►
From YouTube: IETF103-NWCRG-20181105-0900
Description
NWCRG meeting session at IETF103
2018/11/05 0900
https://datatracker.ietf.org/meeting/103/proceedings/
B
What's
up
Ramon,
hello,
everybody?
This
is
the
coding
for
efficient
network
communications
research
group
in
order
to
accommodate
somebody
who's
on
a
very
short
schedule
in
Hong
Kong
we're
just
going
to
go
through
mineral
introduction
and
we're
going
to
go
directly
into
the
first
presentation,
and
then
we
will
finish
whatever
administrivia
and
other
presentations.
We
have
I'd
like
to
thank
Nicolas
for
having
nicely
accepted,
to
become
the
volunteer
to
take
some
notes.
It's
greatly
appreciated
and
without
any
further
delay.
B
Again.
This
is
the
group
that
does
research
into
network
and
application
layer
coding
to
improve
network
performance.
We've
been
going
on
for
a
number
of
years,
and
for
those
of
you
were
there
at
the
beginning,
you
can
see
that
we
have
a
major
move
from
just
looking
at
codes
and
coding
library,
but
also
for
protocols
that
facilitate
the
use
of
coding
and
existing
systems
that
all
the
research
around
this,
as
well
as
the
researcher
is
going
on
in
many
other
fields
that
originally
didn't
have
anything
to
do
with
coding.
But
I.
B
So
this
is
the
note
well
something
that
we
always
have
to
present
to
make
sure
that
everybody
recognizes
issues
of
IPR
and
everything
we
have
50
new
people
who
have
never
been
that
idea-
and
this
is
IRT.
F
follows
the
same
rules
as
the
IETF
in
terms
of
recognition
of
the
processes,
recognition
that
the
IPR
recognition
of
how
to
behave
in
a
meeting
like
I
said
we're
just
going
to
go
directly
to
the
presentation
from
Raymond,
because
he
has
a
very
tight
teaching
schedule
today,
Bremen
I'm,
going
to
give
you
the
speaking
part
now.
C
C
Okay,
how
about
40
now
the
idea
is
very
simple:
you
just
put
a
4G
card
at
each
lamppost,
converting
it
into
something
like
like
a
cell
phone.
Now
the
pros
is
that
using
a
4G
card
is
that
it's
very
easy
to
deploy
and
it's
relatively
inexpensive,
because
the
cons
are,
it
has
a
pretty
high
recurrent
cost.
You
mentioned
that
you
have
a
HD
webcam,
generating
like
5
megabits
per
second
of
data
and
for
the
whole
month,
you're
talking
about
over
10
terabytes
per
month.
C
C
In
this
talk,
I'm
introducing
the
multi-hop
solution
in
this
picture
we
have
a
sequence
of
lampost
for
the
first
one,
on
the
left
being
connected
to
the
optical
fiber
directly
and
for
the
rest
of
the
lamppost.
They
are
connected
through
a
wireless
multi-hop
networks
and
which
is
enabled
by
bypass
code,
which
is
an
acronym
that
we
developed
throughout
the
last
year.
Last
few
years,.
C
Okay,
why
why
do
we
consider
paths
now
we
know
that
multi-hop
is
long
standing
following
wireless
communication
in
industry
and
transmission
can
sustain
no
more
than
a
few
hops.
If
you
go
to
a
few,
a
few
more
hops
than
the
the
at
the
throughput.
Basic
risks
are
quite
negligible,
and
sometimes
in
industry.
People
refer
to
this
as
the
multi-hop
curse.
Best
code
is
a
as
advanced
Nemacolin
technology
that
can
sustain
tens
or
even
hundreds
of
hops,
and
this
is
done
by
recoding
as
the
intermediate
nodes
and
with
that's
a
very
long.
C
The
last
year,
my
collaborator
Shin
Gyung
and
myself
are
published
a
monograph
on
pass
codes
in
the
synthesis
lecture
on
communication
networks
edited
by
a
three-count.
Next,
please.
C
Here's
this
group
comparison
between
bets
and
routing.
Now
here
we
are
talking
about
a
lying
network
with
20%
loss
on
each
20%,
take
a
loss
on
each
each
link
so
that
after
one
heart
you
you
lose
20%
and
a
new
maximum
rate
that
you
can
ever
get
is
0.8.
Okay,
the
the
red
curve
shows
the
throughput
of
routing
versus
the
number
of
hops,
and
if
we
see
that
is
decreasing
exponentially
fast,
so,
for
example,
after
10
hops,
you
only
get
about
10%
of
the
of
the
of
the
rate,
okay
and
with
pass
code.
C
C
So
essentially,
one
can
think
of
a
pass
code,
converting
a
lying
that
multi-hop
line
network
into
a
into
a
single
hop
Network.
The
advantages
of
best
code
are
the
following
in
as
high
throughput,
low
latency,
low
coding
complexity
and
low
storage
requirement
for
the
last
aspect.
I'm
talking
about
that,
you
need
to
store
only
a
in
to
keep
a
very
small
buffer
at
each
intermediate
node
and
for
all
existing
solutions.
C
D
Thank
you
I
think
Wishman
Rick
Tina
Airbus,
first
off.
Thank
you
very
much
for
this
presentation.
This
is
really
interesting.
Technology.
I
totally
agree
with
your
use
cases
that
just
sticking
4G
or
fiber
optic
to
every
lamppost
is
madness.
I
have
a
question
about
I'm,
afraid,
I,
haven't
read
all
this
in
depth
and
I
need
to
go
away
and
do
it.
But
the
picture
you
have
on
the
on
the
on
the
slide
shown
at
the
moment
shows
a
single
linear
topology
for
your
multi-hop.
D
Have
you
done
much
research
into
how
the
topology,
so
we
obviously
know
lampposts,
although
down
one
Street
are
in
a
single
line,
but
when
you
come
to
intersections,
you
start
to
get
start
upon
a
G's.
If
you're
one
lamppost
with
a
fiber
optic,
you
know
backhaul,
is
that
the
is
at
the
center
of
that
star.
How
does
that
scale
to
look
at
that
and
have
you
done
that
research
yet
well,.
C
D
C
D
E
C
E
E
C
E
C
G
H
Devoran
no
existence,
research
and
design
on
to
related
questions
about
your
performance
graph
curve.
Could
you
put
it
back
up?
Okay,
well,
I
can
answer
it
while
you're
getting
it
up.
Does
that
graph
include
or
not
include
hop-by-hop
retransmission.
C
C
H
Does
the
paper
have
all
this
in
it
because
I
mean
it's
useful,
you
know
you've
sort
of
set
up
an
AR.
You
know
best
case
comparison
right,
which
is,
you
know,
high
loss
with
no
hop
by
hop
recovery.
Of
course,
this
is
going
to
do
a
lot
better.
It
would
be
nice
to
show
it
also
does
a
lot
better
when
you
use
a
more
conventional
layer,
2
protocol
scheme
I'm
not
I'm,
not
necessarily
saying
that
doesn't
do
a
lot
better.
It
would
be
really
nice
if
you
could
show
that
yeah.
G
B
B
G
I
wanted
to
say
a
few
words
about
it,
so
you've
noticed
that
we
switch
to
Gita
for
all
the
research
material,
so
I
mention
already
mention
you
in
the
mailing
lists.
This
so
do
not
hesitate.
You
are
looking
this
github
repo.
You
will
find
many
different
things.
So,
first
of
all,
as
I
said,
all
the
recent
good
metal
will
be
posted
there
for
the
various
meetings,
but
as
well
as
interim
meeting,
so
you
will
find
whatever
we're
talking
about
in
this
repo,
but
you
also
have
additional
repositories
in
this
ayah
TF
n
w
mg
organization.
G
You
will
find
the
suite
correct,
so
I
will
talk
a
little
bit
buddies
later
on
for
the
Akutan
project,
but
you
will
also
find
two
additional
repo
for
the
moments
for
the
two
results
could
be
item
documents.
So
this
is
the
place
where
you
can
find
the
latest
version,
whether
it
has
been
committed
or
not
and
uploaded
or
not
to
the
data
tracker.
So
if
you
want
to
have
a
look
at
the
current
version,
ongoing
Russian
have
a
look
at
this.
G
This
is
the
best
place
for
that
and
as
new
Richard
group
eaten
documents
are
accepting,
we
will
create
new
repository
for
each
of
them.
