registrar
is
called.
The
join
server
now
is
also
giving
you
back.
E
B
Richardson,
so
I
was
really
I
was
just
reading.
This
and
I
was
like
I
really
like
option.
One
I
just
Sims
makes
sense,
and
then
I
was
as
thinking
the
six
lbr
assigns
the
bits.
Is
there
optimizations
that
he
can
do?
That
makes
everything
better
because
he
makes
things
work
better
because
he
gets
to
assign
the
bits
exactly.
E
And
I
will
show
you
on
the
next
slide.
What
variation
of
that
for
the
first
survey
that
you
will
give
you
is
the
same
exact.
That's
for
that
right.
It
will
guard
see
that
everything
is
unique.
That's
that's
whether
it
doesn't
need
to
be
a
dad
operation.
That's
the
best
service
that
to
give
you
and
by
the
way
the
bit
is
unique
to
a
do
that
which
means
that
it's
kind
of
correct
that
the
6
of
ER
owns
the
bits.
E
If
you
move
too,
your
daddy
need
a
new
bit
and
some
some
variation
of
that
the
table
won't
spend
too
much
is
if
the
security
system
earns
you
the
shorter
tries,
it
can
also
enter
your
beat
and
the
beat
can
even
be
encoded
as
the
rightmost
part
of
the
shot
address
and
the
rest
could
be
a
group
because
at
some
point
you
don't
want
so
many
bits.
You
have
to
submit
what
you're
doing
so
yeah
those
bits
are
addresses,
and
this
bit
is
the
group.
Well,
the
bitmap
belongs
to
you
already.
E
Yes,
that
there's
another
optimization
is
to
say
I'm
using
two
bits.
If
I
have
two
bit,
zero
is
I'm,
not
a
target,
and
then
one
two
three
is
three
different
addresses
that
you
have
on
my
system.
So
there
are
all
sorts
of
thing
you
can
play
with
this
bit
and
you
can
do
that
with
a
bloom
as
well
won't
be
much
different.
So
so,
basically,
we
end
up
with
the
system
where
all
those
devices
of
the
home
bit
right.
E
That's
that
all
this
allocation
policy
and
stuff
is
to
get
us
there,
and
this
is
where
we
are
classically-
starts:
we'd
get
a
bit
assigned
to
every
device
and
the
route
knows
right.
The
route
has
now
this
mapping
Oh
a
is
benign
these
bit
11,
etc.
So
what
we'll
be
doing
with
those
bits
is
to
basically
the
key
goal
that
we
have
sure
is
probably
to
enable
a
storing
mode
in
the
devices
to
shortcut
things
etc,
which
costs
a
lot
less
than
the
classical
storing
mode
in
ripple.
E
Now,
if
you
represent
all
your
children
as
a
bit
in
the
same
bitmap,
the
amount
of
storage
that
you
need
in
the
individual
device
becomes
how
many
children
do
I
have
directly
connected
to
me,
which
is
in
the
order
of
my
neighbor
cash,
the
order
of
their
beliefs.
So
that's
three,
that's
three.
What
you
are
getting
is
the
big
benefit
of
the
operations
I'm
talking
about
not
depending
on
the
whole
number
of
lease
just
number
of
immediate
children.
That's
the
amount
of
storage
that
I
know.
E
E
You
will
set
your
own
bit
and
past
at
all,
so
the
amount
of
information
that
goes
up
is
always
the
same:
bitmap
just
more
and
more
bits
get
aggregated
inside
this
bitmap.
That's
how
you
save
the
memory
and
that's
how
the
storing
mode
the
size
of
the
amount
of
memory
in
storing
mode
becomes,
the
the
order
of
magnitude
of
your
neighbor
cash.
E
E
As
for
each
of
these
children,
the
bitmap
of
what
the
children
has
told
him
and
it
will
report
your
your
operation
of
all
this
and
where
it
gets
kind
of
cool,
is
when
they
have
to
sound
message
from
from
the
root
to
anybody
actually
from
anybody
to
anybody,
but
when
they
have
to
send
a
message
to
all
my
geoduck,
so
that's
example:
I'm
taking
I
just
have
to
set
the
bits
which
correspond
to
destination.
I
can
use
that
for
unicast
just
set
one
bit.
E
You
can
use
that
for
multicast
I'm
setting
more
bits,
and-
and
this
thing
will
now
go
down
to
the
repo
deck
and
that's
compatible
with
you
know,
do
directories
whatever
you
like
to
build
just
see
the
same
bit
coming
from
two
different,
two
different
zero
and
one
that
that's
because
you're
building
it
your
DAC,
you
can
use
either.
So
that's
basically
how
it
goes
so
in
order
to
forward
what
you
do.
Is
you
take
your
the
bitmap
of
your
child?
You
take
the
packet
you're
on
the
whole
thing.
E
If
it's
zero,
you
don't
pass
it
to
that
children
and
then
it's
a
series
of
unique
eyes
to
all
the
children
for
which
there
is
at
least
one
bit
that
matches
kind
of
a
very
simple
for
operation:
okay
and
well,
this
pretty
much
what
they
want.
So
I
give
example
of
that,
but
I'm
sure
you've,
you've
got
you've,
got
the
story,
I'm,
not
sure
we
have
more
time
so
I'd
rather
take
questions.
