►
From YouTube: IETF102-NWCRG-20180719-0930
Description
NWCRG meeting session at IETF102
2018/07/19 0930
https://datatracker.ietf.org/meeting/102/proceedings/
A
A
A
B
So
we'll
start
slowly,
I
think
we're
going
to
get
more
people
the
problem
that
we
have
right
now
we're
in
parallel
with
the
map
RG,
where
some
of
the
people
are
interested
with.
This
group
are
also
interested
by
that
group.
So
I
think
we're
going
to
see
a
lot
of
moving
in
and
out
and
but
not
to
delay.
The
people
who
are
online
to
do
some
presentations
will
start.
So
this
is
decoding
for
efficient
Network
communication
research,
group,
I'm,
a
virgin
or
PT,
and
the
other
co-chair
is
nice
on
vodka.
B
Obviously
there's
the
note
well
and
I:
seek
it
him
here
and
I
know,
you
guys
still
need
I
guess,
maybe
to
update
some
of
your
IPR
I've
been
told
that
some
of
them
were
expired
or
but
anyway,
I
will
give
you
an
action
item
to
look
into
that.
I
think
the
rest
is
fine,
so
and
if
anybody
else
has
IPR
in
this
field,
please
disclose
it
and
if
there's
the
forms
on
the
web,
this
is
just
stuff
that
you
can
read
so
I
don't
have
to
go
through
this.
B
So
what
are
we
going
to
do
today
after
I'm
finished
talking,
so
we're
going
to
have
a
presentation
remotely
from
Hong
Kong
on
the
bats
codes?
Those
were
presented,
I
think
in
2015,
but
there's
a
big
update
and
Raymond
will
talk
about
that.
We
want
to
have
a
discussion
on
symbol,
signaling
and
FEC
schemes,
and
because
Mesa
and
I
and
I
think
other
people
have
found
that
maybe
there's
a
better
way
to
make
progress
in
this,
especially
with
the
current
draft.
B
We've
made
a
lot
of
good
progress
there
and
yesterday
the
quick
meeting
they
said
they
were
starting
to
look
at
the
next
type
of
work.
Once
the
current
version
of
quick
is
going
to
be
open,
finished.
Sorry
and
they've
also
also
defined
a
new
type
of
frames
inside
quick,
which
is
very
good
for
us
to
encoding.
The
people
on
line
Nicholas
and
Emmanuel
will
do
a
quick
status
on
the
network
coding
and
satellite.
B
This
is
actually
pretty
stable
and
one
question
will
be:
should
we
accept
the
accept
it
as
a
research
group
item
hitter
she
and
Cedric
are
not
there
on
the
IC
n1,
but
there's
also
I,
think
that's
is
also
very
stable
work
and
then
what's
next
we're
thinking
of
a
interim
in
mid-september,
and
thank
you
to
the
IC
NRG
people
to
have
actually
advice
advertised
that
it
was
a
25th.
It
may
be
26,
but
it's
around
there.
B
C
B
So
in
achievements
we
had
our
first
RFC
ETA
and
it's
all
the
taxonomy.
So
now
we
know
where
you
were
talking
about
next,
so
we
have
a
number
of
active
ideas.
We
have
the
tetris
ID,
which
again
we're
going
to
talk
a
little
bit,
we're
not
going
to
talk
about
it,
but
we
would
like
it
to
make
some
progress,
so
we
could
accept
it.
As
a
research
group
item,
we
have
the
draft
idea
that
I
mentioned
the
coding
and
satellite.
The
content.
B
D
Okay,
I
think
you
Maria
say
and
Vincent.
Thank
you
for
accommodating
my
talk.
So
yes,
please,
data
slides
and
it's
the
last
one
yeah
okay.
D
Can
you
guys
hear
me?
Okay,
very
good.
Okay,
hi
everybody,
my
name
is
Raymond
Jung
from
chk
I
gave
a
talk
on
Bansko
some
four
years
ago
and
during
the
past
four
years
there
have
been
a
lot
of
progress
in
the
development
and
so
I'm
giving
it
at
the
top,
and
this
subject
next
slide.
Please.
D
Okay,
I'm
going
to
talk
about
something
called
the
sometimes
referred
to
as
the
multi-hop
curse
in
industry,
and
it's
well
known
that
in
a
wireless
model
of
network,
the
throughput
drops
very
drastically
after
three
or
four
hops,
namely
that
the
throughput
is
very
negligible.
After
for
your
pops,
and
that's
why
we
don't
see
very
often
wireless
networks
with
more
than
a
few
hops.
However,
a
wireless
networks
with
many
hops
are
emerging
in
many
different
applications,
as
we
are
going
to
see
next,
like
this.
D
Okay,
one
of
them
is
the
application
to
smart
lamppost.
This
picture
shows
a
number
of
smart
lamppost
in
a
urban
area.
Okay.
Now,
ideally,
we
should
be
able
to
connect
a
fiber
to
each
of
these
smart
lamp,
oh
so
that
they
can
be
connected
to
the
Internet
backbone,
but
in
a
developed
city
like
Hong,
Kong
and
also
many
other
cities
is
not
quite
possible
to
do
so.
D
The
reason
is
not
only
that
it
is
expensive
to
lay
fiber,
but
also
that
it
takes
a
lot
of
very
long
times
something
2
to
3
years
to
apply
for
all
the
permits
to
do
that.
Ok,
so
one
possible
solution
is
that
we
connect
art.
We
connect
the
an
optical
fiber
to
some
of
the
lampposts,
and
then
we
use
a
wireless
multi-hop
network
to
connect
the
nearby
lampposts
to
death
access
point.
Now,
the
on
this
lamppost
there
we
can
put
a
lot
of
things
on
them.
D
For
example,
we
can
put
census
for
for
weather
conditions
for
air
quality
control
for
real-time
traffic
conditions
and
I
will
also
I
would
like
to
mention
specifically
one
thing
that
will
be
put
on
these
lampposts,
namely
5g
base
stations,
as
we
all
know,
for
5g
we
are
going
to
transmit
in
millimeter
wavelengths,
which
means
that
the
transmission
is
very
directional
and,
as
a
result,
we
need
to
put
the
to
deploy
a
lot
of
5g
base
stations.
Ok,
now,
one
very
natural
place
to
put
these
base
stations
are
lampposts.
D
So
after
putting
that
at
the
base
station,
then
the
following
question
arises:
how
you're
going
to
connect
these
space
stations
to
to
the
Internet
backbone,
and
that's
the
solution
that
I
I
just
talked
about,
namely
that's
using
a
modular
wireless
network,
would
be
a
very
natural
candidate
solution
for
for
this
problem.
Our
next
slide,
please.
D
D
Okay
and
an
example
is
satellite
communication
networks.
There
are
companies
which
have
plans
to
launch
thousands
of
satellites
to
cover
the
globe
all
to
provide
high-speed
Internet
access
next
slide.
Please
and
an
example
is
just
a
v2
example,
which
we
are
all
very
familiar
with
next
slide.
Please.
D
Okay,
yet
another
example
is
underwater
acoustic
communication.
We
talked
to
some
people
working
on
on
underwater
communication
and
I
learned
that
the
communication
environment
there
is
actually
very
volatile.
No
matter
what
coding
scheme
you
use,
what
modulations
give
me
used
the
packet
loss,
we
got
to
be
very
high,
so
it
is
actually
very
challenging
to
build
such
a
neighbor
with
five
hops.
Next,
please,
okay.
D
In
order
to
break
this,
so
called
a
multi
hop
curse.
Next,
please,
we
introduced
bats
code.
Next,
please
what
is
fast
code.
It
is
an
advanced
demo
coding
technology
developed
in
my
group
at
chk
and
their
idea
for
multi
off
networks
with
packet
loss.
Next,
please
best
code,
transmits
coded
packets
in
batches,
which
is
a
parameter
that
can
be
tuned.
D
A
small
number
of
batches
of
thought
and
recoded
at
every
delay
and
Basco
can
dramatically
increases
throughput
in
a
wireless
multi,
Health
Network.
Next,
please,
okay,
I'm,
going
to
show
you
an
illustration,
rich,
mixed
and
mix
a
comparison
between
Pasco
and
fondant
code
are
here.
We
assume
that
the
packet
loss
rate
is
equal
to
20%
on
each
up
without
retransmission
and
the
fire
size
is
equal
to
16.
Please
say
the
fire
animation.
The
mpeg-4
mp4
in
full-screen.
D
Okay
in
this
animation
on
the
top,
we
have
fun
encode
and
at
the
top
and
the
bottom,
we
have
pass
code
transmitting
a
file
from
left
to
right.
Okay,
now
on
the
top
for
fun
encode,
because
each
media
is
doing
store-and-forward
whenever
a
packet
is
lost,
there's
no
packet
to
be
forwarded
to
the
next
hop
at
the
bottom
for
pass
code.
Here
we
use
different
colors
for
different
batches.
Here,
the
branch
size
is
equal
to
4ei.
D
Also
observed
that
when
there's
a
packet
loss,
we
use
the
packers
store
in
the
buffer
at
the
mixed
relay
to
generate
a
new
packet
to
be
forwarded
to
the
mint
to
the
next
hop.