With
a
team
composed
of
the
offers,
we
will
have
full
access
to
their
own
repository
so
as
to
be
able
to
update
and
the
work
from
their
documents,
so
unity
state.
This
is
best
place
to
know
what's
going
on
his
group,
in
addition
to
the
data
tracker.
That
of
course
remains
something
important
for
our
such
group
like
working
groups
of
fight.
Yet,
but
we
are
both
and
we're
working
with.
F
B
B
That
I
saw,
as
put
actually
it
is
on.
That's
we
forget
to
say
this
about
the
github.
Thanks
I
put
a
number
of
I.
We
called
it
FEC
for
dummies,
but
I,
don't
think
it's
dummies
I
think
it's
for
more
for
newcomers,
but
anyway
I.
Don't
think
people
are
done
because
I
don't
know
about
it,
but
anyway
and
there's
a
lot
of
things
that
explains
with
network
coding
is
and
with
network
coding
can
do
for
you
and
I.
B
Thank
all
the
people
who
commented
on
this
and
sent
us
like
comments
and
improvements
to
this
performance
evaluation
is
something
that
we've
talked
about
in
the
past,
and
thanks
was
going
to
mention
it.
We
have
to
talk
about
the
milestones
like
I
said
for
people
who
have
not
been
in
this
meeting
before
this
group
has
been
on
for
a
while.
B
So
we
have
the
items
right
now:
the
ideas
that
are
the
research
group
items.
We
have
the
network,
coding
and
satellite
that
again,
Nicholas
is
going
to
do
an
update
on
and
the
network
coding
for,
content
centric,
which
is
also
a
r2
document,
and
we
are
also
sharing
some
of
the
approval
of
this
document.
With
our
colleagues
in
the
IC
NRG,
we
have
a
individual
ideas.
We
have
a
new
idea
on
facts
coding.
We
have
the
generic
API.
We
have.
B
The
coding
for
quick
coding
for
quick
has
no
progress,
because
the
quick
guys
are
so
busy
putting
things
together.
What
we
intend
to
update
it
so
that
you
won't
disappear
and
detectors
in
RL
NC,
who
straight
enough
where
the
drafts
that
started
this
group
have
now
expired,
and
this
is
something
that
we
have.
We
would
like
the
teams
that
were
behind
that
to
really
consider
restarting
it
as
we
had
docked
and
bidding
to
have
the
beginning
of
this
to
have
one
draft
per
type
of
code
that
was
going
to
be
able
to
be
used
next.
B
Our
future
meeting
are,
we
intend
to
meet
in
Prague
in
actually
after
the
winter
I
guess
it's
early
spring
and
we
intend
to
have
the
market
on
and
I
think
we're.
This
is
the
next
step,
the
next
step
presentation,
but
I
think
we
encourage
people
to
show
up
at
the
hackathon
I
couldn't
be
there
on
Saturday,
but
I
think
the
next
hackathon
one
of
the
goals
that
I
would
have
for
Prague
is
starting
to
write,
use
cases
to
test
the
codec,
and
so
people
who
know
about
Python
and
Python
use
cases.
G
G
That's
much
better
like
that,
so
a
quick
update,
so
the
goals
of
this
Acton
was
to
design
Swift
correct,
switch,
but
explaining
window
fec
correct
in
order
to
facilitate
tests
and
experiments
by
anybody
with
potentially
interested
in
this
technology
and
operated
easy
to
facilitate
the
adoption
technology
itself,
as
well
as
being
a
reference
open
source
implementation,
free,
open
source
plantation
of
these
of
those
solutions.
So
that's
the
first
to
go
of
this
accattone.
The
second
goal
is
to
challenge
our
API
dynamic
API
in
turn.
G
Draft
and
I
will
talk
about
it
later
later
on
and
also
it
was
a
good
opportunity
for
me,
because
there
was
a
newcomer
at
this
Acton
project.
In
fact,
I
was
wrong,
but
I
was
I,
met
quasi,
who
was
dressed
yeah
sitting
here,
and
we
worked
together
on
this
project,
and
since
these
you
bring
new
people
to
ITF.
It
was
a
good
opportunity
for
us
to
for
me
to
test,
and
this
fe
CN
n
c
for
the
means
or
new
comers
call
it
whatever
you
want
and
see.
G
If
it's
appropriate
and
the
feedback
is
positive,
I
would
say
on
this.
If
you
disagree,
do
not
hesitate
to
interrupt
me
and
shop,
but
I
think
okay,
so
those
are
the
goals
we,
of
course,
we
didn't
finish.
It's
just
the
beginning
of
this
project
for
sure,
but
we
already
fixed
a
few
issues
within
this
generic
Fe
C
API,
intern
draft
and
I'm
updating
the
document
yesterday
evening,
just
after
the
accident
take
these
into
account.
So
if
you
have
a
look
at
this
new
version,
you
will
see
many
different
things.
G
I've
reorganized
the
thing
good
some
functions
in
different
way
in
order
to
facilitate
the
completion
of
the
full
staff
fixed,
a
few
people's
removed
one
or
two
functions
and
I
did
another
one.
So
those
are
things
that
you
find
quite
useful
and
necessary
when
you're
working
with
code
at
the
same
time
as
working
on
the
documents,
so
that
that
was
a
very
good
opportunity
to
do
this
work
and
to
add
something
back
on
this
interactive.
Even
if
this
is
something
I
wrote
myself
and
I
was
the
one
doing
the
most
of
the
correct
stuff.
G
Also
at
the
same
time,
even
that
case
was
useful.
So
that's
the
first
tremendous
common
one
is
okay.
We
now
have
this
repository
for
this
suite
correct
and
we
started
to
put
many
different
things
inside,
so
we
reuse
some
existing
code,
either
from
open
factors
or
project
that
I
used
to
work
on
in
the
past,
but
also
come.
F
G
Colleague,
sera
congi,
so
you
you
already,
you
can
really
find
some
interesting,
not
working
code,
but
some
interesting
one
inside
this
github
repository
we've
been.
We
focused
on
the
encoder,
but
first
of
all
for
two
reasons.
First
of
all,
it's
necessary
in
order
to
be
able
to
test
the
full
scheme,
full
solution.
You
need
to
first
welcome
me
and
coda,
and
also
because
it's
a
little
bit
simpler,
you
just
have
to
compute
linear
combinations
of
packets
to
be
able
to
do
encoding.
G
So
it's
something
which
is
quite
simpler
than
the
decoder
which
has
to
deal
with
this
linear
systems
resolution
all
the
stuff.
So
that
was
the
first
goal.
It
still
does
not
work,
but
we
are
not
that
far
from
having
a
solution.
Of
course,
we
focused
on
some
the
most
simplest
assumptions,
one
single
encoder.
For
the
moment
we
will
extend
to
any
network
recording
operations
later
on
in
this
project.
G
G
G
Ok,
so
let's
focus
on
this
generic
FEC,
API
and
documents.
So
this
is
version
offer
that
I
updated
yesterday
evening,
as
I
mentioned,
I
moved
a
few
things
in
within
this
intern.
Draft
I
already
went
through
this
those
modifications,
but
you
can
have
a
look
at
the
document
itself
on
the
dish
with
the
first
version,
you
will
see
that
many
things
moved
now.
I
think
that
we
reach
a
point
where
this
document
is
not
stable,
but
in
a
good
shape,
well,
good
enough
shape
to
be
accepted
as
a
result.
G
Group
item,
if
you
believe
this
is
something
useful
for
this
result
group.
So
I
would
like
to
formally
ask
for
recent
reproduction
for
these
documents,
so
that
can
make
progress
and
in
fact
we
will
add
the
two
activities
in
parallel.
At
the
same
time,
this
academic
project,
which
is
a
good
opportunity
to
check
and
challenge
this
inside
draft
and
the,
of
course,
the
generic
API
instant
draft
that
will
explain
how
to
interact
with
this
correct.
G
H
They
I
don't
see
any
objection
to
adopting
as
a
working
group
as
a
research
group
item.
What
one
of
the
things
I
think
is
a
precondition
to
really
getting
it
to
move
significantly
forward
is
that
there
be
at
least
two
different
FEC
codecs
supported
by
that
API.
Otherwise,
you
can't
tell
whether
you
have
a
generic
API
or
not
or
whether
they
have
an
API
that
only
works
with
one
codec,
so
we
need
to
somehow
get
in
the
work
plan.
H
G
Two
codecs,
yes,
so
a
quick
answer.