E
K
E
E
K
They're
better
coding
ways,
but
anyway
they
should
talk
to
the
net
recording,
and
so
this
is
the
order
of
magnitude
a
little
bit
just
to
give
this
new
idea.
Yes,
the
the
bloom
filter
thing
has
two
changes
here:
one
is
the
things
that
are
listed
in
the
bloom
filter.
Tell
you
not
end
devices.
It's
forward
us
in
your
cases.
K
E
E
What
you
described
and
ripple
now
we
can
avoid
tools
as
well,
but
just
by
itself,
this
thing
stands
pretty
much
and,
and
one
does
not
exclude
the
other.
Just
like
beer
does
not
exclude
beauty.
I
see
a
lot
of
value
in
indexing,
the
hops,
because
you
can
really
decide
how
a
packet
gets
forwarded
if
this
could
allow
you
to
to
design
it
to
your
deck.
But
then
you
would
not
know
when
you
have
a
choice.
If
you
go
left
or
right
so
so
I
mean
storing
against
non-story.
I
E
Index
host,
so
basically,
yes,
you
have
to
do
they
either
you
index
the
vertices
or
you
index.
The
edges
yeah
either
way
right,
so
you
want
to
go
through
a
graph.
You
can
decide
to
index
the
edges
or
all
are
the
so
in
our
case,
the
final
in
G
is
because
we
have
story
mode
or
all
you
mean
that's
the
the
vertices.
That's
your.
I
K
E
Yeah,
that's
the
other
part
of
the
design,
really,
okay,
anything
but
no
I
mean
no
okay,
since
there
is
a
vertex
to
every
leaf
node.
There
is
more
vertices
globally
in
your
craft
and
our
leaf
nodes.
We
have.
We
are
just
addressing
the
leaf
node,
so
we
have
less
thing
to
store
if
you
are
doing
non.
Storing
that
would
be
equivalent
since
we
are
doing
storing.
E
We
just
address
the
leaves
and
because
we
just
addressed
the
leaves
we
have
less
lists
and
vertices
in
the
network,
because
that's
just
the
end
of
some
of
the
vertices
the
edge
was
when
we
are
doing
story.
Well,
this
doing
story
yeah.
So
that's
why
it
does
it
chooses
less
information
than
severe
twin
two
to
all
the
paths.
Every.
B
H
E
B
B
J
B
That
the
question
is
my
mind
is:
if
we
could
do
only
one
thing,
would
you
suggest
we
do
this
or
no
path
or
Dow
projection?
I
mean
they're,
not
comp
they're,
not
mutually
exclusive?
Okay,
it's
for
sure,
okay.
But
but
the
interesting
question
is,
is
you
know
if
we
had
limited
time?
Do
you
think
this
is
a
actually
a
better
way
of
getting
to?
D
C
E
E
E
For
problem,
you
you,
basically
you
have
this-
always
this
problem
filter
of
tuning
them
right.
You
don't
have
the
problem
with
bits,
but
with
bloom
filters
at
some
point
there
are
things
which
which
pollute
the
bloom
and
you
need
to
clean
it
up,
but
that
does
not
really
apply,
because
this
is
rebuilt
anyway
by
the
route.
Every
time
you
send
something
so
I
don't
see
that
such
a
problem,
we.
E
E
Let
me
make
it
reliable
two
minutes
to
it,
so
you
should
thought
the
other
piece
was
fun.
Look
at
this
one,
so
the
other
challenge
of
multicast
is
to
make
it
reliable,
yeah,
and
so
what
you
can
do
with
these
things?
Yes,
okay,
I'm
sending
so
I'm
the
route
and
then
again
you
could
be
doing
from
somewhere
else,
but
on
the
route
and
I'm
sending
this
this
packet
to
this
particular
bitmap.
But
that's
something
that
we
can.
We
do,
but
vo
does
not
have
it
yet.
So
we
are
bit
ahead
of
them.
E
E
So,
as
you
saw
in
the
packet
back,
the
acknowledgements
on
the
multicast
I
also
put
together
on
the
bitmaps
of
old
C,
and
so
if
there
is
a
loss
when
the
overall
acknowledgment
did
I
go
too
quick
here.
So
there
is
a
last
year
and-
and
there
is
a
bunch
of
devices
which
don't
get
the
badge
gas
message,
so
they
will
not
acknowledge,
and
so
what
you
get
is
out
of
this
nation
that
I
Adam,
a
multicast
group
on
these
technicians
will
get
will
get
in.
E
E
E
E
E
E
J
F
Right
so
I'm
Raja,
though
I'll
try
to
make
it
real
quick.
So
this
isn't
continuation
to
what
at
presented
in
Chicago
it's
regarding
employing
the
route,
invalidation
efficiency,
improving
the
route,
invalidation
efficiency
in
context
to
repose,
storing
mode
of
operation
of
course,
route
invalidation
is
there
in
context
only
storing
mode
of
operation.
So
now,
before
going
into
the
details,
you
know
I
just
want
to
talk
about
what
Michael
said.
I
wanted
to
what
I
wanted
to
say
here
is
the
no
part
Dow
optimization.
F
F
So
a
quick
recap.
Quick
recap:
we
have
the
current
ripple
and
peda
mechanism,
the
primal
problem
with
that
is
it
the
NP
Dow
is
sent
from
a
path
which
is
suboptimal.