Because
of
this
we
can
transmit
more
information
through
the
network.
Now
this
illustration
illustrates
what
we
call
it:
inner
code
of
pass
code,
there's
something
which
is
not
illustrated
in
this
animation,
which
is
the
outer
coat,
which
is
something
that
is
performed
at
the
source
node
and
also
at
the
receiving
node.
D
D
D
D
D
So
we
are
running
basco
and
finally
cool
side
by
side.
On
the
left
hand,
side
is
the
window
for
France
code
and
the
right-hand
side
is
the
window
for
funneling
code,
as
we
can
see
on
the
right
hand,
side,
because
there's
pekka
laws
on
every
hop,
although
it
may
not
be
very
significant
after
11
hops,
the
throughput
actually
is
very
low,
and
that's
why
you
know
the
picture
is
not
really
moving
in
particular
forbidden
transmission.
When
you
hit
the
eye
frame,
then
all
the
keyframes
cannot
be
decoded.
D
Okay,
here
the
benefits
of
passcode,
first
high,
throughput,
second
low,
latency
and
and
the
third
is
a
local
coding,
complexity
and
also
low
storage
requirements
that
the
latitude
features
makes
best
code
highly
implementable,
even
for
devices
like
IOT
devices
for
existing
solutions
they
before
shuttle
on
on
each
of
these
on
at
least
one
of
these.
These
features.
Next,
please.
D
D
Okay,
now
in
this
network,
we
assume
that
there's
a
20%
loss
loss
on
each
hops
so
that
after
even
one
hop
the
maximum
throughput
you
can
get
is
0.8.
So
you
see
that
after
50
hops
for
M
equals
64,
you
can
still
get
a
throughput
approximately
for
the
point
7,
which,
which
is
like
a
twenty
twelve
percent
drop
from
the
from
the
capacity.
Next,
please.
D
D
Okay,
so
in
a
nutshell,
basically
essentially
converts
a
multihull
network
into
a
single
hub
network.
It
can
be
an
enabling
communication
technology
for
many
different
things,
including
IOT,
5g,
satellite
networks,
underwater
communication
networks,
power
line,
communication
networks
and
also
for
vehicular
communications.
Next,
please.
D
F
Hi
Dave
or
ran
network
systems,
research
and
design.
What's
you
didn't
compare
the
overhead
and
bandwidth
between
that's
codes
and
uncompetitive
codes?
Can
you
just
give
us
some
first
of
all,
can
you
give
us
some
sense
of
what
that
is,
and
second
is
your
lost
model,
purely
Wireless
loss,
or
do
you
consider
congestive
loss
as
well.
D
D
Actually,
the
overhead
of
Francois's
is
very
small,
because
the
I
didn't
have
time
to
explain
that,
because
the
older
code
is
actually
a
a
matrix
falling
code.
Okay,
we
all
know.
We
all
know
that
one
problem
for
randomly
Ninevah
code
in
terms
of
implementation
is
the
large,
a
relatively
large
overhead
for
the
coding
back
sure,
but
because
the
in
our
case,
the
outer
code,
we
use
a
matrix
Hana
code.
D
G
B
D
Okay,
if
you
are
talking
about
a
very
high
speed,
the
question
is
no
okay,
because
they
currently
the
APS,
have
a
relatively
weak
CPU,
which
are
not
cater
for
for
doing
any
computation
yeah.
So
you
know
not
to
not
to
implement
that,
for
example,
in
the
case
of
smart
lamppost,
we're
actually
engaging
in
a
project
with
a
Hong
Kong
government
for
that
we
actually
have
the
luxury
of
using
a
a
computer
on
the
inside
the
lamppost
to
do
the
computation.
So
it
really
depends
on
the
applications.
I
mean
for
different
applications.
G
Because
an
apple
coding
has
been
proposed
to
for
almost
10
years
or
more
years,
but
is
it
is
very
difficult
to
deploy
in
the
commercial
networks?
So
if
we
count
serve
this
problem-
and
is
it
still
very
hard
to
do
to
see
in
the
practical
networks-
and
since
you
have
mentioned
IOT
and
5g,
so
do
you
envision?
Some
difference
are
some
benefits
in
these
two
networks
that
can
make
it
easier
for
the
network
coding
easier
to
be
deployed.
G
D
A
very
good
question:
in
fact,
one
of
the
obstacles
or
implementing
level
coding
is
complexity,
and
you
can
think
of
Basco
as
being
a
an
official
implementation
of
our
random
need
in
our
coding.
It
is,
in
fact,
a
radius
code
which
combines
the
the
nice
features
of
random,
meaning
our
coding
and
also
foreign
code.
You'll,
see
they
say
all
echo
is
a
is
a
generalization
of
common
Kosar
in
it
inherits
the
efficiency
of
onam
core
as
well.
B
G
B
B
D
B
D
B
E
So
it's
my
turn,
so
in
Vincents
we
especially
where
the
feeling
that
was
right
timing
to
have
some
discussion
on
well
FEC
cuts
on
signaling
on
protocols,
but
more
from
a
practical
perspective.
What
to
do
in
this
group?
How
to
make
this
lose
idioms
move
forward,
perhaps
more
easily,
so
I
have
just
a
few
slides
and
when
then
we
can
discuss
about
it
next.
E
So,
as
I
said,
we
have
a
bunch
of
internet
draft
that
are
related
to
these
topics.
We
have
the
tetris
draft,
which
is
a
very
detail
on
some
aspects
which
proposes
a
full
solution
that
includes
protocol
aspects.
That
includes
coding
aspect
that
includes
scenario,
aspects
with
also
a
specification
of
slamming
window
HC
codes.
E
That's
one
thing:
we
also
have
work
on
Ireland
see
with
what
that's
just
the
first
draft,
so
it's
obvious
that
we
cannot
have
only
case
we
we
may
want
to
have,
but
this
document
focuses
on
a
bunch
of
two
potential
curves
for
doing
network
culling.
It
focuses
also
with
a
lot
of
details
on
others
running
other
formats.
E
So
that's
another
perspective.
So
it's
very
detail
on
some
parts
and
very
more
elusive.
Evasive
on
some
other
parts
next-
and
we
also
have
this
work-
that
I
made
with
I've
done
with
colleagues
on
algae
for
fake
friends.
So
this
document
has
passed
working
group:
let's
go
through
death
pass
working
group.
The
school
will
see
that
good
afternoon
and
we
got
some
feedback,
including
from
the
group.
E
This
is
a
full
specification,
but
in
a
different
context,
and
then
we
have
this
new
work
on
algae
effects
scheme
for
quick
that
we'll
talk
about
later
on
during
this
meeting,
which
level
ranges
on
the
first
one,
we're
using
all
the
code
specification,
all
the
coding
journals,
but
adding
some
things
took
its
specific
to
quick,
signalling,
quick
adolphus
techniques,
and
we
also
at
this
potential
but
internet
draft.
So
the
question
we
are
with
mozzie
is
how
to
make
this
easier,
how
to
move
forward
more
easily.
That's
the
question.
E
E
What
some
work
on
most
signaling
aspects
and
went
work
on
protocols
and
which
something
that
could
be
reasonable
would
be
to
separate
those
things
in
different
parts,
maybe
in
same
document
or
in
those
on
documents,
but
keeping
them
separate
more
or
less
focusing
on
one
and
trying
to
do
as
far
as
we
can
on
this
topic
and
then
doing
the
same
for
the
other
one.
Next,
and
from
this
point
of
view,
the
fec
scheme
is
a
good
approach.
E
So
if
it's
a
scheme,
if
you
are
not
real
well,
what
is
this
at
the
same
time
a
code
specification
with
all
the
internal
details
of
this
code?
The
PNG,
for
instance,
we'll
talk
about
it
later
on
or
all
the
other
aspects.
All
the
technical
aspect
that
make
it
possible
to
implement
this.
Using
these
documents
in
a
back
free,
compatible
way
having
interoperable
is
guaranteed
if
defects
Kimmie's
well
fine.
So
that
is
at
the
same
time,
this
got
specification
but
also
signaling
expects.
E
So
it
explains
how
to
use
this
code
in
the
context
of
a
given
protocol,
and
what
I
would
like
to
say
here
is
that
maybe
this
is
not
exactly
what
we
need.
Maybe
you
should
do
things
a
little
bit
different
in
terms
of
signaling.
So,
instead
of
trying
to
explain
and
detail
how
we
can
use
these
facts,
God
in
a
given
protocol,
we
may
do
things
will
be
different.
We
may
explain
how
to
use
this
fact
code
in
general,
but
not
specifically
for
a
given
protocol.
That's
the
expression
bottom
of
the
slide
illustrates
this.
E
So
on
the
Left
we
have
classically
the
format
adder,
that's
the
kind
of
thing
that
we
find
so
that
we
found
so
far
in
FEC
schemes.
So
the
proposal
would
be
to
leave
that
aside
and
instead
of
that,
trying
to
specify
what
the
code
itself
needs.
So
in
case
of
algae,
we
need
to
specify
what
is
the
composition
of
this
cooling
window,
and
if
there
is
not
yet
in
the
screening
window
in
symbols,
then
we
can
just
say:
okay,
it
starts
at
this
so
symbol
which
can
be
identified
in
several
different
ways.