Yes,
of
course,
this
is
one
of
the
goals
of
this
document
to
have
something
generic
that
can
be
used
by
several
different
types
of
physicals.
I
have
said
during
previous
meetings.
We
are
focusing
on
sliding
window
codes,
but
not
only
one
solution,
but
hopefully
all
potential
sliding
window
solutions
with
end
to
end
Kareem,
but
also
with
in-network
recording,
as
in
networking
systems
like
in
bad
schools.
So
we
are
really.
We
really
want
to
address
all
those
the
use
cases,
all
those
different
cuts,
Ric.
D
Taylor
follow-up
to
that
absolute
plus
one
and
can
I
recommend
I,
give
you
sort
of
answered
this
already,
but
that
you
pick
to
make
sure
your
API
is
generic
enough.
You
pick
two
contrasting
if
you
see
income,
so
a
sliding
window
and
maybe
you're
trying
to
pick
something
else
random,
but
at
the
other
end
of
the
spectrum,
to
make
sure
that
your
API
is
is
good
enough.
Just.
G
G
G
G
G
G
We
adoption
founders
lots
of
comments.
There
were
lots
of
harmonization
issues
in
the
figures,
so
we
try
to
polish
all
these
presentations
figures
and
also
had
lots
of
grammar
mistakes
and
that's
something
we
hope
we
have
improved.
Also
the
way
from
wasn't
clear
from
the
Willing.
What
is
a
crew
infectious
agents
at
comm
systems?
So
we
just
try
to
make
it
clearer
by
adding
one
or
two
sentences.
We
hope
it's
clear
enough,
but
don't
not
agitate.
Let
us
note
it
is
not.
G
We
can
point
to
more
dvb-s
to
solids,
for
example,
but
basically
we
try
to
make
it
clear,
also
once
opposite
of
text
for
the
multicast
sections,
we
have
tried
to
include
all
that
and
thank
Universal
such
as
proposed
by
evensong.
We
added
research
challenge
section,
so
we
somehow
cheated
because
we
moved
the
conclusion
to
we
change
the
conclusion
to
a
research
research,
science
section.
Basically,
we
have
identified
two
main
research
challenge
that
are
specifically
related
to
succumb
and
network
coning.
G
First
one
being,
how
do
you
make
it
deployable
today
in
the
systems
we
have,
because
that
is
one
of
the
goal
we
had
in
the
document
we
identified
use
cases
and
where
we
could
have
agreed
that
were
coding
in
satellites,
but
there's
always
a
question
how
you
get
it
deployed.
So
we
think
that
this
should
be
could
be
doable
in
TCP
proxies.
We
have
because
then
we
split
TCP
connections.
G
We
are
at
the
user
space
that
also
you
can
do
lots
of
funny
things
and
we
have
fun
geology
same
thing
at
the
terminal
side,
there's
also
a
challenge
that
is
already
inherent
to
satellite
systems
in
how
efficient
we
are,
as
opposed
to
a
much
with
gradual
resource.
We
use
that's
eternal
debate
and
trade-off
that
is
going
on
in
SATCOM
industry,
and
this
is
even
more
relevant
for
network
coding
in
high
layers
because
we
don't
because
wider
resource
is
cast.
G
So
we
have
to
save
it,
so
that
is
something
that
really
always
be
open
and
we
kind
of
tackle
it
in
generic
way
and
just
depending
on
the
systems
you
have
and
the
objective
of
in
terms
of
latency
and
loss
tolerance
of
the
applications.
You
have
different
solutions,
another
thing
and
now
that
energy
working
group
is
over
and
I
think
one
of
the
thing
that
we
had
discussed
and
I
think
in
Alps
also
the
API
document.
G
G
Network
coding
turns
to
be
vnf.
You
could
easily
deploy
that
would
be
very
useful
for
us
and
that
would
define
meant
of
the
network
audience
attempted
systems,
something
also
that
we
did.
We
I
think
we
kept
me
may
had
some
content
to.
It
is
a
Security
section
because
basically-
and
there
are
lots
of
SATCOM,
specific
security
aspects,
but
I
think
the
goal
of
document
is
not
to
tacko
satellite
system
security
issues.
G
More
death,
Network
coding
has
any
specific
issues
with
security
in
certain
systems,
so
we
precise
that
in
the
draft
I
mean
we
can
make
some.
We
have
had
some
since
the
document
was
released
as
working
up
item.
We
had
some
feedbacks
from
the
internal
from
people
from
our
own
company,
so
we
may
want
to
add
some
more
details
on
sat-comm
system
security.
B
D
A
really
good
DTN
section
in
this
document
I
had
a
quick
look.
It's
it's
really
good.
Could
I
ask
you
to
update
some
of
your
references
to
the
IETF
RFC,
somehow
that
draft
still
that
are
going
forwards
rather
than
the
IRT
F
stuff?
And
if
you
need
some
volunteers,
can
we're
meeting
on
Thursday
morning
put
a
copy
of
this
onto
the
mailing
list.
I'm
sure
there
are
people
there.
You
know
we
all
work
in
satellites
in
DTM
as
well,
so
I'm
sure
there
are
people
with
lots
of
industrial
experience.
G
G
G
G
I
From
an
IC
k,
so
let
me
talk
about
quick
status.
The
tracker
is
our
network
coding
for
shishun
Indian
requirement
and
challenges,
so
the
main
objective
of
this
route
is
to
provide
a
sufficient
equation
and
the
clarification
requirements
and
challenges
regarding
foreign,
genetic
coding
into
Chien
and
any
CNN
Vienna.
So
in
order
to
you
know
that
provider
establish
common
common
understanding
and
hopefully
to
provide
useful
insights
to
develop
a
through
a
pry
and
she
in
decision.
I
So
well
year
ago
we
presented
our
initial
draft,
both
network
holding
research
group
and
I
shine
research
group
and
the
in
the
previous
meeting.
It
was
discussed
regarding
the
the
adoption
draft
as
a
research
study
group
draft.
So
then,
thanks
to
the
chairs
and
already
had
the
balls,
Network
coding,
misogyny
panda
energy.
This
draft
has
now
become.
I
So
here's
a
current
structure
of
this
draft,
so
this
is
a
group,
so
we
I
think
we
need
to
I
think
it's
important
to
introduce
context
and
semantics
of
Chien
in
the
section
one
and
the
two.
So
in
section
one
we
introduced
both
network
coding
and
decision
over
B
and
in
Section
C.
We
provide
the
incision
equivalent,
so
so
the
current
way
in
this
part,
we
we
provide
in
section
3.
We
provide
the
advantage
given
by
NEC
on
decision
in
Vienna.
F
I
After
that,
we
describe
why
the
requirement-
and
we
describe
the
research
potential
research
challenges,
so
we
we
now
cover
the
requirement
for
offering
energy
into
CM.
So
one
is
the
naming
and
the
others
are
transport
in
networking
and
seamless
mobility
and
security
and
the
privacy.
So
in
this
section
we
we've
already
modified
based
on
the
obtained
comment.
I
So
far,
so,
for
instance,
we
add
the
relationship
between
the
coding
information
head
down
the
security
envelope
and
then
we
add
design
choice
who
determines
the
encoding
vector
and
we
describe
its
impact
on
ratings
here
and
we
were
already
updated
regarding
some
editorial
updates.
So
as
a
potential
research
challenge,
we
will
now
cover
these
four
points.
Other
thing,
compulsion,
coding,
region
and
congestion,
control
and
the
security
and
privacy
and
the
rating
is
rooting
scared
of
it
here.
I
So
in
this
draft,
there's
no
description
about
a
specific
protocol
solution,
because
it's
out
of
the
scope
of
this
net,
so
actual
protocol
proposal
will
be
indicated
indicated
in
another
draft.
So,
as
a
next
step,
we
consider
heater
update
is
if
there
are
other
other
requirements
to
be
considered,
we'll
add
in
section
3
and
we
we
plan
to
refine
the
potential
challenge,
especially
about
a
cigarette
gear,
looting
in
section
3
and
if
any
we
had
important
potential
changes.
I
G
I
B
H
This
is
an
IC
energy
chair,
Hat
comment,
not
a
personal
comment,
which
is
that
we're
about
to
get
the
CCN
drafts
published
as
experimental
RFC's.
It
would
be
really
helpful
as
you
do
this
work.
If
you
discover
changes,
you
think
ought
to
be
made
in
the
CCN
protocols
in
order
to
make
this
work
better.