The
node
decides
to
switch
a
path
to
another
parent
and
it's
to
tries
to
send
an
NP
route
through
its
previous
parent,
which
is
suboptimal
by
design.
So
that
is
what
we
are
trying
to
change,
so
nobody
here
tries
to
send
a
regular
Dow
tries
to
establish
a
new
path.
F
There's
a
common
ancestor
know
which
is
node
a
which
detects
that
there
is
a
change
in
the
next
stop.
It
detects
the
change
in
a
desktop
and
sends
an
NP
Dow
downstream.
So
this
is
the
first
time
in
Peter
and
our
message
is
going
downstream.
I
think
now.
Projection
also
requires
now
to
be
going
downstream,
so
yeah.
F
So
few
clarifications
to
be
provided
here,
compatibility
with
the
current
NP
Dow
we
can
have.
So
there
is
a
scenario
for
current
in
pit
out
to
work.
So
the
scenario
is
in
case
of
a
gracious
node
shutdown.
You
would
still
need
the
current
and
peered
out
to
be
working,
there's
no
regular
Dow,
so
you
need
to
still
have
that,
so
the
new
proposition
has
to
be
compatible
with
the
with
the
current
ripple
specification.
So.
F
Another
clarification,
the
multipath
routing
I,
won't
go
into
the
details
of
this
example.
What
I'm
trying
to
say
here
is
in
case
of
multipath
routing,
you
select,
multiple
preferred
pairings
and
you
send
down
from
multiple
preferred
parents.
So
what
I'm
trying
to
say
here
is
the
downstream
NP
drought,
the
proposed
in
downstream
NP.
Now
it's
still
compatible
with
this
current
multipath
routing
semantics
as
well.
The
example
shows
you
there's
a
node
DG.
F
Clarification
three
now
this
was
a
question
raised
by
Pascal
on
the
mailing
list.
Whether
should
we
call
it
a
dau
at
all
now
my
rationale
for
regarding
to
calling
it
a
dau
was
that
we
are
reusing
most
of
the
existing
headers
and
the
containers
and
the
transmitter.
Almost
everything
is
reused
here.
So
why
not
call
it
Dao
itself,
but
there
is
one
important
change
here.
So
the
current
specification
says
that
the
NP
Dao
changes
I
mean
basically
it
invalid
did
the
last
downward
route
to
the
target,
so
it
work.
F
F
F
B
Michael
Richardson,
so
sorry
that
your
last
slide,
yeah
yeah,
so
I
just
want
to
say,
I
think
that
that
that
we
have
enough
type
codes,
that
I
think
you
should
just
ask
for
new
type
code.
It,
the
structural,
were
the
same
in
the
code.
It's
like
you
know,
case
Tao
case
NP
Dow,
do
all
the
decoding.
If
NP
Dow
do
this
extra
stuff
and
I
think
that
works
really
well
and
I.
Don't
think
that
that
we
have
a
such
a
drought
of
numbers,
we
hundred.
L
I
B
Hundred
and
thirty
left
or
two
hundred
100
and
100
fifteen
secure
ones
and
115,
unsecured
ones,
so
I,
don't
think
I,
don't
think
it's
a
problem.
Just
ask
for
a
number
and
I
think
that
you
just
keep
the
same
structure
exactly
as
you
would
and
I.
Think
that's
good
and
say
it's
exactly
like
a
dowel
except
it
has
this
other
semantics
sure.
E
I
would
be
actually
better.
I
would
feel
better
because
I
just
wonder
about
the
interaction
of
this
with
note,
which
just
our
back
level
on
how
or
that
place.
No,
it's
a
clear
message
that
they
don't
understand.
They
won't
react
badly
on
them,
so
sure,
otherwise.
You
have
to
test
all
possible
cases
and
prove
me
that
doesn't
have
an
effect
right.
E
The
the
other
thing
that
would
like
to
see
and
have
not
seen,
but
maybe
I,
missed
it.
There
is
a
reason
why
this
thing
is
not
in
the
original
repo.
That's
because
well,
the
design
of
the
whole
thing
is
to
say
we
should
not
have
more
control
than
they
tell
you
right
when
something
is
broken
and
it
may
come
and
go
quite
often
if
there
is
no
traffic
about
that
subtree
or
whatever
else
we
don't
want
to
spend
our
energy,
repairing
and
breaking
and
repairing
rods
which
are
not
being
used
alright.
E
So
so
there
is
a
rationale
behind
for
the
fact
that
trip
was
designed
this
way,
and
this
is
compensated
by
the
FBI.
When
I
said
there
was
all
those
side
effects
of
not
implementing
FBI.
One
of
them
is
that
LPR
is
designed
to
say
oh.
If
we
did
not
fix
the
route,
then
we
can
still
discover
that
we
are
broken
on
the
data
path
and
fix
the
situation
from
there
and
the
method
to
do
that
is
effectively
provided
by
RFC
6550.
E
So,
okay,
we'll
lose
a
packet
of
two,
but
we'll
discover
the
situation
and
we'll
fix
it,
and
so
this
comes
it's
not
really
an
ultra
naté.
It
really
depends
on
what
you're
optimizing
your
network,
for,
if
you
want
to
quickly
repair
report,
does
not
have
it.
It
will
wait
for
traffic.
This
enables
to
repair
it
now.