E
F
Favor
and
I
I'd
like
you
to
talk
a
little
bit
about
what
you
when
you
use
the
word
signaling.
That
means
a
lot
of
different
things
to
a
lot
of
different
people
and,
for
example,
in
the
in
the
real
RTC
world.
You
know
they
tied
a
sign,
a
separate
the
the
set
up
style
signaling,
which
allows
you
to
sort
of
like
instantiate
enough
state
to
get
going
versus
the
some
people
call
it
in
band.
F
E
Yeah
you're
right
right,
we'll
do
the
afford
to
keep
them
separate
and
be
more
specific.
In
this
case,
signaling
for
me
was
I
saw
one
audios,
so
it
was
a
little
bit
ambiguous.
So
it's
clear
that
we
need
to
synchronize
and
code
and
decoder
for
sure
and
we
encoder
as
well.
So
this
is
part
of
signaling.
E
So,
probably
that's
a
good
way
to
move
forward,
something
that
can
try
at
least
so.
That
was
the
first
point.
Second
point
is
interflow
coding,
so
there
are
use
cases
for
that.
We've
seen
in
the
past,
we
were.
There
is
also
a
desire
to
go
into
this
direction
with
Island
C.
That's
great
I
have
no
objection
for
that.
It's
just
raises
interesting
questions,
interesting
problems
that
we
need
to
to
investigate.
E
So
that's
also
something
that
we
could
adopt
as
a
working
group,
research
group
topic
of
interest
and
maybe
add
it
in
the
document
with
Ireland
C
in
a
separate
document.
I,
don't
know
no
matter.
It's
not
in
assumption
third
item
parameter
the
relation.
That's
things
that
we
tried
to
do
as
a
thousand
concern
in
the
FEC
for
effect
frame
in
the
oil
see
for
fake
frame
document.
E
So
this
is
also
a
topic
that
we
would
like
you
to
have
to
be
discussed
in
this
group
and
for
next
and
finally,
yes,
there
is
always
this
idea
of
protocol,
something
that
could
be
more
generic
than
what
she
was
done
for
tea
trees.
So
can
we
step
back
and
try
to
devise
to
design
something
to
design
a
generic
protocol?
That's
also
talking
that
we
would
like
to
see
the
discussed
and
developed
in
this
group,
and
maybe
they're
also
topics
as
well,
that
we
don't
have
in
mind
today,
but
I
forgot,
I,
don't
know
so.
E
H
Karyam
fully
from
code
on
so
I
I,
don't
know
when
is
a
good
time
to
give
a
small
status
update
on
the
Jana
cider
draft
as
I'm
gonna
give
it.
Now,
if
you
want
me
to
give
it
later,
also
it's
fine
you
under
mic,
so
let's
create
it's
looser,
so
I
mean
I,
haven't
talked
with
all
of
the
authors,
but
I
think
we
have
a
pretty
good
idea.
What
we
wanted
to
I
wanted
to.
Thank
you
very
much
for
your
comments.
H
H
Hasn't
come
back,
especially
the
technical
ones,
but
the
general
question
of
the
scope
on
the
Haider
draft
is
important
and
I
might
give
a
few
notions
today,
so
really
that
the
main
point
is
that
it's
actually
a
very
narrow
job
definition
we're
not
trying
to
be
white.
So
there
are
three
sections
in
the
draft.
The
third
section
really
has
what
is
amounts
really
a
payload
definition,
as
they've
said,
or
maybe
something
more
of
the
data
format
definition
it
doesn't
have
any
algorithms
in
it
or
anything.
H
Any
protocol
aspects,
everything
you
have
it
in
two
first
sections
is
really
background.
Information
and
presentation
of
the
few
very
few
fields
that
are
defined
in
the
in
the
payload
definition,
which
is
the
simple
format
really
so
so
we
didn't
want
to
make
it
wide
and
I
think
there
is
a
misunderstanding
there
specifically
on
this
point
on
the
inter
intra
session,
we
agree,
I
mean
there
are.
There
are
complications
in
intercession
when
you,
when
you
implement
it,
we
still
think
that
that
format
will
apply
to
most
of
the
applications
that
you
have.
H
So
your
thinking
here
is
to
take
a
step
back
and
to
say
all
of
these
applications
that
are
using
network
coding,
specifically
randomly
recording
whether
they
have
in
common
and
really
what
they
have
in
common
is
a
format
and
everything
else
can
be
designed
differently
by
different
people.
So
we
focused
on
the
format.
So
it's
a
very
narrow
format
and
that's
the
point.
I
have
to
say
and
I'm
happy
to
discuss
that.
E
H
No
I
agree
with
you
what
this
missing
really
was
a
scoping
part
and
the
introduction
which
says:
here's
what
we're
doing
in
part
1
part
2-
and
this
is
poetry-
part
3-
is
really
what
we
present,
which
is
just
a
format
a
few
a
few
and
that's
as
reusable
as
possible.
Yeah,
it's
different
I'm,
not
sure
it
is
that,
but
we.
H
H
B
Just
like
why
I
think
also
get
him
I
think
a
lot
of
times.
You
guys
send
us
messages,
I
think
it
would
be
cool
if
you
were
also
copying
the
list,
because
a
lot
of
the
questions
that
have
been
raised
actually
other
members
of
the
list
is
also
contribute,
so
I
would
encourage
you
to
copy
everybody.
Okay,
it's
actually.
B
B
I
You
Brandon,
Williams,
Akamai
I
think
the
question
of
evaluating
whether
or
not
a
common
framing
for
the
for
the
coded
packets
etc.
Whether
or
not
that's
feasible
is
a
valuable
area
of
exploration
for
the
group.
Unless
people
are
you
know,
unless
everybody
is
already
at
consensus
on
the
point
that
you
make,
that
it
probably
isn't
feasible,
then
then
it's
worth
evaluation
and
I
think
the
way
that
you
frame
it
of
looking
at
sort
of
what
are
their
requirements
for
these
different
codes
and
in
trying
to
form
some
sort
of
basis
that
that
spans.
I
You
know
the
variety
of
codes
that
people
are
looking
at
makes
a
lot
of
sense
as
an
approach
to
that
evaluation.
I
think
you
know
to
to
Dave's
point:
Dave
they've
talked
about
two
different.
You
know
aspects
of
signaling,
but
with
the
kind
of
session
establishment,
as
well
as
the
what
what
he
didn't
want
to
refer
to
as
a
in
band
and
and
the
session
establishment
part
is
also
I.
I
Think
a
valuable
area
exploration
for
the
research
group,
whether
it's
actually
meaningful,
to
establish
some
sort
of
session
establishment
protocol
that
would
allow
for
some
kind
of
coding.
Flexibility
between
the
two
end
points
involved
in
the
session
is
I
think
meaningful,
I'm
unconvinced,
based
on
the
way
that
I've
seen
this
stuff
getting
implemented.
I
Whether
or
not
anybody
would
actually
be
in
a
position
to
use
that
session
establishment
protocol,
but
I
think
it's
interesting
to
ask
the
question
and
evaluate
whether
or
not
anybody
would
actually
implement
it
and
be
prepared
to
sort
of
negotiate
aspects
of
the
session.
Yeah
yeah
and
in
the
the
third
thing
I
was
gonna.
Mention
in
in
this
is
a
comment
that
I
almost
made
during
the
bats
discussion,
but
it's
more
appropriate
here,
I
think
when
we
see
their
presentations
about
the
individual
codes
and
the
benefits
that
they
provide
and
how
they're
applicable,
etc.
I
E
That's
a
good
idea.
We
already
mentioned
this,
but
yes,
I
forgot
to
have
it
in
the
list.
Yes
looks
clearly
on
Craiglist
yeah
concerning
the
negotiation
of
parameters.
As
far
as
I
know,
on
the
examples
I
have
in
mind
concerning
flag
frame,
int
and
now
quick,
it's
very
basic
enjoy
basics,
just
matter
of
sharing
one
two
values
that
are
required
to
synchronize,
encoder
and
decoder,
and
and
that's
all,
it's
more
I
propose
to
use
this
and
I
share
this
information
to
everybody.
It's
more.
E
Negotiation
negotiation
aspect
can
be
okay,
I
support
this
and
this
code.
What
usually
and
you
send
this
information
to
me,
what
sign
no
remote
site
says:
okay,
we
will
use
this
one,
it's
more
or
less.
This
way
it
works,
but
do
we
need
something
more
complex,
more
flexible
to
really
negotiate
I,
don't
know,
but
parameter.
Derivation
is
something
extremely
important
that
cool.
So
we
put
some
way
into
this
description.
That's
on.
F
F
B
E
Me
once
again,
so
let's
continue
with
this
discussion
and
this
generic
API
for
slanging
window
SEC
codes.
This
is
something
that
we
initiate
in
the
six
months
ago,
the
first
a
third
time
I
introduced
this.
We
made
a
lot
of
progress,
in
fact,
which
means
that
we
are
moving
for
one
and
we
now
have
a
not
a
strong
proposal,
but
at
least
a
proposal.
So
just
to
remind
you,
this
is
what
we
are
considering
here.
Is
this.
E
The
thought
of
the
figure
is
not
visible.