Just
remember
you
don't
need
to
treat
the
CCN
protocols
as
cast
in
stone
right.
We
can
change
them.
J
Mr.
sidearm
so
commentary
I
think
we
should
not
ask
to
modify
any
CC
NX
format,
which
is
already
in
a
stage
for
the
just
before
the
LOC.
So
but
Observatory
we
need
to
add
some
packet
format
change
or
we
need
to
change
something,
but
that
one
is
can
be
categorized
as
something
like.
A
extension
of
the
basic
packet
former,
so
I,
say
notice
doesn't
need
to
care
so
much
at
this
moment
for
the
formalizing,
the
CNX
message,
Walmart
or
semantics,
but
in
your
future.
J
G
G
So
a
few
words
about
this
document,
efficiency
for
all
newcomers
will
say,
but
I
I,
like
dummies,
also
because
this
is
a
good
theory
anyway.
So
just
to
remember
you,
okay,
we
mentioned
during
previous
I,
kept
meeting
the
fact
that
was
important.
You
have
some
quick
and
easy
introduction
into
networking
FEC
to
make
people
believe
that
they
understand
what
is
behind
so
that
they
can
more
easily
accept
the
technology
and
the
very
important
force
for
everybody.
I
would
say
so.
G
It
was
very
good
comment,
so
we
started
thinking
about
it
and
trying
to
put
a
few
IDs
kept
as
simple
as
possible,
but
at
least
the
goal
put
together
some
ideas,
simple
ideas
on
what
we
want
to
address.
What
is
the
background?
The
minimum
knowledge
you
need
to
have
you
know
that
to
be
able
to
understand
more
or
less
what
we're
talking
about.
So
that's
an
attempt
I
sent
on
the
email
only
mainly
sorry,
an
email
two
weeks
ago
on
the
topic
with
draft
of
IDs.
G
So
what
I'm
presenting
now
is
in
line
with
what
I
sent
only
lists
with
a
few
modifications
based
on
comments
that
I
receive
I
received
two
comments
to
the
comments.
People
were
in
tracing
this
topic,
but
I
received
two
suggestion
for
modifications
improvements.
So
it's
in
line
is
that
it's
small,
that's
a
requirement,
so
I
have
eight
IDs,
so
the
first
one
is
to
stress
the
fact
that
we
are
working
on
packet
loss
networks,
so
networks
where
packet
will
be
lost,
perhaps
or
if
they
arrived
to
the
destination.
G
G
Consent
to
explain
what
encoding
means
and
what
decoding
means
so,
when
coding
means
adding
redundancy
in
the
packet
flow,
I
mean
one
or
more
repair.
Packets
and
decoding
means
using
this
redundancy,
those
additional
repair
kits
to
be
able
to
recover
from
packet
losses.
I
think
it's
quite
easy
to
understand.
The
third
idea
is,
okay
is
to
tell
people
that
they
don't
necessarily
need
to
understand
what
is
behind
the
technology.
What
is
mathematical
aspects,
of
course
we
are
working
on
finish
fields,
so
we,
if
you
want
to
perfectly
understand
what
we
are
talking
about.
G
We
also
need
to
know
what
is
a
finite
field
or
with
the
multiplication
of
all
these
mathematical
paths.
But
if
you
just
want
to
have
a
row
idea,
general
idea
of
what
FEC
and
networking
means
you
don't
necessarily
need
to
add
this
knowledge.
It's
not
required
at
all.
Just
know
that
you
have
a
linear
system,
it's
just
mad.
The
coding
is
just
matter
of
solving
a
linear
system,
and
economy
is
just
matter
of
computing,
a
linear
combination.
So
that's
those
are
two
basic.
G
Aspects
that
everybody
can
understand
Paraiso,
the
fourth
ID-
is
about
a
physical
store.
Roughly
roughly
speaking,
there
are
two
categories
of
a
physical.
One
of
them
is
block
codes,
the
other
one
is
sliding
window
cuts,
so
block
codes
with
block
codes.
You
had
this
flow
of
data
packets.
You
cut
this
rule
into
among
these
packet
flow
into
blocks
and
you
do
efficient
coding
on
each
block
independently.
So
that's
for
block
codes
and
forth
landing
window
codes.
G
Once
again,
if
you
have
comments
about
those
ie,
if
you
believe
that
they
are
not
put
in
the
right
way,
if
we
can
add
something
that
is
simpler,
more
easily
constable
or
if
you
have
suggestions,
do
not
hesitate.
The
goal
here
is
not
to
tell
you
what
we
are.
What,
if
not
to
teach
you
what
is
networking
and
50
it's
just
to
get
essentially
to
get
your
feedback
so.
D
Rick
Taylor
again
sorry
I'm
talking
a
lot
just.
Can
you
go
back
to
idea
to
slide
well
I
completely
agree
with
you.
Apart
from
you,
use
the
word
repair
which,
if
you
go
to
idea
one
you
said
you
know
we're
not
a
filer,
we
don't
care
about
bit
error.
You
know
we're
assuming
our
packets
like
that
arrived
or
don't.
The
word
repair
implies
that
you
are
fixing
broken
packets
rather
than
I.
Think
use
the
word
recover
correctly.
Can
you
call
them
recover
packets?
You
know
it's
you're,
not
fixing.
G
G
Okay,
so
the
fifth
ID
is
okay,
so
what
type
of
code
for
what
type
of
use
case
so,
basically,
very
basically
speaking
records
I
create
for
all
situations
when
you
don't
have
so
much
real
time
requirements.
So
five
transfer-
that
kind
of
this
case
is
great
in
for
those
kind
of
use
cases
using
broken
is
not
an
issue
usually
and
slagging.
We
look
at
some
much
better
for
real-time
traffic.
That's
the
general
idea!
Of
course
you
can
find
exceptions.
You
will.
G
H
First
of
all,
a
block
code
with
very
small
blocks
is
going
to
work
way
better
than
a
sliding
window
code
with
very
small
blocks.
So
if
you
have
lots
of
small
objects
that
are
independent
and
you're
spraying
them
out
on
the
network,
a
sliding
window
code
is
actually
going
to
enforce
certain
types
of
ordering
properties
on
the
data
and
actually
slow
things
down.
So
we
have
to
find
another
way
to
say
this.
I
think
I
know
what
you're
getting
at
right,
which
is,
if
you
have
large
bulk
data
block
codes,
are
win
yeah.
G
H
G
B
G
Good
point
next
ID
ID
six
is
about.
Yes,
there
are
two
situations.
Basically,
if
you
see
a
net
working
with
the
GC,
we
had
a
single
encoder
and
civil
decora
and
with
networking
the
way
we
understand
it,
we
have
the
possibility
to
do
we
incur
in
within
the
intermediate
nodes.
So
basically
you
have
multiple
encoders.
G
Next
ID
is
seven
well.
Okay,
there
are
situations
where
network
coding
can
help
improving
network
usage,
and
this
is
a
very
simple
example
with
Alice
and
Bob.
Both
of
them
want
to
send
packet
to
the
other
one,
and
you
have
this
intermediate
routing
equipment,
the
Wi-Fi
water.
This
Wi-Fi
widow
is
if
first
packet
p1
from
a
listeners
like
@pq
from
Bob
and
instead
of
sending
p1
to
Barb
and
p2
to
Alice
this
Wi-Fi
water
just
sent
the
combination
of
the
two
packets,
so
Alice
remembers.
G
G
G
G
G
G
So
that's
the
just
to
avoid
this
kind,
this
type
of
objection-
and
that's
all
so
we
don't
want
to
have
an
intern
draft
on
this
topic
or
we
may
I,
but
there's
no
strict
requirement
for
that.
What
I've
done
so
far
is
I've
extended
the
github
readme
documents.
Well
fine
in
the
github
repository
in
order
to
add
those
simple
ideas
in
this
with
me
and
make
people
mostly
understand
what
we're
talking
about.
So
it's
already
in
the
github
in
we
in
the
coop
material
with
me,
fine
and
in
the
Swift
correct
readme
file
as
well.
G
Okay,
so
do
not
hesitate.
If
you
have
some
more
feedback
to
tell
us,
we
will
continuously
improve
those
ideas.
If
you
have
new
ideas
that
you
think
be
within
this
quick
introduction
to
fenc
notice
date
as
well,
she
find
that
the
where
they
are
organized
is
not
to
create
once
again
send
us
send
us
a
comment.
That's
more
than
way.