There
must
be
some
words
of
caution,
don't
implement
this
right
away
and
blindly,
because
all
of
a
sudden,
you
will
see
your
energy
being
drained
on
in
places
where
you
don't
have
traffic
all
right.
E
F
You
for
raising
this
Pascal,
so
you
know
with
regards
to
the
overall
control
overhead
that
is
added
here.
Actually,
it's
decreased.
It's
decreased
the
reason
being.
If
you
see,
can
you
see
all
right
yeah
if
you
see
before
the
NP
da
used
to
go
all
the
way
till
border
or
it
used
to
be
all
the
way
till
the
modern
order?
Here
we
are
actually
reducing
overall
controller
overhead.
It
actually
gets
reduced,
so
we're.
F
F
Yeah,
really,
you
know
the
other
prophets
that
I've
mentioned
in
the
in
the
draft
which
talks
about
why
it
would
be
required.
For
example,
one
case
is
the
neighbor
churn.
Then
the
neighbor
the
neighbor
table
efficiency.
There
are
lot
of
other
things,
so
I've
removed
those
problem.
Use
cases
from
I'm
not
mention
two
scenarios
here.
I
would
really
encourage
the
working
craft
working
group
to
go
through
the
raft.
Thank
you.
H
A
N
N
So
so
we
raised
that
issue
and
some
people
think
that
we
should
put
in
the
same
draft
and
some
people
think
that
we
should
put
it
in
a
different
draft.
But
nobody
said
so
far
that
we
shouldn't
do
it
so
I'm,
not
absolutely
chomping
it
a
bit
to
go.
Do
it,
but
if
it
needs
to
be
done,
must
we'll
do
it
now
and
so
I
wanted
to
raise
that
issue
here.
Discussion.
D
E
N
O
N
N
Okay,
no
problem,
so
unless
there's
a
additional
request,
I'm,
not
gonna,
suggest
doing
anything
more
and
then
oops.
Then
there
was
another
question
about
really.
What
do
you
think
about
links
when
you're
first
starting
operation?
In
other
words,
you
have
your
network
and
it's
going
and
you
want
to
get
a
peer-to-peer
route.
What
do
you
think
about
the
links?
Well,
the
way
the
draft
is
currently
written.
We
just
think
Tudor,
it's
unknown.
Well,
it
K
is
symmetric.
N
E
You
know
sketchy
well.
Iii
would
like
to
keep
this
the
asymmetric
capability,
just
because
it
may
be
that
there
is
enough
bandwidth
on
the
way
in
which
I
want
to
do
a
bi-directional
traffic.
Just
the
way,
the
way
and
already
chaos,
the
bandwidth
and
so
did
the
metrics
goes
down,
and
do
you
get
two
better
metrics
once
you
have
installed
your
stuff
now?
What
is
the
bad
news
laughs?
Would
you
steal
pass
this
way
the
way
back
traffic
because
you're
using
the
same,
reduce
to
Samia?
It's
not
like.
E
N
And
then
the
last
question
it
had
to
do
with
with
other
regions
should
consider
the
routes
have
a
lifetime,
and
here
I
don't
suggest
that
we
should
do
there,
but
I
wanted
to
raise
the
issue
since
it
was
brought
up
on
the
mailing
list.
So
unless
anybody
here
wants
to
investigate
that
I
and
I
think
that
actually
specifying
it
would
be
relatively
straightforward.
E
Charlie
you're
presenting
me
here
because
I
think
it
just
shouldn't
care
and
the
reason
is
you
should
have
objective
functions
if
you
just
designed
classical
repo,
which
means
that
there
are
a
lot
of
those
decisions
which
are
left
to
something
that
you
plug
in
your
traffic
but
which
is
not
inside
your
draw.
If
you
say.
Oh,
that's,
part
of
what
the
objective
function
will
tell
me
when
it
says.
Look
that
20
right
wants
to
look
from
the
McLaren
below.
It
will
be
very
happy
about
this
and
that
it
will
tell
me
pick
up
yeah.
N
E
N
Next
steps:
well,
since
there's
not
much
health
left
to
do,
I
suggest
that
the
draft
is
ready
for
last
call
and
I
would
sure
we
could,
unless
there's
a
rising
for
these
new
features.
I
think
we
could
easily
be
done
and
then
go
back
and
make
them
well
enough,
three
or
four
small
editorial
Corrections
in
the
next
version
of
the
draft
and
and
be
done.
So
that's
four
for
the
document.
There's
also
a
more
a
little
bit
of
news
about
implementation.
There's
you
can
download
it
and
have
it
and
there
was
a
paper.
N
I
was
moved
to
a
conference
and
the
reviewer
comments.
Basically,
so
we
they
wanted
to
see
information
about
the
how
the
simulation
results
when
there's
asymmetric
links,
so
that
is
also
in
the
process
of
being
done
anyway.
Please,
if
you
are
interested,
please
take
to
take
our
code.
Please
and
that's
it.
A
A
Well,
we
would
like
to
have
the
review
results
to
make
the
last
call
if
it's
ready
or
not,
we
don't
know
yet,
for
example,
in
the
Security
section,
I
would
like
to
add.
If
you
can
please
add
more
information
like
a
ODB,
because
you
mentioned
in
secret
accounts
security
considerations.
You
mentioned
about
ripple,
but
if
you
add
some
information
about
ODB,
oh
I,.
A
N
N
P
P
I
put
this
raft
on
line
on
March
this
year,
but
not
parting.