I
should
not
have
used
right
Rex,
so
what
we
are
considering
is
just
the
codec
API,
that
is
to
say
the
API
for
the
very
low
level
codec
and
not
an
API
for
the
scheme
itself.
So
the
fact
scheme
is
much
broader
than
this
low
level
codec
and
this
fact
scheme
is
in
itself
included
in
a
much
broader
piece
of
software,
which
is
the
application
or
transport
protocol
or
whatever.
E
E
So
what
is
in
scope
of
this
API?
Well,
we
have
six
five
items
session
management,
what
codec
initialization
and
session
management,
which
concerns
the
encoder,
the
encoder
and
the
decoder.
We
also
have
this
management
of
coding
window.
We
also
have
the
management
of
coding
coefficients
so
initializing
them,
giving
them
to
the
codec
or
asking
to
collect
to
generate
them
by
itself,
so
an
algorithm
for
doing
that.
E
E
Okay,
and
in
this
is
the
case,
for
instance,
of
this
8002,
so
symbols
mapping,
which
means
which
is
the
essential
key
feature,
which
is
what
this
mapping
is
meant
to
map
what
we
see
from
the
application
or
from
network
if
you
are
really
not
and
what
you
will
produce
so
symbols
that
are
required
by
the
correct
itself.
So
this
mapping
is
not
thought
of
this
API,
so
that's
the
same
for
transmission
or
assumption
of
packet
signaling.
E
All
those
aspects
are
not
proven
by
deception
next,
so
the
goal
of
this
work
is
to
be
compatible
with
sliding-window
codes.
Only
we
are
not
considering
blood
clots
because
they
are
working
in
a
different
way
and
we
felt
that
was
not
possible
to
ever
satisfying
API.
That
would
cover
both,
but
within
sliding-window
codes
we
want
to
be
compatible
with
any
code
that
we
work
this
way
so
I'll
see
our
land
sea
baths.
All
those
codes
should
be
should
be.
E
We
should
be
able
to
use
this
API
for
all
those
codes
and
we
I
want
to
make
it
clear.
This
API
must
be
compatible
with
both
end
to
end
use
cases,
situations
where
you
have
only
one
encoder,
one
decoder,
but
also
in
situations.
While
you
have
this
Rhian
coding
capability,
we
want
to
have
an
API
show
all
those
features
next,
so
the
API
is
structured
in
five
different
sections.
The
first
one
is
for
the
general
definitions.
E
I
will
just
mention
the
fact
cut
point,
because
it
is
something
that
is
important
then
so
the
second
section
is
for
the
two
is
about
opening
window
management.
The
next
one
is
about
coding,
coefficients
management
and
then
encoder
functions
whose
that
are
specific
to
the
encoding
side.
Re-Encoding
sign
I
forgot
to
mention
this
in
the
slide,
and
then
we
finish
with
the
decoder
specific
functions,
so
I
will
go
through
some
of
them
rapidly.
I,
don't
want
to
bother
you
too
much,
it's
a
bit
technical,
but
the
interesting
aspects
I
think
next.
E
Okay,
so
this
notion
of
a
physical
points
can
be
useful
and
important.
This.
The
idea
is
to
have
an
identifier
of
codec,
a
correct
identifier
that
is
used
only
locally.
It
won't
be
used,
it
won't
be
exchanged
with
the
network
and
it's
very
important
because
for
this
FC
cut
points
you
will
desire,
you
will
point
to
a
given
correct
in
what
is
available.
Maybe
you
have
more
than
one
if
we
include
also,
it
will
determine
also
some
key
parameters
like
the
finish
field,
if
you
have
multiple
codecs,
multiple
codecs,
with
specialized
with
different
fields.
E
By
using
this
code
point,
you
can
point
to
one
particular
one
about
one
particular
correct
sorry,
it
will
also
point
to
internal
techniques
like
the
way
the
coding
coefficients
are
managed,
so
you
can
design
a
product
that
assumes
that
those
coding
coefficients
are
provided
by
the
application
and,
at
the
same
time
you
may
add
also
a
codec
that
whoosh
that
includes
a
generator
function
that
will
generate
those
coefficients.
So
you
can
distinguish
between
them
using
this
efficient
code
point
so
at
the
bottom.
E
Just
to
trace
these
points,
you
can
have
the
same:
a
physical
increment
in
two
different
codecs
which,
for
instance,
different
implementation
levels,
and
you
will
have
in
that
case
to
cut
points
one
for
each
of
them.
So
that's
the
idea.
That's
used!
Only
locally
and
that's
an
example
of
this
list,
you
may
have
additional
ones.
This
is
not
sent
to
the
remote
site.
Only
the
efficient,
clean
ID
will
be
centrally
one
outside,
because
the
f-15e
ID
identifies
effects
scheme
yeah.
This
is
just
manage
a
nice
fair
to
do
that
locally
within
the
an
application.
E
F
Different,
no
resistance
restrictions
design.
This
is
not
what
the
crypto
folks
did
with
cipher,
suites
and
I.
F
Don't
know
whether
this
is
the
right
thing
to
do,
or
the
approach
they
took
is
the
right
thing
to
do.
The
my
understanding
is
the
reason
they
went
in
the
direction
they
went.
Is
that
when
you
have
all
these
degrees
of
freedom
in
describing
the
the
interface
between
the
upper
layer
and
the
codec,
you
get
a
combinatoric
explosion
of
options
yeah
and
it's
very
difficult
to
figure
out
whether
two
things
will
work
together
or
to
do
any
of
the
kind
of
certification
that
it's
actually
working.
F
E
E
Okay,
so
then
I
have
cutting
window
management
functions,
oh
well.
The
idea
is
that
the
it
works
differently
at
under
@erisa
voice
under
this
anchoring
window
is
managed
progressively
you
add
more
symbols
as
they're
as
they
arrive
from
the
application,
and
since
the
culling
window
has
a
maximum
size
when
this
cunning
window
moves
to
the
left
to
the
right.
Oh
sorry,
right.
E
Sorry,
when
the
cleaning
window
moves
to
the
right,
you
have
to
remove
the
old
symbols
dies
mechanism
for
doing
that
automatically
and
the
application
is
aware
thanks
to
a
call
back,
then
you
have
this
similar
functions
for
the
cooling
coefficient
management.
The
idea
is
that
you
have
basically
two
possibilities:
either
the
application
provides
the
cooling
coefficients
because,
for
instance,
this
reckoning
points
on
the
application
knows
what
to
do.
What
coefficients
to
you.
E
So
that's
one
possibility
and
you
have
a
set
coding
coefficient
that
function
for
doing
that
or
you
may
ask
the
correct
to
generate
the
coding
questions
for
you,
because
these
effects
can
be
associated
effects.
Schemes
comes
with
a
coding
coefficient
generation
function,
so
it's
just
matter
of
asking
the
codec
to
generate
them,
who
seed?
Usually,
there's
a
seed
following
that,
so
you
provide
the
seeds.
This
is
the
current
wave
generator
encryption
function
with
the
key,
which
is
the
the
seed
for
the
PNG,
and
you
you
lived
in
the
correct
generator
of
them.
E
You
can
retrieve
them
we'll
get
coding
coefficient
tap
function.
If
you
need
it,
it's
not
missile
case
next
and
then,
once
you're
cunning
window
is
ready.
Once
you
clean
coefficients
are
ready,
you
can
ask
to
generate
a
new
to
build
a
new
with
us
and
you
have
a
function
folding
that
so
that's
basically
the
way
it
works
and
the
unclean
side
may
be
a
bit
differently
at
the
Rankine
insight.
We
need
to.
E
Okay,
next,
then,
for
the
decoder,
it
depends
on
whether
you
receive
we
pass
symbol
or
so
symbol.
If
it
is
a
so
symbol,
then
that's
it's
pretty
easy.
You
just
have
to
decode
ask
the
decoder
to
decode
with
what
has
been
received.
So
there
is
a
function
funny
in
that
decode
with
new,
so
symbol
provide
the
buffer,
the
size
and
that's
all,
and
for
the
in
the
case
that
you
receive
a
new
reaper
symbol,
you
just
have
to
inform
the
correct
what
the
cunning
window
that
has
been
used
to
generate
this
Reaper
symbol.
E
What
does
what
are
the
coding
coefficients
either
by
providing
them
by
asking
the
collect
to
provide
to
generate
these
loose
questions
thanks
to
a
key
that
has
been
carried
in
the
packet
either?
And
then
you
call
the
decoder
is
Nuri
passing
ball
and
that's
so
maybe
the
code
with
newspaper
symbol
will
decode
something
actually
or
not.
It
depends
on
the
linear
systems.
E
One
very
important
feature
is
the
presence
of
callbacks
or
callbacks
are
functions
that
are
implemented
within
the
application
like
duck
that
are
called
each
time.
Some
event
happens,
so
this
event
can
be,
for
instance,
with
an
encoder
the
fact
that
so
symbol
has
been
removed
from
the
killing
window.
So,
as
I
said,
the
cunning
windows
maximum
size
most
of
the
time.
So
if
it's
shifts
on
the
right,
then
maybe
most
probably
there's
an
old
saw
symbol
that
has
been
removed.