B
Think
the
sliding
windows
or
the
flexible
window
or
variable
size
windows
have
shown
better
performance
when
the
losses
are
correlated
so
that
that
could
maybe
I'm
suggesting
it
I'm
not
saying
you
should
do
it,
but
I
think
a
lot
of
people
think
that
they
have
a
very
wrong
or
a
very
bad
view
of
what
their
their
loss
profiles
are.
And
it's
not
that
you
know,
I
have
losses.
I'm
going
to
put
the
coat
over
it
and
everything's
got
to
go
well.
Well
depends
a
little
bit
of
what
this
looks
like
as
a
profile.
H
H
G
G
I
do
like
yep,
okay.
That
will
be
perfect
for
the
next
presentation,
because
in
this
presentation
we
are
talking
about
performance
simulation
and
personal
salvation
requires
to
have
lost
models.
So
if
you
have
indeed
satellite
driven
loss
models,
then
that's
very
great
because
we
I
don't
have
any.
As
far
as
I'm
concerned,
we
have
different
roles
models,
but
yes,
that's
quickly
appreciated.
So
you
want
to
speed
my
comments
and
my
question
in
this
presentation.
That's
fine!
Thank
you!
Nikolas,
okay,
so
this
presentation
is
about
efficiency
performance
simulation.
G
G
Okay,
it's
just
mentioned
that
this
is
not
the
first
time
we
are
mentioning
the
possibility
of
doing
consideration
for
efficiency.
The
at
least
boost
two
references,
one
from
Morton,
if
I
remember
in
IGF
eighty-ninth
some
time
ago
and
another
one
I
made
introduce
work
on
a
fake
frame
and
I
see
last
year
anyway.
G
So
there
are
several
aspects
to
the
program.
The
first
of
all
the
question
of
how
do
you
manage
your
correct?
What
are
the
correct
parameters?
How
do
you
initialize
those
parameters?
So
you?
What
are
good
values
to
give
them
to
provide
them?
So
that's
the
first
point.
The
second
mention
is
about
performance
metrics,
the
different
performance
metrics.
So
you
have
to
select
what
you
want
to
do:
the
right
metrics.
So
there
is
some
discussion
about
this.
There
is
also
a
methodology
discussion.
That's
something
very
important:
how
do
you
conduct
your
performance
variation?
G
And
the
fourth
item
is
bad.
Fourth,
dimension
is
about
the
communication
channel,
so
this
is
what
I
mentioned
just
before
after
suggestion
and
proposal
to
share
satellites
loss
models.
So
that's
something
very
important
because
you
want
to
test
your
solution.
Your
fe
co
NC
solution
for
a
given
news
case,
so
you
need
to
have
lost
model
that
is
representative
of
this
first
case.
So
that's
something.
G
And,
of
course,
it's
a
fifth
I'd
mentioned:
you
need
to
use
tools,
so
this
Acton
project,
which
focuses
for
the
moment
on
writing,
designing
Nephi
and
NC
solution.
So
next
step
could
be
to
design
tools
to
evaluate
this
performance
of
this
is
incorrect
solution,
so
it
could
be
one,
it
could
be
a
new
academic
project
after
so
that's
not
right.
That's
not
number
one
the
first
priority,
but
it
could
be
something
that
we
know
second
step.
G
Okay,
so
I
will
go
through
the
first
four
items,
four
dimensions
now
when
details,
so
the
first
topic
is
about
the
correct
parameters.
So
what
are
the
parameters?
And
how
do
you
derive
these
parameters?
So
it
of
course
depends
on
the
features
of
the
sauce
flow.
So
there
are
several
possible
parameters,
so
I
mentioned
some
of
them.
Now
this
slide
is
about.
How
do
you
manage
these
parameters?
G
Do
you
consider
the
parameters
in
terms
of,
for
instance,
their
window
size
in
terms
of
number
of
symbols,
or
do
you
express
this
window
size
in
terms
of
time
seconds
number
of
seconds,
so
whatever
fits
inside
this
time
frame
will
be
part
of
this,
including
window.
So
that
those
are
the
two
possibilities
I
can
imagine
and
depending
on
the
situation,
whether
you
have
a
fixed,
great
sauce
flow,
viable
rate,
sauce
flow,
real
time
constraints
or
not,
you
will
do
one.
We
will
select
one
of
those
indepence,
so
you
have
to
be
quite
flexible.
G
G
Still
for
those
correct
parameters,
so
you
the
uncoated
side.
We
had
this
anchoring
window
size
parameter,
which
is
one
of
the
key
input
parameters.
You
have
the
code
rate,
the
amount
of
energy
that
you
put
within
this
data
flow,
and
you
also
have
in
case
of
real
time
flows.
This
F
is
related
latency
budget.
So
that's
the
amount
of
time
that
can
accommodate
for
encoding
and
decoding
operations,
so
this
latency
budget
accounts
for
this
time.
G
It
does
not
include
the
consummation
time,
the
propagation
time
all
the
types
of
delays
that
you
made
parents
understand
or
disfavor
on
using
the
network.
It's
just
for
FEC,
encoding
and
decoding
operations.
That's
something
very
important
in
case
of
real-time
flow,
because
this
is
one
of
the
first
parameter
that
will
be
used
to
in
that
can
be
used
to
initialize,
for
instance,
the
and
canoeing
window
size.
So
at
the
B
color
size
were
this
Nino
system
maximum
size,
and
you
also
have
in
case
of
red
and
blue.
This
is
the
way
I
understand
it.
G
What
I'm
saying,
but
you
won't
decode
those
packets,
too
late,
to
be
useful
to
the
application.
So
this
is
forward
time
flows.
So
those
are
the
main
two
parameters
that
I
can
imagine
for
the
current
there's,
a
relationship
between
the
two
once
again,
I
don't
have
time
to
go
into
all
the
details,
but
something
we
can
discuss
offline
and
we
try
to
explain
how
to
derive
those
parameters
depending
on
the
situation
in
this
document.
G
So
this
is
the
oil
CFX
scheme
document
that
I'm
working
on
in
the
context
of
in
the
context
of
the
GS
v
working
group,
I'm,
not
free,
satisfied
by
this
those
guidelines.
Maybe
we
can
improve
this
description.
We
can
find
better
ways
to
initialize
or
clearly
I
have
the
feeling
we
need
to
work
together
on
this
topic,
because
it's
not
easy.
It's
not
trivial
and
perhaps
we
can
go
and
find
something
or
a
better
way
to
express
this
mid.
Rho
G
parameter
duration,
stuff
together.
So
that's
for
the
first
step.
G
G
All
those
metrics
are
quite
classic,
but
you
can
also
find
more
practical
conic
oriented
metrics
like
en
Coningsby
in
the
decoding,
speed
maximum
above
memory
and
the
maximum
amount
of
memory
required
during
decoding
the
number
of
field
operations
you
want
to
abstract
to
bits
instead
of
working
on
speeds
and
column
upon
each
page
can
also
consider
the
number
of
physical
operation.
That's
something
that
can
be
quite
useful
in
some
situations.
To
compare
different
decoding.
Algorithms,
for
instance,
so
lose
art
was
a
classic
performance
metric.
G
G
We
connected
some
research
from
this
topic
in
section
static.
What
is
the
required
could
rate
in
order
to
be
able
to
achieve
a
certain
quality
in
terms
of
transmission,
and
since
we
are
considering
real-time
data
flows,
this
quality
is
not
perfect,
but
you
have
an
objective.
This
objective,
expressed
in
terms
of
residual
packet
erasers
after
doing
whatever
FEC
and
coding
the
canoe
you
may
want.
So
what
is
the
required
rate
to
achieve
this
objective
quality
objective
for
given
to
maximum
returns
so
that
there
is
a
methodology
to
carry
out
this
kind
of
experiments?
G
It's
well
I've
three
ideas
on
how
to
do
that,
but
you
I
would
be
happy
to
share
with
you
also
may
different
methodologies
different
IDs.
So
but
yes,
that's
something
extremely
important
to
consider
that
we
are
not
always
targeting
a
full
reliability,
but
having
a
reasonable
packet
loss
rate
can
be
unavoidable
in
some
circumstances,
especially
if
you
have
real
time
constraints.
G
Next,
whenever
we
have
to
consider
time
makes
a
variation
quite
complex,
so
that's
abuse,
but
there's
no
way
to
avoid
it
if
you
are
considering
where
time
flows,
and
so
we
have
to
accumulate
this,
so
the
first
dimension
is
but
the
metallurgy,
or
do
you
do
that
you
can
carry
out
a
thorough
analysis?