Last
meeting
since
there's
only
one
hour
in
the
working
group
and
did
not
present
it
and
it's
time,
I
got
a
chance
to
your
s.
Pretty
nice
trout,
so
two
problems
are
addressed
in
this
draft
one
on
a
heard
problem
and
the
other
is
a
random
area.
P
So
what
is
randomly
and
honest
man
working
here?
We
have
two
figures
in
the
net
figure
node
a
and
they'll
be
has
has
on
several
potential
and
charging
and
in
the
common
coverage
they
are
two
sub
nodes
and
if,
let's
say
log,
a
and
not
B
has
table
same
rank
in
the
same
path
cost
two
days
to
submit,
then
these
two
Sumner's
will
select
apparent
randomly
and,
moreover,
those
are
greedy.
P
Take
a
ets,
for
example,
if
the
missed
half
of
the
function
is
used,
then
these
two
nose
prefer
to
choose
a
best
passed
cost
to
a
root.
So
if
not
a
is
just
a
little
better
than
and
then
they
would
be,
then
these
two
nose
in
the
common
coverage
we
are
all
chose
to
select
root,
not
a
even
if
they
don't
be
is
also
is
also
acceptable.
P
So,
as
a
result,
there
will
be
this
kind
of
unbalanced
Network,
and
if
this
network
expands
I
mean
there
will
be,
let's
say:
if
more
unsub
knows
connect
to
this
network,
then
node
a
is
likely
to
be
overloaded
and
then
the
taiga
and
and
then
P
on
switching
happens.
So
what
we
would
like
to
do
is
to
balance
the
number
of
children
from
the
very
beginning.
Let's
say
here
in
the
right
figure:
node
a
and
no
B
has
the
same
children
and
this
minimize
the
possibility
of
parents
reaching.
P
So
our
our
goal
is
to
balance
the
number
of
children,
and
here
I
would
like
to
emphasize
here.
I
would
like
to
emphasize
that
this
is
an
indirect
way
of
reflecting
the
traffic
load,
so
you
know
address.
We
do
not
mention
anything
about
the
load
balance.
We
think
it's
not
that
they
know
balance
just
honest
and
a
number
of
children,
but
this
method
can
be
very
useful
when
all
those
send
packets
in
a
same
size
and
frequency
and
in
this
case
and
I'm
not
showing
you
cross
traffic
build
and
we
what
sorry.
E
P
E
P
So
here
we
have
a
very
important
meaning
scenario
and
it's
a
advanced
metering
infrastructure
that
the
smart
meters,
let's
say
for
electricity
for
gas
rule
for
water,
a
timely
upload
data
to
obtain
call
an
injury
with
the
same
size
and
same
frequency.
So
in
this
case
financing
the
number
of
children
can
be
very
useful.
P
And
then
we
would
like
to
combine
the
number
of
children
with
other
metrics
or
constraints
like
a
media
account
or
latency,
and
here
are
three
figures
to
illustrate
the
whole
procedure.
So
as
the
at
the
very
beginning,
the
node,
a
and
the
B
are
unbalanced,
and
once
that
yeah,
we
send
down
word
together
with
a
number
of
children.
Fighting
one
and
then
these
two
nodes
in
between
we
all
know
not
B,
is
a
better
choice
and
then
they
also
get
to
not
be
so.
Here
is
album
modification.
P
So,
in
the
general
option
field
we
have
da
Jamaica,
we
have
a
DHD
metric
container
and
in
this
container
we
would
like
to
create
a
new
metric
called
CNC.
We
we
used
to
bias
one
bias
for
CNC,
namely
channel
count,
which
is
a
number
of
children
for
anode
and
the
other
bite
for
the
sigma
cnz,
namely
the
maximum
number
of
children
apparent
Kahoot
and
with
the
use
of
maximal
CNC.
P
We
can
minimize
the
use
of
jail,
negative
8,
because
sure
you
already
know
that
their
potential
parent
is
able
to
hold
more
chosen
or
not
so
in
the
following.
I
will
show
some
simple
demo
to
show
where
it
is
very
powerful
to
combine
a
sensei
with
other
metrics
or
constraints.
If
we
put
this
sentencing
and
a
new
metric,
so
first
it's
like
a
combine
that
EDX
and
CNC
together.
If
he
acts
clearly
is
a
constraint
by
your
CNC
is
a
metric,
and
here
we
have
three
potential
parents.
P
D
P
P
So
if
here
they
say
on
D
and
the
potential
parent
for
E,
then
because
they
both
have
no
charging,
then
the
note
e
will
select
among
them
randomly.
But
if
the
number
was
chosen
or
not
a
and
B
has
sent
down
word
to
e,
then
the
not
a.
We
all
know
that
happy
will
be
a
better
choice.
Then
it
will
said
select
D
as
its
parent.
P
Why
are
in
our
method,
will
compute
the
day
number
of
a
children
from
the
neighbor
table
and
we
sent
this
number
of
children
together
with
GI
Oh
downwards,
so
that
the
downstream
nose
is
able
to
select
that
near
best
parent
and
in
the
end.
Here
we
come
there,
our
children,
to
on
modifications
so
in
amongst
raft
he
put
the
parent
address
in
the
deal
option
field
and
he
would
like
to
create
a
new
repo
control
message
option.
Well,
you
know
our
on
drought.