E
So
by
this
way
by
this
callback,
you
can
inform
the
application
that
this
symbol
has
been
removed
from
the
killing
window
and
then
the
education
is
is
free.
Well,
whatever
is
needed,
freeing
the
buffer
entry.
That's
next,
and
you
have
a
few
more
callbacks
for
the
decoder
I'm,
not
completely
sure.
This
is
the
final
list
we
will
see
later
on,
but
okay,
we
have
a
callback
typically,
each
time,
so
symbol
has
been
decoded.
That's
very
important
and.
E
Yeah,
that's
me:
I
skip
the
details
so
now
what
we
need
to
do
is
to
challenge
this
API
with
a
real
implementation.
That's
probably
the
next
big
thing
to
do
for
this
API.
We
already
we
discussed
with
co-authors
and
Cedric,
pointed
to
me
that,
okay,
maybe
something
was
missing,
was
done
wrong
way
in
this
API.
So
we'll
fix
this
I
don't
go
into
test
once
again:
I
skim,
the
ESI
and
then
yes.
The
final
point
is
that
I'm
not
sure
this
API
is
great.
E
When
you
have
Hardware
collects
that's
something
that
was
already
mentioned
raised
by
Dave
next
time
last
time,
that's
something
important,
but,
honestly
speaking,
I
don't
know
how
to
do
that,
given
the
way
it
works.
So
we
need
to
think
about
it.
Morgan
again
with
this
approach,
since
we
move
symbols
independently
from
one
another,
it's
maybe
not
the
best
solution,
because
the
MA
will
take
time
to
set
up
and
all
these
lose.
E
E
So
this
is
very
basic.
We
have
this
formula
so
use
it
works.
This
way
you
have
a
scene.
So
once
you
have
seen
the
sequence,
then
if
you
need
one
more
so
the
random
number,
it's
just
matter
of
computing
this
formula
eight
times
the
previous
PNG
number
modulo
M,
where
a
and
M
are
constants
that
has
been
chosen
by
the
designers
by
talk
and
meter
a
equals.
E
Sixteen
thousand
eight
hundred
and
seven
and
what
has
been
chosen.
But
it's
the
right
value
to
use
and
M
is
this
value.
So
it's
pretty
easy
to
do
and
we
had
no
problem
with
LDPC.
But
in
fact,
the
way
it
was
used
in
LDPC
was
quite
different
from
what
we
were
used.
What
were
what
we
need
in
I'll
see
in
LDPC.
We
see
this
PNG
only
once,
and
then
we
produce
a
very
large
number
of
pseudo-random
numbers
from
seed,
and
we
also
scale
I
forgot
to
mention
this,
but
it's
important.
E
We,
the
result
of
this
PNG
is
between
0
and
maximum
value
and
in
practice
we
need
to
scale
this
between
0
and
a
smaller
value.
So
this
is
the
motivation
for
this
max
P.
So
we
have
this
scaling
and
in
case
of
LDPC,
this
maximum
value
is
quite
large.
It's
several
thousands,
because
this
is
a
large
block
and
there
is
a
dependency
blocksize.
E
Next,
okay,
but
with
I'll,
see
it's
totally
different,
it's
totally
different
because
we
need
to
produce
a
limited
number
of
random
pseudo
random
values
for
given
scene.
So
typically,
this
is
meant
to
the
nine,
the
compute,
the
cutting
coefficient
so
which
we
have
a
cunning
window
of
size
20.
Then
we
need
to
produce
22
the
random
numbers
using
these
PNG
and
the
program
is
that,
most
probably
if
the
application
generates
to
repair
symbols
for
the
same
killing
window,
this
application
will
use
some
seed
s
and
then
for
the
next
with
a
symbol.
E
It
will
use
seed
plus
1
s
plus
1.
And
if
you
come
back
to
the
previous
slide,
you
will
see
that
the
difference
between
the
two
produced
values
will
be
only
eight
10s
or
eight
times
s
plus
1,
and
this
difference
is
not
that
large.
So
next,
yet
not,
and
so
we
you
can
do
a
very
easy
test.
You
test
with
seed,
1,
2,
3,
4,
etc,
and
you
see
what
you
get
as
a
result
when
you
scale
the
result
between
0
and
255,
because
we
are
working
on
finish
field
20
point
eight.
E
So
we
need
a
value
between
0
and
255,
and
you
see
that
the
first
value
is
always
0
for
certain
time,
then
always
1,
then
us
to
then
well
3,
etc,
etc.
So
it's
typically
very
bad
behavior
for
us,
because
it
means
that
the
two
with
symbols
will
use
exactly
the
same
coefficient
for
for
the
first,
so
symbol
and
it's
bad
does
not
bring
you
enough
diversity.
It's
it's
totally
useless
to
send
to
repel,
in
that
case,
concerning
the
first
sumo
and
there
are
also
bad
properties
in
terms
of
well.
Why?
It's?
E
Not
it's
not
a
good
Connie,
it's
not
a
good
PNG
for
our
usage,
so
we
searched
next
and
we
decided
to
use
something
else,
especially
as
sometimes
we
need
to
scale
between
0
and
15
and
of
0
1
255.
So
the
situation
is
even
worse
in
that
place
and
we
finally
came
with
this
teeny
mt-32
PNG,
which
is
a
well-known
and
now
yes,
it's
one
of
the
most
we
used
I
think
PNG.
E
Currently
it
is
a
variant,
a
compact
giant
of
NASA
and
Twista
PNG,
which
have
trouble
quality,
and
the
good
side
also
is
get
comes
with
a
reference
implementation
that
is
included
in
our
documents,
and
we
solved
all
problems
with
this.
Well,
we
think
we
saw
forums
with
this
PNG.
So
the
question
is:
is
it
fast
fast
or
not?
That's
next.
E
The
the
answer
to
this
question
is
yes,
it's
pretty
fast,
so
the
initialization
is
bit
longer,
but
otherwise,
when
you
just
need
to
produce
random
values,
it's
quite
fast
in
this
environment,
in
the
cortex
a15
embedded
board,
we
are
most
was
almost
2
times
faster
than
the
previous
one
in
a
commercial
PNG.
So
this
is
not
a
problem.
We
also
did
tests
on
the
MacBook
Pro,
but
then
it
was
a
little
bit
the
opposite.
It
was
not
exactly
2
times
slower,
but
almost
we
don't
try
to
investigate
much
more.
E
This
difference
and
in
fact,
doesn't
matter
so
much.
The
idea
is
that,
in
terms
of
pseudo-random
number
generator
speed,
it's
fast
enough
for
our
usage.
Next
one
wants
specificity
with
this
teeny
mg42
geonji.
We
need
to
specify
three
key
values:
three
magic
values.
There
is
a
complex
program
for
determining
what
values
you
need
to
provide.
E
You
have
millions
of
possible
triples
for
initializing
this
we
chose
one
of
them,
which
is
a
one
that
was
mentioned
in
paper
and
also
generating
big
lists
of
values
was
the
first
one.
So
we
decided
to
choose
this
one,
so
it's
very
important
to
keep
to
keep
these
those
values
or
image
with
something
else,
but
those
values
are
working
and,
finally,
to
finish
when
this
I'll
see
for
fake
frame
document
will
be
published
as
RFC,
we
will
have
this
full
specification
full
specification
of
this
T
on
T
for
T
to
PNG.
E
That
could
be
reused
in
also
documents
and
I.
Think
that
having
a
good
PNG
is
essential
for
for
this
research
group
for
documents,
so
you
may
use
something
else,
but
keep
in
mind
that
PNG
is
is
not
trivial,
I'm,
not
mathematicians,
I,
don't
try
to
change
anything
in
the
PNG
myself,
I'm
using
them
in
the
way
the
authors
think
is
the
most
appropriate
and
we
we
should
stay
with
this.
E
This
approach,
because
at
some
point
of
time
we
imagine
removing
the
first
sequence,
the
first
pseudo-random
number
on
skipping,
the
first
one
or
the
first
ten
once
and
using
the
other
ones
with
a
partner.
It
was
not
a
good
approach
to
good
idea.
So,
stick
with
this,
it's
a
good
approach.
I
think
it
makes
sense
it's
reasonably
fast.
It's
just
proved
all
quality,
so
that's
the
point
that
the
feedback
here
questions
know
the
internal
state
of
the
East
Indian
to
mt-32
is
quite
large,
so
this
is
not
something
you
can
send
on
the
network.
E
H
B
E
H
Made
a
big
mistake,
but
it's
fixed
now
so
just
to
connect
yeah
gentleman
from
Akamai's
point
I.
Think
it's
really
good
idea.
Also
to
look
at
the
the
impact
I
mean
it's
a
general
point,
of
course,
to
look
at
the
impact
of
these
choices
on
what
we
decide
to
benchmark
as
the
performance
of
these
codes.
So
in
this
case
you
mentioned
the
speed
which
is
great,
but
then,
if
we
have
a
common
framework
for
know,.
E
B
We
now
pushed
it
I
think
I'm
going
to
talk
about
it
at
the
end
in
the
milestone,
I
think
it's
been
pushed
further
out.
However,
I
still
think
it's
a
very
important
aspect
to
look
at
the
gentleman
who
was
talking
about
the
issues
of
implementation
of
network
coding.