That's
quite
useful.
G
We
never
say
the
opposite,
but
this
document
is
not
about
this.
We
can
also
carry
out
real
world
experiments.
That's
once
again
very
important,
but
the
metallurgy
is
quite
used
case
by
use
case
at
specific.
Oh,
you
can
also
try
to
do
simulations
where,
with
this
idea
of
fully
controlled
and
reproducible
experiments,
that's
extremely
useful
and
accurate.
G
G
There
are
two
following
that
we
have
our
own
tools.
Make
some
work
in
this
domain.
I
would
like
I
will
be
happy
to
share
with
the
group
it's
already
available
in
pots,
but
I'm
pretty
sure
that
you
guys
have
your
own
tools.
Holding
that
we
can
exchange
and
share
knowledge
from
this
aspect
can
be
quite
useful
when
we
are
doing
simulations.
We
need
to
be
very
careful
in
the
way
we
carry
out
those
simulations,
because
can
we
must
take
into
account
the
correct
specificities
and
the
way
it's
implemented
can
change
totally
in
the
conclusion.
G
If
you
don't
take
them,
if
you
don't
take
that
into
consideration,
can
take,
you
can
easily
go
into
wrong
conclusions
just
because
you
only
get
to
think
that.
Okay,
if
you
implement
that
feature
with
one
puts
Q
or
two
out
to
choose,
that's
the
example
I'm
vining
and
the
bottom
of
this
slide.
This
is
for
blood
cuts,
but,
depending
on
the
way
you
implement
efficient
cleaning
and
blood
cuts,
you
can
either
consider
one
output
out.
Sorry
I
forgot
to
say
something:
we
want
to
have
a
constant
bitrate
output.
G
So,
in
order
to
achieve
this
constant,
bitrate
output,
you
can
either
use
a
single
output,
Q
or
two
alkyd
use
each
of
them
betrayed,
so
it
depends
how
you
assign
source
and
repair
packets
to
this
output
Q
or
to
use
to
output
Q's.
The
result
will
be
completely
different
in
terms
of
packet
ordering
in
the
link
in
the
connection
link
between
the
sender
and
the
receiver
and
which
will
lead
to
completely
different
conclusions.
G
G
G
This
is
a
framework.
The
fourth
topic
is
about
the
communication
channels
that
we
need
to
consider.
So
when
you
want
to
assess
the
performance
of
a
given
solution,
forgiven
use
case,
so
you
need
to
have
a
model
for
your
channel
loss
model
and
that's
typically,
what
we
already
mentioned
to
the
morning
discuss
to
this
morning.
So
it's
very
important
to
have
traces
that
are
representative
of
your
target
use
case.
G
So
as
far
as
I'm
concerned,
and
so
far,
I
was
only
aware
of
this
3gpp
essay
for
mobility,
traces,
which
is
quite
a
school
that
I've
been
using
quite
a
lot
in
the
past.
It's
convenient
because,
instead
of
having
this
to
state
in
that
model,
where
you
have
two
parameters,
two
probabilities
and
you
have
to
find
appropriate
values
for
those
two
parameters,
which
means
that
you
need
to
explore
a
two-dimensional
space,
infinite
space
and
you
end
up
with
some
values
that
are
not
necessarily
able
to
detective
with
your
case.
G
Instead
of
that,
you
have
those
traces
that
will
tell
you.
Ok,
if
you
are
a
pedestrian,
what
working
in
the
street
at
that
speed,
then
you
may
experience
either
a
1%
average
lost
rate
or
5
percent
or
10
or
20,
but
in
that
case
those
are
the
four
loss
finds
they
provide
for
lost
files
that
explain
you.
What
are
the
packet
that
will
be
received
and
what
are
the
packets
that
will
be
lost
on
the
network?
So
this
is
the
figure
on
the
right.
G
Each
sign
indicates
a
packet
loss,
so
you
can
follow
this
fine
and
see
if
this
packet
that
has
been
sent
by
the
sender
will
arrive.
The
receiver
or
not
so,
depending
on
the
average
loss
rate,
will
have
more
or
less
packet
losses
and
since
you're,
a
pedestrian
when
you
live
via
an
obstacle
will
stay
on
this
obstacle
for
quite
a
long
time,
and
therefore
you
will
see
burst
of
packet
losses
on
the
opposite
when
you
are
in
a
vehicle
at
120
kilometer
per
hour.
G
Different
model,
and
once
again,
they
provide
four
different
files
with
an
average
one
person,
packet
loss,
5,000,
10
and
20,
and
you
have
to
slice
very
convenient
use
files
that
will
tell
you
what
packets
are
received
on
what
packet
that
we
must.
So
that's
something
it
is
available
on
the
web.
You
can
go
to
this
website
and
we
treat
them
and
use
them
for
your
own
equations,
but
it's
just
for
free
PPSA
for
every
Asians.
For
a
specific.
G
Frequency
bound
to
get
something
like
that,
it's
not
necessarily
representative
of
all
the
use
cases.
So
this
is
why
I
mentioned
earlier.
When
we
grab
propose
to
share
saturate
lost
models.
I
was
very
happy
because
it's
exactly
the
kind
of
thing
that
we
need
in
order
to
do
tough
conservation
for
satellite
communications.
B
G
Model,
so
it's
a
loss
model,
yes,
its
loss
model,
the
result
for
us
model,
but
yes,
the
probably
also
lost
price
under.
If
we
have
access
to
some
of
them
are
publicly
available,
then
that's
great
to
share.
That
would
be
great.
So
the
idea
is
to
might
get
some
feedback
and
if
you
are
well
something
then.
G
To
wrap
it
okay,
so
that's
that
consideration
something
very
important
for
the
group.
I
would
say
we
need
to
be
careful,
but
of
course
we
are
talking
together
on
we
try
to
do
our
best,
most
probably
in
the
case.
We
need
to
write
in
you
into
the
draft
on
this
topic.
I
think
it's
useful,
I
cut
off.
Of
course,.
G
I
need
to
shake
hands
directly,
I
need,
sharing
and
trying
to
find
good
ways
to
initialize
products
find
good
ways
to
run
performance
relations
on
Otis,
yes,
questions
comments,
just
a
comment
because
I
tell
you
agree
about:
do
we
need
lost
models
or
because
it
depends
a
lot
on
where
you
are
in
the
oscillator,
because
if
would
you
take
layer,
two
oscillators
measurements
from
silver
networks?
You
have
all
these
fake
and
things
already
going
on,
so
you
have
to
be
sure
about
where
packet
size,
if
it
is
measured.
G
B
So
I
have
anything
about
the
codec
is
number
one
I
think
a
lot
of
us
again
chair
chair
hat
away.
A
lot
of
us
have
done
n,
F,
Street
and
NS
to
simulation
models,
and
what
I
think
is
missing.
A
lot
of
time
in
these
model
is
actually
good
simulation
of
the
codec
itself.
There's
not
a
lot
of
very
good
ns-two
or
NS
tree
code
simulator.
However,
there
is
a
lot
of
I.
B
B
B
B
B
Okay,
thank
you
so
anyway,
like
I
said,
we've
gone
through
a
number
of
iterations
of
these
milestones
and
I
think
it's
time
to
review
where
we
are,
and
also
we
will
like
I
said
at
the
beginning,
send
these
to
the
list
also
to
make
sure
that
everybody
agrees
with
potential
changes.
So
we
had
all
of
this
that
were
there,
we
have
some
that
I
think
are
in
pretty
good
shape.
There
are
things
in
red
that
we
need
to
either
review
change
or
delete,
and
that's
what
I'm
going
to
talk
about
in
the
next
few
minutes.
B
B
Frankly,
rekindle
the
work
on
RL
and
C
and
Tetris,
and
so
I
knew
that
km
is
on
meet
echo.
So
take
that
as
a
nation
and
based
on
says
he
can
actually
oh.
We
would
also
like
the
tetris
team
to
update
their
their
draft
and
we
would
like
to
adopt
both
of
those.
As
as
our
G
item,
we
will
also
have
dragon
cast
to
update
and
and
adopt
as
a
RG,
and
so
we
still
waiting
for
these
inputs
and
and
now
with
bats.
B
B
Actually
it
this
turned
out
to
be
more
way
more
complex
than
than
we
expected,
because
it
really
has
specifications
and
quick
sources
and
experiment
that
or
a
vote
that
evolved
at
the
same
time
as
we're
trying
to
put
them
them
down
and
put
a
coat
over
it.