P
E
So
we
have
discussed
at
all
times
and
I'm,
sorry
that
Michael
just
left,
but
we
have
discussed
that
six,
that
there
is
a
need
to
express
the
willingness
to
get
some,
which
is
some
abstraction,
that
you
have
some
form
of
willingness
to
join
a
do
dark.
For
instance,
we
are
looking
at
the
joint
process
in
as
part
of
the
joint
process.
E
There
may
be
multiple
join
parents
even
before
you
do
repos
you
might
select,
who
you
use
as
your
giant
proxy,
and
we
figured
out
that
there
should
be
a
there
is
a
need
for
an
expression
of
the
willingness,
something
very
abstract
of
beings
used
as
a
joint
proxy
and
of
the
number
of
people
you're
serving
at
this
very
moment.
The
more
children,
if
you
like
this
part
of
that.
But
if
you
realize
it's
something
which
truly
comes
from
the
top,
for
instance,
this
all
rippled
geotag,
maybe
saturated,
and
we
want
that.
E
The
metric
that
expressed
this
willingness
comes
from
the
top.
So
that
the
root
can
say,
hey
we're
not
taking
anyone
anymore,
it's
not
just
this
guy
with
his
children,
it's
the
whole
dyrdek,
and
so
it
would
express
a
willingness
and
unless
you
propagate
this
down
and
it
needs
to
be
propagated
down,
that's
my
second.
E
E
So
the
notes
down
can
only
say:
oh
I
want
even
less
than
the
root
oh
I'm,
just
like
the
root,
and
so
so
we
are
talking
about
an
abstract
property
that
comes
from
up
that
is
propagated,
just
like
the
rug,
but
that
has
nothing
to
do
with
the
rank
because
you
build
your
dog.
That's
that's
for
your
watching,
and
this
thing
is
about
joining.
This
thing
is
about
how
much
bandwidth
how
much
resources
you
get
left
in
this
deal
like
to
accommodate.
So
you
must
not
mix
this
because
the
semantics
are
so
different.
Yes,.
P
E
That's
also
something
is
that
and
it's
not
this
local
decision
to
this
node
second
thing
does
to
be
propagated,
just
like
the
rack,
but
it's
not
a
rack,
and
then
you
have
to
think
like
we
thought
when
we
built
3po
it's.
You
cannot
make
the
decision
of
what's
good
for
this
dyrdek.
There
must
be
something
like
an
objective
function,
but
it's
not
the
traditional
life.
Now
it's
like.
E
And
this
joint
function
would
be
just
like
an
objective
function
and
it
may
play
with
the
number
of
children
as
you're
doing
and
that's
where
you
have
something
right.
Oh
it
may
play
with
the
amount
of
traffic
that's
going
through.
Oh,
it
may
play
with
the
saturation
of
what
order
table
and,
and
so
right
now.
E
What
really
is
would
be
good
for
this
group
is
to
provide
you
abstract
mechanisms
for
the
route
to
send
that
down
the
dyrdek,
for
what
are
the
rules
for
node
to
change
it
if
they
want
to
expose
something
else
to
their
children
and
a
basically
the
API
I
mean
the
abstract
interface
for
one
objective
function
that
we
decide
actually
what
it
does
based
on
what
it
got
from
the
parrot.
So
this
is
the
thing
that
we
really
need.
P
J
O
P
P
P
O
P
That
is
possible,
okay
I
mean
well.
If
this
parent
cannot
hold
any
more
even
you
even
this
child
sent
at
the
ideal,
then
you
did
yeah
we're
only
you
saved
by
a
negative,
it's
okay,
right,
okay,
yeah,
so
I
will
use
the
maximal
CNC
so
that
the
children
already
know
there's
no
more
room
for
that
children,
so
it
we
are
not
in
any
to
deal
so
we
saw
this
problem
in
the
first
place:
okay,
yeah
and
one
more
thing.
P
I
can't
mention
here
is
that
the
mechanism
for
computing,
the
number
of
children
is
based
on
Rajas
chart.
So
in
a
story
mode
he
used
the
ideal
and
the
no
pastel
should
you
able
to
do
the
child
during
registration
in
a
mortgage
wire
in
an
Australian
mode.
He
use
the
NS
cross,
L
&,
NS,
plus
L,
with
zero
lat
trying
to
do
the
register
and
the
D
register.
E
So
this
is
the
projection
thing,
so
the
name
comes
from
the
fact
that
in
classical
writing
you
inject
a
rod
from
where
you
are,
and
that
spreads
just
like
something
in
your
blood
till
the
guy
gets
it
and
can
write
to
you
and
with
projections,
we're
basically
sending
across
the
network
a
rod
from
A
to
B
from
the
perspective
of
CC.
So
that's
why
we
call
it
projection
and
pretty
much
that's
the
model
that
you
know
off
with
Sdn
any
sort
of
thing.
E
So
today
we've
been
talking
a
lot
about
classical
ripples
touring
an
on
sterling,
which
is
just
no
more
writing
operation.
Then
we
talked
about
proactive
rotary
operation.
Now
we
talked
about
a
odv
which
brings
repo
into
the
reactive
world
and
with
this
draft
we
are
now
bringing
as
the
end
into
inter
repo,
that's
pretty
much
what
this
does
well
now,
we've
got
the
story
complete
right.
Why
not?
Okay,
let's
be
the
data
center.