This
is
actually
an
issue
that
keeps
coming
back.
How
that
were
coding,
influences
or
is
influenced
by
the
performance
of
TCP
in
particular,
so
I
wanted
to
raise
it
today,
because
I
knew
that
we
had
online
and
in
the
room
some
of
the
people
hadn't
been
involved.
B
I
think
it's
important
that
they
can
support
it,
that
we
continue,
but
I
don't
know.
I
would
like
to
know
if
anybody's
ready
to
contribute
if
the
work
that's
been
done
or
if
a
manual,
for
example,
if
you're
ready
to
present
your
version,
you
know
as
some
kind
of
a
draft
in
the
group
which
could
be
one
solution
or
if
I
come
I
who
had
been
interested
at
the
beginning
would
be
interested
to
continue
this.
It's
more
of
an
open
thing,
because
it's
part
of
our
milestones
and
I
thought
I
wanted
to
raise.
It.
B
Well,
I
think
it's
worth
it's
worth
reopening
it
and
see
if
there
is
now
maybe
more
more
interest
again,
because
I
think
we
are
moving
towards
like
real-world
implementations
but
I
think
it's
worth
it.
Maybe
at
least
send
a
message,
and
you
know
see
what
other
people
respond.
Okay
and
again,
I
think
because
worst
I'm
going
to
start
and
later,
but
I
was
also
on
the
line
to
start
they
it
draft
on
issues
for
implementation
and
if
I
could
refer
yours,
I
wouldn't
have
to
write
that
part.
B
E
A
B
K
So
this
is
an
update
on
coding
for
quick,
first
I
want
to
think
where
you
say
and
for
for
all
their
help
in
this
obviously
I
know
a
lot
about
quick
I
know
very
little
book
coding.
So
you
know
it's
kind
of
a
you
know,
one
of
those
things
where
we're
trying
to
get
people
together,
who
are
experts
in
book
fields
and
and
produce
a
combined
solution
that
works
next
slide.
K
So,
at
least
at
this
point,
the
the
main
principles
are
that
the
protection
is
at
the
stream
level.
This
is
mostly
done
at
the
moment
for
simplicity.
We
can
certainly
imagine
cases
when
you
would
want
to
do
cross
stream
protection,
but
I
think
the
the
framing
and
other
things
increase
in
complexity
quite
a
bit
and
we
can
kind
of
go
into
why
that
might
be
true
later
you
can
negotiate
FEC,
we'll
talk
a
little
bit
about
the
trade-offs
there.
K
K
We
are
not
leveraging
the
quick
packet
number
field
which
originally
we
talked
about
doing
and
actually
the
original
quick
forward
error
correction
scheme
did
actually
use
the
packet
number
field
and
it
did
everything
on
a
packet
granularity.
The
nice
part
is
that
the
stream
is
typically
the
thing
you
actually
want
to
deliver.
K
So
quick
has
something
called
transport
parameters.
These
are
just
basically
a
list
of
key
value
pairs
that
are
sent
in
the
handshake.
The
client
sends
there's
the
server
since
there's
back
and
in
the
case
of
the
this
negotiation,
it's
just
a
matter
of
the
client
says:
I
can
use
all
of
these
things.
The
server
says
I
can
use
all
of
these
things
and
the
intersection
set
is
sort
of
like
what
you
can.
K
You
can
agree
on
so
because
it's
a
fairly
open-ended
format,
it's
possible
to
put
parameters
and
other
like
scheme,
specific
information
in
the
header
in
the
kind
of
value
portion
of
the
header.
So
you
can
say
like
have
a
reserved
header
for
I
want
to
negotiate.
You
know,
afford
error
correction
scheme
and
have
the
metadata
within
it
be
somewhat
rich
and
complex
depending
on
the
scheme.
So
obviously,
like
you
know,
you
have
to
both
agree
on
something
book
size
that
it
provides
quite
a
bit
of
flexibility.
K
K
So
the
RLC
configuration
information
there's
an
encoding
ID
which
identifies
the
scheme
with
an
eye
in
a
registry,
and
we
have
the
simple
size
which
is
at
least
initially
fixed
over
the
course.
The
connection,
that's
what
it's
also
worth
noting
we
now
have
an
or
registry
or
going
to
have
an
eye
in
a
registry
for
all
of
the
quick
extensions.
So
the
definition
of
the
frame
that
we're
going
to
be
talking
about
as
well
as
these
and
the
transport
parameters
goes
in
and
I
in
a
registry
as
well.
So
yes,
so
is
it?
K
You
know
they
may
have
different
trade-offs,
real
time
content
versus
like
bulk
content,
but
there's
a
lot
of
extra
complexity,
because
then
you
have
to
identify
which
scheme
you're
using
every
single
time,
I
use
it
or
you
have
to
set
up
a
stream
and
say
like
when
I
create
this
stream.
I'm
gonna
use
this
type
of
every
pair,
simple
purchase
that
time.
F
B
K
B
Character,
you
really
have
to
have
the
request
sent,
but
you
can
maybe
wait
for
the
reply.
Yep
so
and
then
you
would
use
two
schemes
because
you
may
not
need
to
have
the
same
requirement
so
I
think
yeah.
Thank
you,
David.
Yes,.
K
Yeah
good
point,
thank
you.
So
we
map
from
a
frame
data
to
the
source
symbols,
so
the
application
data
is
a
variable
size.
So
this
is
all
kind
of
in
order:
sequence
of
bytes.
It
arrives
in
multiple
stream
frames
and
then
the
symbols
kind
of
map
on
to
second
sub
segments
of
that
stream
and
protect
different
portions
of
the
of
the
stream
and
so
we're
using
the
stream
offsets
as
the
kind
of
what
key
into
what
we're
actually
trying
to
protect.
K
So
on
the
choice
of
the
symbol,
size,
you
can,
in
theory
use
anything
longer.
Symbols
are
kind
of
fit
into
frames,
better
easier
to
manage.
They
have
less
overhead
because
you
don't
have
multiple
frame
overheads
from
each
one,
but
as
it's
out
the
larger
the
symbol,
the
more
it
covers
in
the
more
you
might
hit
a
boundary
that
you
can
actually
use
and
recover,
and
so
that's
also
unfortunate.
K
There's
no
need
for
in
encoding
symbol,
ID
I'm
less
familiar
with
this
the
subject,
but
my
idea,
my
understanding
is
that
essentially
we're
using
the
stream
offset
instead
of
an
including
civil
ID.
Is
that
correct
and
so
we're
basically
using
a
built-in
quick
mechanism?
This
dream
offset
offset
end
ID
instead
and
you'll
see
that
I
believe
in
the
next
term
in
the
next
slide.
When
we
talk
about
what
the
frame
layout
actually
looks
like
for
the
repair
frame,
I.
E
Cannot
something
historically
speaking
in
arm
T
in
fact
frame?
We
always
consider,
including
some
more
ideas
that
identify
Isis
was
all
reversible
the
investigation
case
of
quick.
We
can
get
rid
of
this
identifier
and
replace
it
with
this
offset
offset
within
the
source
flow,
which
makes
things
much
simpler
in
fact
more
effective.
So
that's
a
good
point
and
a
good
feature
of
quick
that
will
reuse
in
this
context.
K
K
It's
gonna
have
some
common
features
that
we're
gonna,
see
like
the
stream
ID
that
it's
trying
to
repair
and
the
offset
and
then
kinda
variety
of
things
that
I
think
are
like
kind
of
pulled
in
from
the
feck
frame
design
kind
of
a
comment:
it's
like
a
hybrid,
it
seems
of
the
the
fact
frame
design
and
the
quick
stream
frame
design
next,
one
yeah.
So
this
is
the
the
wire
for
MATLAB.
The
the
stream
ID
is
a
variable
length
integer,
which
is
a
standard
kind
of
quick
currency,
of
how
we
encode
stream
IDs.
K
The
next
is
an
offset
in
the
stream,
so
that's
also
kind
of
a
standard
like
quick
stream
ID'd,
our
quick
stream
frame,
sort
of
item
and
the
next
one,
of
course,
is
the
length
of
the
novice
stream
data.
It's
actually
the
repair
data,
that's
in
the
payload
below,
and
then
you
have
the
p
rng
seed
and
the
I
guess
they
are
LC
density,
prim.
So
Brandon.
E
E
In
case
of
fake
frame,
you
will
have
those
42
bits
and
not
in
case
of
fake
frame
and
offset,
but
is
the
including
single
ID
and
a
dot
Phi
of
the
first
so
single
in
decoding
window,
and
then
you
add
the
full
40
to
world
42
bit
world
that
provides
the
key,
the
DGD
NSS.
It's
exactly
the
same.
So
both
what
some
say.
K
So
one
issue
is,
you
may
actually
like
have
a
last
symbol
that
you
want
to
send.
You
don't
have
enough
data
to
fill
up
the
window
and
so
you're
kind
of
stuck
there
like
waiting
to
send
it,
and
then
you
just
never
end
up
sending
it
because
you
never.
You
know
you
go
into
questions
for
for
too
long
of
a
period
of
time.
I'm,
probably
the
easiest
solution
of
this
is
just
adding
a
timer.