So
that
is
currently
still
on,
however,
there's
no
way
we
could
finish
that
this
year
and
we
plan
to
reschedule
it
for
next
year
as
quick,
probably
in
another
room
right
now,
are
really
busy
issuing
there.
B
One
and
obviously
the
code
will
be
part
of
the
next
version,
the
draft
and
satellite.
What
this
is
is
almost
done:
Thank,
You
Nicholas,
and
so
the
proposal
is
probably
to
be
done
by
December,
1st
and
time
for
Christmas
shopping,
and
so,
if
the
authors
can
address
that
it
would
be
really
great
the
common
API.
B
We
also
want
to
reschedule
it
because,
essentially,
as
I
think
they
sound
really
showed
is
that
it
was
one
thing
writing
about
the
API
and
it's
another
thing
about
putting
it
in
the
code,
and
so
we
we
plan
to
reschedule
it
as
it's
heavily
dependent
on
the
hackathon.
So
yesterday
we
were
discussing
about
this.
B
So
this
is
an
interesting
thing
we
had
at
the
beginning
when
we
bit
retarted
the
group
said
ok,
so
we
should
look
at
some
of
the
adoption
challenges
and
in
effect
it's
been
done,
but
not
in
a
draft.
It's
been
done
in
all
the
drafts
that
have
been
issued
about
applications
via
the
satellite
B
and
the
CC,
and
they
have
addressed
in
there
the
challenges
to
adoption
that
the
adopt
the
challenges
to
implementation.
B
So
we
think
that
actually
we
can
give
it
up,
but
we
shouldn't
we're
not
giving
up
the
idea,
we're
still
thinking
that
applications
of
network
coding
should
have
something.
Every
time
you
write
a
draft
about.
You
know
what
are
the
adoption
challenges
but
in
it
would
be
like
and
over
it
would
be
just
doing
cut
and
paste
from
existing
documents
to
create
another
draft.
So
we
would
plan
to
think
that
this
is
not
very
useful
anymore.
B
Now.
This
is
the
that
the
rabbit
that
comes
out
of
the
hat-
and
they
saw
mentioned
this
ID-
that
you
can
actually
do-
network
coding
or
FEC
coding
that
is
going
to
have
good
performance
even
within
a
using
protocols
that
have
congestion
control.
The
status
of
this
is
uncertain
because
of
other
work.
B
I
had
no
time
frankly
to
to
do
this,
so
the
surgeon
that
the
status
is
in
surgeon
I
got
in
a
discussion
yesterday
with
a
bunch
of
people
who
were
interested,
but
they
I
don't
think
they
also
have
time
the
exact
goal
may
have
to
be
reformulated.
I
think
you
missed
a
line
here,
no
yeah!
We
are
looking
for
a
champion,
we're
looking
for
people
who
would
collaborate
to
this,
especially
people
who
have
done
it
and
the
big
thing
is
always.
How
does
how
do
you
deal
with?
B
Essentially,
all
the
time
Network
coding
with
TCP,
so
I
think
we'll
put
that
on
the
list.
We
need
to
have
more
discussions
and
decide
what
we
want
to
do
with
this
again,
there's
been
a
lot
of
work.
There's
been
new
flavors
of
TCP
that
have
it
there's
been
all
kinds
of
tunneling
solutions
that
have
been
proposed.
So
what
do
we
want
to
do
with
this,
and
maybe
we
just
wanna
have
something
simple,
just
to
say:
okay,
frankly,
we
can't
do
anything
more
than
what
people
have
done
already.
B
We
may
discover
that
there's
more
research-
and
we
didn't
know
so-
maybe
again
this
is
going
to
the
list.
The
DNC
and
ICN
will.
Obviously
thank
you
guys.
This
is
on
track.
This
is
perfect.
It
will
be
a
small
reschedule
to
make
sure
that
it
goes
to
the
the
last
call,
and
actually
you
can
again
converge
to
the
final
version,
so
the
questions
is
to
with
the
author's.
What
do
you
think?
I
B
B
The
also
that
we
had
talked
about
looking
at
that
we're
coding
and
robust
tunneling,
a
bunch
of
people
were
interested
in
this
I
think
we
we
have
ended
it
and
the
proposal
is
to
take
it
out
as
a
milestone.
Because
again,
this
is
now
part
of
specific
implementations
to
end
every
code
that
has
been
proposed
and
I
think
even
Nikolas.
You
address
it
into
the
satellite,
so
maybe
it
is
something
that
has
now
been
embedded
like
the
other
ideas,
some
of
the
either
ideas
have
been
embedded
in
to
other
people's
work.
B
B
B
The
new
milestones
may
be
that
we
want
to
add
this
forward
or
correction
that
were
performance,
evaluation,
performance
evaluation
if
Morton
and
company
are
interested
in
the
draft
and
if
other
people
are
ready
to
collaborate-
and
this
has
to
deal
again
with
loss
profiles
and
everything.
There's
this
idea
of
the
computer
and
the
network
for
net
for
network
coding.
Actually
there's
some
people
on
the
list
who
have
said
that
they
had
started
working
on
implementations
of
network
coding
and
p4,
for
example,
so
that
they
would
be
inside
switches
and
they've.
You
have
a
question
no.
H
B
H
B
There's
the
the
P
for
network
coding,
which
is
how
do
we
re
encode
stuff
inside
the
network
and
then
there's
what
I'm
presenting
in
two
minutes,
which
is
what
can
network
coding
do
to
help
things
that
are
computed
in
the
network
or
the
application?
So
I
yeah,
that's
two
things,
there's
there's
there's
one
is
helping
and
the
other
one
is
actually
you
doing
it
so
we'll
see.
The
idea
is
that
we
wait
some
more
and
we'll
see
what
happens
and
I
would
say
for
me
personally,
I
would
say.
B
Probably
the
performance
evaluation
could
be
actually
a
very
valid
milestone
because
of
what
is
needed,
and
because
also
of
your
of
the
coding
for
dummies,
which
has
defined
what
things
are
and
then
having
this
thing
is,
how
do
you
evaluate
the
performance
could
be
really
well
the
computer
in
the
network
computing
in
the
network.
The
reason
I'm
not
pushing
that
much
in
this
group.
It's
because
we're
trying
to
start
a
new
group
which
is
not
what
I'm
going
to
talk
a
little
bit
about
questions.
J
I'd
like
to
suggest
you
to
add
one
milestone,
it's
related
to
network
coding
with
ICN
as
a
solution,
because
currently
it's
just
defined
a
requirement
that
our
challenges
and
the
most
important
I
don't
say
most.
But
one
of
the
most
important
point
is
that
how
we
express
that
network
earning
function
within
IC
or
CCM
network.
So
this.
J
B
Thank
you
very
much
well
taken.
Ok,
so
I'm,
still
long,
I
guess,
if
you're
on
the
other
I'll
be
faster
next.
B
Ok,
so
for
those
of
you
were
at
the
ICN
RG
interim
yesterday,
this
will
mean
like
it
will
sound
like
stuff
that
you've
seen
before,
but
that's
going
to
do
it
and
by
the
way
this
is,
the
picture
is
from
magic
leap
who
actually
made
their
name
by
showing
this
picture.
It's
actually
with
goggles
of
a
whale
that
jumps
out
of
a
basketball
court.
B
So
there's
a
draft
that
I
wrote
about
the
review
of
what
I
call
the
requirements
for
what
I
call
extended
reality
for
me.
Extended
reality
is
the
Cromwell
is
actually
trying
to
find
the
commonalities
between
augmented
reality,
virtual
reality,
360
degree,
video
experiences
and
stuff
and
the
ID,
because
we
are
suggestion
we'll
suggesting
the
creation
of
a
new
research
group
on
in
network
computing,
whether
the
idea
was
actually
to
see
once
you
have
networked
X
our
experiences.
B
What
are
the
challenges
and
what
is
needed
in
terms
of
adding
computing
in
the
network
to
to
be
able
to
meet
these
requirements
and
also
to
look
at
some
of
the
open
issues
and
the
research
aspects
and
a
lot
of
them
have
not
been
resolved
yet.
So
the
idea-
and
you
see
that
I've
put
like
packet
loss
and
loss
and
use
delay
here,
because
this
is
this
research
group,
but
a
lot
of
other
things.