E
Okay,
so
actually
it's
we
are
here
because
by
your
house
Falls,
because
he
told
me
I
want
stirring
I
won
on
stirring
in
there.
So
I
had
to
do
it
now,
so
I
did
it,
and
so
I
did
it
for
last
time.
But
since
I
did
three
we're
really
before
the
meeting
I
did
not
have
much
time
at
the
last
meeting
to
just
mention
what
was
going
on
so
Frank's
round.
E
Now
we
have
an
on
storing
mode,
which
remained
means
that
instead
of
projecting
Adel
when
doing
the
normal
double
process
along
a
series
of
diode
that
is
in
the
packet
will
be
creating
a
turn
on
with
the
encapsulation
and
everything
that
you're
the
good
thing
you
find
in
non
storing
mode
so
that
you
don't
have
to
rely
the
Dow,
but
you
just
send
to
a
given
note.
There
is
a
ton.
I'll
use
it
in
order
to
do
that.
I
had
to
restructure
the
document
quite
a
bit.
E
So
a
lot
of
the
extraneous
text
now
is
appendix
yeah.
That's
where
we
go
so
I
have
a
number
of
question
for
the
group
or
now
to
do
that.
Well,
because
conceptually
it's
quite
easy,
but
now
how
you
express
it
in
the
options,
etc.
There
are
ways
of
doing
it
and,
at
the
moment,
I'm
worried
not
too
happy
with
what
is
written
there,
how
they
totally
know
what
to
do.
That
would
look
better,
so
I
would
really
like
to
get
some
help.
E
So
initially,
with
the
normal,
storing
mode
project
it
down,
you
would
get
a
via
information
option
perhap.
So
what's
going
to
happen,
is
you
sound
the
doubt
to
the
exit
of
the
part
that
you're
creating
of
there
artichoke
rating
and
the
thing
is
forwarded
to
the
ingress
by
harp
and
the
sequence
indicated
by
the
sequence
of
those
options?
E
So
the
way
I
found
to
address
rounds
prom,
which
is
to
just
put
every
state
in
a
single
node
and
do
a
tunnel
where
the
source
routing
is
to
to
instead
of
putting
multiple
via
option,
I
put
a
big
one
with
all
the
address
in
sequence
in
there
I'm
not
saying
it's
a
good
thing,
it's
just
what
I
wrote
that
day?
Maybe
it's
better
to
indicate
it
differently.
Maybe
we
should
have
a
mode
of
operation
endow.
E
Maybe
it
should
not
even
be
a
dhow
I,
don't
know
so
we
we
need
to
discuss
this,
and
so
I
wrote
these
tags,
which
are
really
not
like
too
much
which
which
tells
you,
based
on
what
I
just
showed,
how
you
recognize
a
new
path
now
starring
versus
a
new
path
down
on
story,
knowing
that
this
thing
can
be
used,
whether
you
overall,
no
more
repo
infrastructure
is
storing
or
not
stories,
so
this
can
be.
There
is
all
sort
of
use
cases
I'm,
sorry!
E
Well,
you
can
add,
storing
our
non,
storing
mud,
projection
and
I,
don't
know
what
I
just
did.
Okay
on
the
network,
which
is
self
restoring
or
not
sorry,
so,
we'll
have
to
find
if
we
want
to
cross
all
the
possible
use
cases
etc
and
then
now
further
discussions
and
I'm
not
even
sure
whether
they
belong
to
this
draft
or
not,
for
instance,
when
you're
doing
at
the
end.
E
There
is
something
like
teaser:
what
which
will
tell
you
about
the
topology
of
your
network
so
that
you
can
compute
all
those
nice
tags
when
we
have
a
non
storing
we've
got
an
idea
of
a
do
tag,
but
if
you
want
to
build
transversal
rods,
you
know
orthogonal
to
the
do
tag
well,
even
on
storing
Montel
enough
information
that
you
would
like
to
see.
So
should
we
be
it?
Should
that
be
another
document?
Should
that
be
something
else
entirely?
E
Should
that
be
part
of
free
poll
which
I
don't
think,
but
we
need
to
answer
those
questions.
Do
we
want
to
both
on
the
passable
mixed-mode?
So
these
are
a
number
of
questions
and
only
if
you
have
time
to
debate
it
now
or
come
back
at
the
other
or
make
them
the
meaningless.
But
there
are
a
number
of
questions
we
need
to
address,
and
then
there
was
the
six
inflation
very
how
again,
which
was
a
if
you're
doing
all
those
things
together.
How
can
you
avoid
rots
and
it's
a
good
thing?
E
And
the
question
is
particularly
true:
if
you
are
not
doing
a
strict,
if
you're
doing
a
strict
non
stirring
rod
a
to
be
ups
either
you
get
there
or
you
don't,
but
you
don't
look
right,
mostly
with
six
hours
shifts.
It's
you
remove
all
the
addresses
on
the
way
as
you
go
now,
if
you
are
doing
something
news
and
you're
using
another
writing
on
the
way,
then
this
other
writing
may
take
you
back
where
you
were
the
recommendation.
Here
is:
oh,
we,
oh
by
the
way
we
thought
about
this.
E
When
we
built
ripple,
there
is
the
concept
of
a
little
bit
in
the
API.