K
You
know
and
quickly
have
some
experiments
with
adding
a
proactive
timer
after,
but
a
quarter
of
an
RTT
of
just
sending
out
kind
of
like
whatever
is
remaining,
and
it
seems
to
in
practice
like
work
pretty
well
so
I
think
I
think
this
is
a
fairly
solvable
problem.
I
mean
obviously
like.
However,
you
set
this
timer
is
a
trade-off
between
sending
it
too
early
and,
like
you
know,
closing
out
a
group
prematurely
when
there's
more
data
coming
and
waiting
so
long
that,
basically,
the
simple
is
useless.
I
I
I
think
the
the
greatest
value
in
being
able
to
apply
a
single
session
to
multiple
streams
is
probably
going
to
be
in
the
context
of
low
packet
rate
like
if
you
have
a
multiplicity
of
low
packet
rate
streams
being
able
to
apply
to
multiple
streams
and
that
context
will
be
really
valuable,
and
you
would
have
similar
issues
in
that
context
to
this.
So
it's
probably
a
trade-off
between
well.
K
Thanks
I
think:
that's
it
great!
Thank
you
and
yeah
again.
If
you
have
people
have
thoughts
on
kind
of
effective
ways
to
do
multi
recovery,
you
know
I
think
that's
certainly
I
think
there
are
a
variety
of
applications
that
would
benefit
from
it.
But
it's
that's
a
lot
of
complexity,
I'm,
not
really
sure
exactly
how
you
shoot
horn
in
without
making
this
design
quite
a
bit
less
elegant.
So
when
well
done
next
steps,
Yin.
K
So
now
that
there
are
quite
a
few,
there
are
quite
a
few
options:
I
don't
have
a
candidate
for
that,
but
certainly
in
my
case,
if
you
know
we
always
have
spots
for
interns
at
Google.
If
you
have
someone
in
mind
who
you
think
can
you
know
actually
can
go
through
the
interview
process
unscathed,
then
we
would
appreciate
a
referral
and
I'm
happy
to
host
them.
B
B
E
K
Ya
know
so:
yeah
definitely
feel
free
to
report.
People.
D
F
K
Agree,
I
would
definitely
agree.
I
think
this
is
gonna,
remain
and
kind
of
a
ongoing
work,
for
you
know
probably
a
while
to
come
until
you
know
at
least
we
have
two
or
three
implementations
on
real
applications.
I
mean
from
my
perspective,
like
what
Bart
see
is
a
great
use
case,
but
currently
Webb
Berg
see
uses
a
lot
of
little
streams
and
so
I
don't
really
understand
exactly
how
dit
like
rectify
the
fact
that,
like
you
know
that
this
single
source,
the
single
stream
issue,
it's
not
something
I'd
like
to
figure
out.
D
J
B
L
L
So
since
the
last
version
the
song
gave
us
all
the
comments,
so
we
need
to
assess
them
and
submit
another
draft,
so
we
can
on
doing
that
soon,
but
the
cutoff
date
was
passed
so
we
have
to.
We
have
to
work
on
it,
but
basically
we
still
have
to
work
on
some
figures.
We
want
to
add
research
challenges
section
before
the
conclusion,
and
also
we
may
want
to
give
couple
of
words
on
security.
L
There
are
security
aspects
in
SATCOM
that
I'm
not
sure
to
what
security
aspects
and
network
what
aspects
in
net
security
and
network
OD'ing
in
Satcon
are
very
relevant
or
I.
Think
and
more
security
and
metal
cooling
in
general
I'm,
not
sure
there
are
specific
cities
for
SATCOM,
but
we
may
want
to
propose
the
conclusions
if
there
are
any
security
issues
in
the
use
cases
proposed
in
the
draft
yeah.
J
L
E
Yes,
thank
you
Nicholas
yeah,
it's
great,
so
my
main
comment
was
about
maybe
the
lack
of
research
challenges
discussion
in
this
document.
So
if
you
can
improve
this
aspect,
that's
that's
be
very
useful
and
as
far
as
I'm
concerned,
iPhone
document
quite
easy
to
read
and
pressing
and
brings
new
well
I'm,
not
familiar
with
this
domain
and
I
learned
something
by
winning
so
I
think
it's
useful.
M
Good
morning
it's
to
card
new
participant
I'd
be
happy
to
have
some
conversations
with
you
about
the
SATCOM
security
issues
and
now
one
that
I'll
throw
out
there
right
now
is
satellite
really
facilitates
reflection
based
denial
while
distributed
denial
of
service
attacks
through
spoofing
negative
feedback.
So
that's
so
you
know
one
of
the
ones
that
can
have
catastrophic
effects,
that's
fairly
easy
to
trigger.
We
can
talk
about
that
and
other
things
actually.
B
It
would
be
great
if
you
started
a
discussion
on
the
list,
because
the
list
has
a
number
of
people
from
the
European,
Space,
Agency
and
I.
Think
John.
Are
you
a
member
of
our
list
and
so
John
is
from
Hughes
so
or
what
you
yeah.
So
there
would
be
I
think
that
could
be
a
very
good
discussion
on
the
list
and
while
I
have
the
mic.
Yes,.
L
Thank
you
thank
you
and
for
the
just,
a
quick
comment
on
the
security
I
would
be
very
happy
to
have
discussions
on
security
issues
in
Satcon
and
the
problem
with
denial
of
service
attacks
on
all
the
attacks.
It
just
I'm,
not
sure
it
is
a
scope
of
this
draft,
because
it's
more
on
metric
coding
and
security
and
that
calm.
But
if
we
can
start
the
discussion
on
security
and
SATCOM
systems
in
the
IETF
and
be
happy
to
start
here,.
E
E
N
B
E
B
B
Here
would
you
like
to
just
say
a
few
words
anyway:
I
was
I
had
talked
to
Cedric
about
it
the
he
was
supposed
to
send
me
a
few
slides.
It
did
not,
but
it's
okay.
This
is
another
one
where
there's
been
a
lot
of
progress,
and
this
is
another
one
that
we
were
ready
to
put
to
working
group
item,
but
he
does
she
do
you
want
to
comment
on
what
you've
done
recently
well,.
O
Actually,
she
said:
I
don't
know
which
research
you
I
will
have
the
task
to
decide
the
let's
say:
network
coding
and
research
group,
working
draft
or
ICN
researching
instructor
I,
don't
know
which
this
attitude
we
take
a
log
of
that
document,
but
in
any
case
the
requirement
for
network
coding
for
ICN
or
CCN
is
important
to
decide
the
solution
because
they
were
recording
for
the
ICN
is
really
one
of
the
most
important
use
cases.
So,
in
any
case,
we
definitely
need
to
have
solution
but
to
define
before
defining
the
solution.
O
B
P
P
F
So,
first
of
all,
the
last
thing
we
need
is
a
political
football
and
secondly,
these
two
groups
tend
to
have
not
had
a
problem
in
the
past
getting
along
so
I
think
in
terms
of
of
theory.
It
doesn't
really
matter
in
terms
of
practice.
I
think
I
was
the
one
who
made
the
comment
that,
if
it's
from
a
practical
standpoint
better
to
land
the
document
in
a
group
that
has
the.
F
F
F
So
I
can't
necessarily
officially
speak
for
the
other
two
co-chairs,
but
you
know
I,
don't
think
we
should
obsess
too
much
about
this
and
if
you
know,
there's
there's
energy
in
NW
c
RG
to
really
work
on
this
IC
NRG
also
has
a
pretty
not
full
plate,
but
you
know
we
have
lots
of
stuff
going
on
so
we're
you
know
we're
not
in
you
know,
grat.
You
know
we.
We
need
to
grab
this
to
have
something
to
do
so.
I'm
I'm,
happy
either
way.
B
Okay,
so
let's
go
back
to
what
we
had
at
the
beginning,
the
slides
we
were
talking
about
the
milestones
and
everything
it's
lights,
that
they'll
stop
the
slides.
At
the
beginning,
the
initial
slides
we
stopped
and
said:
let's
keep
that
for
later.
Okay,
so
we
you
know,
we've
been
doing
this
for
a
while.
So
I
look
at
some
of
the
milestones,
and
so
there
are
good
news
and
there's
less
by
there's
less
good
news.
We
would
like
to
have
a
and
again
I
hope
to
I
mean
you're
on
the
call.
B
We
would
like
to
have
a
a
new
update
to
Tetris
and
we
would
like
to
move
it
to
a
working
group
item.
We're
a
coding
group
and
we
have
very
little
coding
drafts
except
the
one
from
NASA
and
the
same
thing
for
the
Ireland
see
I
get
him.
You
guys
were
the
initial
people
for
the
formation
of
this
group
and
I
think
we
still
need
a
good
draft
moving
to
probably
working
group
item
and
RFC
on
the
nor
L
and
C,
which
again
started
this
thing.
So
I
think
we'd
like.
E
H
Is
what
it's
doing?
Okay,
so
that
that's
a
good
point,
I
think
it's
it's
a
good
addition.
It's
not
you
scope
really
specifically
of
this
draft,
but
I.
Think
it's
a
good
point.
The
what
this
draft
is
trying
to
do
is
to
basically
introduce
the
the
items
that
are
going
to
be
used
later
in
Section
three.