B
Are
there
I
thought
it's
the
multi-source
multi
destination
problem,
but
you
have
a
lot
of
sources
for
your
experience
being
part
of
an
IOT,
some
video,
some
overlaying
graphics,
a
lot
of
things
that
are
bringing
together
and
they
come
from
different
sources
and
they're
going
to
different
destinations,
and
especially
if
you
do
it
in
a
multi-party
and
that
everybody
is
participating
into
that
experience,
then
you
get
a
lot
of
interesting
networking
challenge.
So
it's
this
shared
experience,
it's
interconnected.
B
It's
distributed,
you
need
some
Federation
and
so
you're.
Creating
this
global
immersive
experience,
and
so
some
of
the
principles
is
allowed.
Joint,
collaborating
and
X
are
when
the
best
one
I
saw
was
actually
joint
surgery.
Joint
manufacturing
IKEA
actually
has
a
very
cool,
augmented
reality
things
where
people
are
building.
B
No
tables
and
chairs
in
na
R,
so
the
multi-view
also
like
when
you
start
looking
at
it
for
a
different
point
of
view
and
all
the
extra
streams,
and
also
that
becomes
also
another
issue,
is
that
some
of
these
streams
are
very,
very
large
streams
with
very
large
bandwidth
requirements,
and
some
of
them
are
very
small
but
they're,
the
ones
who
actually
are
driving
the
experience.
So
the
challenges-
and
thank
you,
take
your
day
for
suggesting
to
take
the
word
that
was
there.
B
So
it's
actually
the
Andres
is
on
time,
delivery
of
all
these
multiple
streams
and
their
rendering
across
multiple
participants.
So
you
have
you
want
these
things
to
happen
at
the
same
time,
because
if
these
people
are
sharing,
they
have
to
share
together.
I
will
tell
you
that
I
have
experience
of
something
that
we
used
to
call
social
television,
which
was
also
either
thing
about
having
people
were
sharing
the
same
video
experience
and
it
was
a
big
thing
of
having
them
having
at
the
same
time
and
actually
network
coding
helped
in
that.
B
Obviously,
a
lot
of
these
streams
that
are
coming
have
a
lot
of
sensitivity
to
packet
loss
and
lots
induced
delay,
especially
strangely
enough
for
the
non
video
components.
If
you're
going
to
have
something,
that's
overlaid
and
is
graphics.
If
these
things
are
delayed,
then
the
whole
experience
falls
apart
or
if
their
loss
is
even
worse,
so
at
low
end-to-end
delay
and
delay
variations.
B
It
is
very
well
known
that
when
things
in
immersive
experiences
are
delayed
and
that
you
move
your
head
to
try
to
see
them,
you
get
seasick,
because
your
brain
expects
the
experience
to
move
with
you
and
it
doesn't.
And
obviously
this
is
this
idea
of
optimizing
the
cache
and
the
rendering
this
is
actually
inheritance
from
gaming.
The
gamers
have
started
doing
that.
You
start
putting
elements
in
the
network
that
will
not
change
non
variant
elements
and
then
whatever
is
the
activity.
The
dynamic
stuff
goes
over
it,
but
all
of
this
is
part
of
the
experience.
B
So
what
would
we'd
like
to
have?
This
is
actually
I'm
just
figuring
out.
This
is
extremely
busy,
so
you
would
like
to
optimize
the
location
that
type
of
resources
that
you
want
to
have.
You
would
like
to
distribute
the
functionalities
between
data
center
and
edge.
This
actually
requires
functional
decomposition
of
some
of
the
elements.
Again.
I
said
that
some
things
are
really
well
done
in
the
in
the
core
in
the
in
the
cloud.
B
There
are
things
that
cannot
be
done
there,
which
are
the
things
that
really
happen
now
you
may
want
to
have
Federation's
of
these
edge
nodes.
That
goes
back
to
the
idea
of
how
do
you
distribute
the
functionality,
and
how
do
you
disagree
the
load
across
all
these
things
and
then
the
evaluation?
Again,
that's
actually
from
the
gaming
world,
the
local
caching,
to
make
micro
data
centers
with
some
of
the
pre-rendering
that
is
needed
to
for
the
experience
you
may
want
to
have
multicast
distribution.
B
Another
way
that
is
in
bold
anyway,
you
and
again
one
of
these
things,
I
think,
is
interesting
here.
If
you
have
this
idea
that
you're
going
to
put
things
in
the
network
to
help
these
things
to
help,
these
experiences
is
that
there
are
it's
not
about
just
putting
it
in
the
cloud
which
was
what
people
thought
in
the
beginning.
B
B
Well,
baton
was
very
interesting.
You
want
to
have
dynamic
allocation
of
all
the
functionality
as
people
move.
I
speak
as
the
group
changes,
as
the
experience
evolves
and
you
may
want
to
have
performance
optimization,
and
then
we
go
back
and
some
of
the
things
that
we
talked
here
on
tunneling,
sufficient
virtualization
and
loss
protection,
and
so
I
will
not
go
into
this,
because
we
will
not
have
time.
B
But
this
is
essentially
the
architecture
or
one
of
the
architectures
and
I
made
it
very
complex
to
actually
highlight
that
a
lot
of
these
things,
although
there
is
ICN
and
there
or
CCN
there
is
in
the
network
computing
that
helps
some
of
the
fake
functionality
coined
by
the
ways
for
computing
in
the
network.
There
are
things
that
come
from
the
cloud
and
interact
with
the
cloud
and
go
back
to
the
cloud.
B
I
think
a
lot
of
time
that
I
think
one
of
the
mistakes
of
all
these
architectures
is
you
have
the
cloud
you
have
the
transit
network
and
you
have
the
destination,
not
thinking
half
way
that
there's
interactions
between
things
come
to
the
cloud
or
process
go
back
to
a
cloud
go
to
the
network
and
when
they
come
to
the
edge,
they
have
to
actually
net
synchronize,
with
incoming
streams
that
are
coming
from
other
parts
of
the
network.
So
I
think
this
is
what
I
wanted
to
highlight
in
that.
B
What's
the
link
to
this
network
XR
is
obviously
very
delay
and
lost
sensitive.
We
want
to
use
a
racial
coding
to
provide
this
packet,
the
racial
coding
to
maximize
things
like
related
to
peer
to
peer
and
multipath
efficiency,
and
reduce
the
need
for
format
for
retransmissions.
Actually,
you
want
to
prevent
as
much
as
you
want,
as
you
can
to
have
three
transmission
of
the
lost
information
you
would
like
to
recreate
it
and
it
has
to
have
it
really
has
a.
B
We
were
mentioning
this
right
now:
the
link
to
current
in
network
computing
and
programmable
network
elements.
Once
you
can
actually
program
the
in
the
network
elements
to
facilitate
the
the
streams
that
that
facilitate
the
fact
that
the
streams
need
to
have
very
low
loss
and
very
low
delay,
you're
going
to
be
very
happy
with
this,
so
there
is
a
strong
link
to
this
group.
However,
there
is
going
to
be
a
request.
B
B
So
it's
a
proposal
for
a
proposed
research
group
we're
meeting
on
Friday.
My
co-chair
is
in
the
back
there
Jeffery
from
Huawei,
and
the
idea
is
to
look
at
the
whole
research
that
is
happening
right
now
in
computing
in
the
network,
both
in
the
data
center
at
the
edge
and
again,
like
I,
said
in
the
middle.
A
lot
of
the
work
is
research
of
what
you
called
things
that
heart
in
the
network
in
terms
of
how
they
reprogram,
but
because
of
my
own
interest.
B
I
also
think
that
we
will
look
at
some
of
the
applications
that
are
enabled
once
you
can
do
that
and
with
then
can
justify
all
the
other
implementation
so
again
join
us
Friday
and
for
this
group,
I
think
there
is
a
strong
possibility
for
for
FEC
and
network
coding
to
basically
be
part
of
one
of
the
great
XOR
solutions.
Thank
you.
G
G
B
Thank
you.
Everyone,
merci
beaucoup,
the
question
from
Phil
meeting
of
Jose.
We
you
should
you
should
have
known
this
is
for
people
who
came
in
late
for
the
first
five
minutes
of
this
meeting.
Everybody
was
here
about
it's.
Almost
everybody
spoke
French.
So
thank
you
very
much.
We
will
see
you
in
Prague.
We
will
see
you
on
the
list
and
I
can
see
that
there
was
a
lot
of
new
people
here,
and
this,
for
us
is
absolutely
fantastic.
Please
collaborate.