If
people
care
to
implement
the
API-
and
this
orbit
will
tell
you
whether
you
are
going,
we
are
doing
a
pass
which
makes
sense
or
patch
which
does
not
and
if
you're
doing
a
loop
at
some
point,
you're
doing
something
which
doesn't
make
sense,
okay,
which
is
going
down
and
then
going
up.
That
does
not
exist
in
repo
and
that's
what
the
the
OB
tells
you
and
that's
how
he
did
eclipse.
E
So,
yes,
we
have
ways
of
doing
loose
and
non
end-to-end
rods.
Does
it
mean
we
should
be
doing
this
I,
don't
know
if
we
are
using
either
the
this
new
specification
with
the
street
source
route
or
you're,
using
what
we
did
previously
in
the
specification
which,
which
also
gives
you
an
end-to-end
path
and
forces
that
the
endpoint
already
has
a
dowel
rod
where
you're
going?
You
cannot
make
roads
route
loops.
E
Now,
if
you're
trying
to
mix
the
two
in,
we
have
fashions
and
I,
don't
know
what
you
may
end
up
with
loops,
so
at
the
moment
I
think
we're
safe,
but
we
can't
do
anything.
We
like
there
is
enemy
to
what
you
want
to
do
with
this
sort
of
game,
I'm
just
illustrating
what
this
with
the
with
this
bike.
So
we've
got
how
much
time
can
have
taken
to
build
this.
Now,
it's
even
a
different.
Well,
it
not
even
reuse.
The
old
one
you
see
so
here
is
here
is
the
the
non
stirring.
E
So
you
get
basically
the
path
that
from
from
a
note
to
another
one
you
see
in
this
case
there
is
a
huge
stretch:
that's
the
price
to
pay
for
not
adding
optimized
rods
for
everybody,
that's
how
we
save
so
much
energy.
As
long
as
we
don't
choose
this
path,
and
if
you
are
in
storing
mode,
then
we
will
be
using
this
path
here.
Right
would
save
a
little
bit
still.
E
It
would
be
quite
and
optimized
and
that's
when
the
this
rod
projection
come
into
play,
because
if
you
go
through
the
root
the
root
will
see
every
packet.
It
will
know
a
stalking
to
be
there's
a
lot
of
traffic
and
by
the
way,
it's
really
an
optimized,
because
I
have
opted
up
ology
in
my
head,
so
I
can
see
that
doing
something
wrong.
So
in
this
particular
case
here
is
what
the
drive
now
allows
you
to
do.
It
will,
it
will
basically
say:
hey
mister
43.
If
you
want
to
go
to
mr.
E
41,
you
better
build
a
tunnel
here
and
since
now
this
drafts
tables
non
storing.
It
really
means
you
have
to
are
caps
like
we
do
in
non.
Storing
and
then
ever
writing
it
down,
which
is
so
well
compressed
now
with
six
leverage.
Thank
you
and
the
cool
thing
compared
to
the
storing
mode.
Is
that
the
note
can
acknowledge
already.
E
You
don't
need
to
forward
the
Dow
along
the
path
to
install
the
storing
mode
on
each
hub
before
you
AK,
so
the
arc
is
direct
from
the
node
who
got
the
Dow,
so
it's
really
metrical
to
the
long
story
classic
on
one
story
and
then,
when
you
get
a
package
for
41
coming
to
43
now,
43
has
a
rod
to
41
using
this
multi-hop
sauce
rod,
so
it
down
calculates
the
packet
and
thought
okay.
So
this
is
stupid.
E
I
said:
look,
this
is
what
the
drafters
you
to
do
now,
so
you
can
shortcut
the
whole
thing,
but
now
this
time
we've
sorted
are
supposed
to
a
set
of
rods
along
the
path.
What
the
draft
does
not
allow
you
to
do
yet,
and
that
would
be
something
we
may
or
may
not
want,
is
send
something
more,
which
will
basically
say,
Oh
mister
43.
E
If
you
want
to
build
a
rod
to
somebody
if
yard
Stano,
you
can
also
do
it
right
very
clear
today:
come
do
it
so
if
I
want
to
build
a
path
from
43
to
51,
I
have
to
install
a
second
full
and
Torrence
sequence
of
node
in
an
eternal,
eternal
I
can't
tell
mr.
43,
with
this
new
virtual
interface
packet.
So
far,
51
can
go
through
it
and
the
question
to
the
group
whether
we
want
to
take
that
path.
That
I
always
acknowledge
good.
Now
we
could
do
this
play
it
again.
E
D
E
Same
thing,
right
I
mean
still
Dow
and
but
but
you
you
basically,
you
would
have
to
tell
well
for
the
source
route
you
will
have
to
resort
to
what
custom
has
been
doing,
because
you
need
to
index
the
links
and
if
you
use
the
classifier,
the
classical
no
path,
the
classical
projected
out
the
one
we
had
before,
then
you
can
use
what
edges
show.
So
it's
pretty
or
token
or
you
would
shake.
Oh,
the
uncle.
E
E
E
Okay,
but
you
see
I
mean
same
question.
I
saw,
should
we
be
doing
this
show
with
between,
should
we
be
doing
that
and
actually
switch
keep
asking
that
to
this
crowd
that
we
are,
we
ask
you
a
technician.
We
love
to
do
everything,
but
but
first
I
need
to
market
which,
where
should
we
end
and
all
fashion,
can
we
go
well?
Who
could
we
ask.