Basically,
there
is
a
background
section,
that's
true,
but
we
don't
want
to
expand
it.
I
mean
we
could
expand
it,
but
we
need
a
lot
of
help
doing
that
I
mean
we
are
stretched
well.
B
Okay,
so
there's
again,
I
would
actually
put
it
out
on
the
list
because,
there's
probably
you
know,
as
you
know,
murals
code
that's
been
used
by
a
ton
of
people
and
a
lot
of
these
people
are
on
this
list.
A
lot
of
people
have
worked
with
the
code.
You
know,
I
could
actually
write
it.
There's
a
lot
of
people
who
know
you
could
write
it
so
I
think
again,
I
think
you're,
not
using
the
list
to
help
you
guys
so
I
think
I
would
do
that.
B
B
H
B
Cool,
it's
the
tetris
people,
I
think,
is
the
same
comment.
You
guys
are
very
advanced,
there's
still
some
things
to
be
done,
you're,
one
of
the
authors,
so
so
it
needs
attention
so
NC
and
quitting
it.
We're
coding
and
and
quick
well
ie
and
presented.
We
have
the
initial
drafts.
There's
we
decided
to
make
two.
One
is
how
to
do
it
and
the
other
one
is
a
specific
code
doing
it
so
that's
cool
and
we
could
have
from
there.
Maybe
others
and
it's
not
problematic.
We
have,
as
we
just
said,
an
initial
implementation.
B
N
B
B
So
we
actually
just
said
that
for
the
satellite
we're
going
to
probably
make
it
into
a
working
group
item,
so
this
is
good
progress,
the
ICN
same
thing,
common
API.
We
did
it
okay
so-
and
this
is
one
that
I
was
supposed
to
do
and
in
the
past
few
months
have
been
doing
other
things
and
that
was
really
sick.
B
So
I
didn't
really
would
do
anything,
but
I
really
want
to
do
something
for
the
interim
and
the
ITF
103
on
adoption
challenges
and
or
colleague
who
asked
you
asked
questions
and
if
you
want
to
participate,
if
you're
more
than
then
welcome,
I
mentioned
where
some
of
the
adoption
challenges
were,
and
there
were
many
there's
technical
challenges.
There's
implementation
challenges,
the
availability
of
codecs
and
everything.
B
Remember
it
was
a
huge
ABBA
thing
about
one
night
in
Bangkok
probably
shows
my
age.
We
need
to
prepare,
possibly
I
think
was
a
90s
thing.
Yes,
yes,
one
night
in
Bangkok,
the
by
the
way,
the
video
is
fantastic,
I
think
it's
the
only
thing
of
the
whole
chest
that
is
worth
watching
it's
one
night
in
Bangkok
and
it's
from
a
guy
from
Montreal.
By
the
way
you
wrote
the
song
anyway
hackathon
for
IETF
103.
B
We
are
starting
to
think
about
it
because
of
the
API
and
I
think
we're
getting
a
few
people
excited
about
doing
that.
We
know
that
it's
a
lot
of
work
but
I
think
we're
ready
to
do
it
and
the
interim
again
it's
interesting
because
at
the
energy
and
Tuesday
they
announced
that
we
were
going
to
have
a
interim
meeting
which
was
cool.
B
It
was
suggested
to
be
on
the
25th
in
Boston,
because
a
lot
of
the
people
of
the
IC
energy
who
are
interested
in
coding
are
going
to
be
in
Boston,
because
it's
right
after
this
ICN
conference-
and
it
could
also
be
totally
virtual.
It
could
also
be
because
of
other
issues
and
also
to
maybe
get
you
know.
We
did
it
in
Boston
last
year,
we're
thinking.
Maybe
we
should
do
it
in
Europe.
B
It's
not
a
bad
time
to
go
for
people
who
want
to
go
and
also
there's
a
there's,
a
critical
mass
of
network
coding,
people
in
France
in
Toulouse
and
India
and
other
places,
and
some
people
also
in
Germany.
So
we're
going
to
ask
the
list.
It
would
then
be
probably
the
26
so
that
the
people
who
are
in
Boston
have
time
to
go
back,
including
me,
and
what
we're
going
to
put
it
on
the
list
and
the
reason
that
we
want
to
have
a
people
always
say.
Would
you
like
to
have
an
interim?
B
B
Q
M
C
B
Yeah,
okay:
well,
you
make
a
point.
The
goal
here
is
not
to
have
a
transport.
We
talked
about
it.
Okay,
so
this
been
a
discussion
about.
Do
we
want
a
network
coded
transport
and
we've
discussed
this
in
the
past
extensively
and
the
quite
the
reason
that
we
did
not
do
it
exactly
because
of
what
you're
doing
and
also
because
it
is
so
easy
to
not
use
it.
You
know
you
can
have
all
the
new
transport
you
want,
but
nobody
uses
it
its
uses.
B
So
the
question
a
the
issue
here
is
not
that
is
we
want
the
network
coding
and
the
transport
to
be
fairly
independent,
or
so
it's
not
because
you
use
TCP
X
that
you're
not
going
to
be
able
to
use
network
coding
it.
Those
two
things
are
different.
However,
the
fact
that
you
use
network
coding
because
you're
hiding
the
losses,
then
the
whole
TCP
congestion
control
starts.
B
And
the
goal
here
again
is
not
to
propose
a
new
transport
I'm,
actually,
first
I
think
it's
really
hard
to
be
accepted,
and
second
I
always
think
that
somebody
will
go
back
into
the
Linux
stack
and
change
the
transport
just
because
they
don't
know
what
this
new
transports
about
and
I
know.
My
applications
works
better
with
that
version
and
something
we
are
going
to
use
it
so
so
I
get
your
point
on.
F
So
I
think
we
may
have
worked
the
discussion
from
do.
We
need
documents
that
will
get
people
more
excited
about
adoption
and
more
comfortable
with
adoption
and
I
think
we
just
confounded
it
with
do.
We
need
to
have
a
network
heard
of
Transport
and
yeah,
so
one
comment
I'll
make
that
we
think
in
supporters
of
I
think
what
Stu
is
getting
at
here
with
people
coming
from
industries.
We
had
this
horrible
experience
with
parity
codes
for
real
time
that
simply
got
ahead
of
the
curve
and
they
basically
published
something
that
people
actually
started.
F
F
We
actually
want
to
get
stuff
out
there,
but
and
but
we
also
need
to
know
if
somebody
else
is
getting
something
out
there
around
the
side
and
make
sure
that
we
don't
have
to
like
affect
frame,
wound
up,
having
to
take
a
big
hammer
and
redo
everything
that
simply
did
and
then
DVB
actually
was
the
one
that
that
went
to
synthy
and
say
no,
no,
no,
no,
no
and
then
DVB
had
more
international
power
than
cynthy.
So
we
finally
sorted
it
all
out,
but
it
was.
It
was
really
unfortunate.
B
There's
also
there's
another
thing
I
would
like
to
add,
and
so
maybe
we
can
take
that
to
the
list.
Also
again,
when
our
colleague
earlier
had
talked
about
the
fact
that
there
was
known
that
were
coding
that
was
working
in
real
networks,
we
all
we
are
all
under
NDA
with
I,
don't
know
how
many
companies
who
have
actually
done
it,
and
maybe
it
would
be
interesting
for
these
companies
to
come
forward
and
say
this
is
our
experience.
B
I
know
sometimes
there's
question
of
competitive
advantages
of
having
these
solutions
and
their
network
because
it
gives
them
in
advance
to
be
better
than
the
others.
There's
all
issues
of
obviously
issues
by
PR
issues
of
everything,
but
I
think
this
is
something
we
should
put
on
the
list,
a
discussion
about
who
has
done
this
in
a
real
network
and
would
like
to
come
forward
and
talk
about
the
issues
they
had.
Was
it
in
terms
of
development?
Was
it
in
terms
of
wealth?
B
We
all
think
that
the
performance
improved,
but
I
would
cost
so
it
was,
was
the
was
it
a
hit
on
their
time
of
development?
Was
it
a
hit
on
the
time
to
market
there's
all
kinds
of
things
like
that?
Yes,
I
know
an
industry,
it's
always
for
yesterday.
So
if
you
come
with
I,
have
this
great
idea
we'll
improve
performance
by
25-30
percent,
but
it
will
take
two
years,
I'm
sure
nobody's
going
to
even
listen
to
me
for
a
while.
B
It
will
Center
me
and
one
year
and
a
half
plus
three
months
when
it's
going
to.
Thank
you
so
I
think
it's
all
issues
like
that
that
I
want
to
put
in
the
draft,
but
also
I,
would
like
maybe
yet
the
discussion
on
the
list
to
say
this
is
what
we
think
are
the
issues
and
it's
been
an
ongoing
thing
about
the
congestion
control.
It's
been
an
ongoing
thing
about.
You
know
the
fact
that
we're
looking
at
this
open
that
hackathon
for
the
open,
API
implementation.
B
B
Google
well
I
talked
to
Ian
about
yeah
about
doing
some
internship.
Well,
you
could
order
Akamai
to
it's
actually
their
neighbors,
as
you
view,
maybe
you're
interested
in
being
grinded
other
questions.