►
From YouTube: IETF104-ICNRG-20190329-1050
Description
ICNRG meeting session at IETF104
2019/03/29 1050
https://datatracker.ietf.org/meeting/104/proceedings/
C
A
A
This
is
the
main
ITF
104
I
see
NRG
meeting
we
had
an
interim
last
Sunday,
we've
seen
a
number
of
you.
Almost
all
of
you
then,
but
some
of
you
not
so
we'll
take
a
little
bit
of
time
to
go
through
that.
So
we
have
a
note-taker.
Everyone,
thankfully,
shows
in
with
the
tea
who
has
who
took
notes
on
Sunday
and
has
agreed
to
take
notes
today.
Saves
us
a
lot
of
time
to
not
have
to
embarrass
people
into
trying
to
do
it
at
the
last
minute.
A
To
start
off,
you
all
should
be
aware
of
the
IRT
F
policy,
which
deals
with
the
disclosure
of
intellectual
property
and
develop
behavioral
and
other
aspects.
So
if
you
haven't
looked
at
it
recently,
please
look
at
it.
Oh,
so
here's
all
our
administrative
information
where
the
mailing
list
is
where
our
are,
where
the
dress
is
where
our
wiki
is,
there
might
be
a
couple
of
people
in
the
room.
A
Don't
know
us
I'm,
Dave
or-
and
this
is
buh
da
or
among
this-
is
their
cuiture
and
we,
as
a
team
have
been
co-chairing
I,
see
NRG
for
eight
years.
Yes,
seven,
eight
years,
something
like
that:
no,
not
that
long,
maybe
six
or
seven
years,
at
any
rate,
if
you'd
like
to
contribute
to
the
note,
oh
you're,
not
using
the
ether
pad
right,
you
know
what
these
neat
about
nevermind.
A
A
A
We
have
a
technical
talk
on
hop-hop-hop
authorization,
a
bunch
of
document
updates,
which
hopefully
will
go
reasonably
quickly,
just
to
make
sure
people
are
up
to
date
on
where
the
documents
are
and
what
our
plans
were
progressing
them
is,
and
then
the
one
sort
of
longer
section
trying
where
we're
going
to
try
to
make
progress
on
how
we
slide
introducing
quality
of
service
ideas
and
mechanisms
into
the
into
our
ICN
architecture.
And
for
that
we
have
sort
of
like
to
reasonably
content
talks.
A
We
had
updates
on
on
the
name
resolution
services,
I'll,
say
very
briefly
on
that
we'd
like
to
complete
that
work,
but
there
hasn't
been
a
lot
participation
and
the
folks
doing
on
it
doing
it
are
going
to
be
moving
on
to
be
doing
other
things.
So
we'd
like
to
wrap
that
up
as
quickly
as
possible
by
having
people
review
the
latest
drafts
and
reaching
closure
for
our
G
last
call
and
progression
of
those
on
mailing
list.
A
So
look
at
the
slides
there
and
then
we
started
a
discussion
of
management
tools
which
most
of
the
time
went
into
getting
people
back
up
to
speed
on
one
of
the
main
management
protocols
under
consideration,
which
has
been
adopted
as
an
RG
document
called
CCN
info.
So
questions
or
comments
on
the
sunday
in
room
yeah.
A
C
A
B
C
C
D
A
For
those
of
you
who
don't
know
all
the
administrative
crap,
so
any
document
that
has
asked
working
group,
research,
group,
consensus
and
and
people
agree
it's
ready
to
go.
Those
are
a
review
by
the
entire
flag.
People
on
the
irst
internet
research
steering
group
provide
comments
when
not
satisfied.
A
There's
a
final
publication
poll
to
the
IRS
G,
which
is
normally
pro
forma
and
relatively
quick.
Then
there's
two
stages
after
that:
the
next
stage
is
it
sent
to
the
iesg
there's
for
them
to
confirm
that
it
does
not
conflict
with
work
going
on
in
sustained
ITF
working
groups.
That
also
is
pretty
much
pro
forma
if
in
fact
that
doesn't
conflict
and
then
one
to
the
RFC
editor
for
final
editing
and
publication.
So
these
have
finished
all
of
our
review
processes
and
are
mostly
are
at
this
point.
C
C
A
B
E
E
So
what
may
be
interesting
to
you
is
so
the
TPC
has
decided
to
slightly
change
the
say,
approach
the
conference
this
time
to
reflect
the
maturity
of
the
field.
So
if
you
look
at
the
cup
of
papers,
you
will
see
that
this
different
set
of
topics.
We
think
it's
a
good
approach
to
strengthen
the
the
program,
so
we
also
have
to
you
know,
get
new
communities
interested
and
submitting
papers
to
the
conference.
E
C
E
D
G
G
The
publisher,
consumer
and
computer
guru
can
be
the
catchy
food
or
in
establishing
the
consumer,
will
consider
many
to
Canova
cats.
Why
is
continent
convenient
hell?
This
is
chapter
can
impersonate
a
currently.
The
content
is
only
signed
by
the
public
Alicia,
so
it
is
impossible
to
verify
the
publishers,
affinity
and
all
the
hookers
used
was
indicated.
Studies
of
the
CA
and
another
captain
would
consider
is
to
increase
the
flagging,
kappa
death
at
least
a
conclusion.
G
G
This
file
shows
the
influence
of
the
content
of
protein
in
camp
Lisa,
pepper,
Rooter
of
caches
effect.
The
data
this
is
a
coup
d'etat
and
this
rule
is
close
to
the
consumer,
so
that
consumer
always
get
to
the
wrong
data
that
is
pepper
and
I,
also
respect.
The
data
can
be
cached
that
we,
the
cache
language
group
at
the
butcher
as
the
bureau's
of
sprays
and
at
the
consumer
food,
which
could
also
always
to
be
killer,
fix
the
data.
So
the
Reuters
need
to
modify
the
data
before
cashing
need
to
be
vocation.
G
G
Merchants
they
have
a
cap
in
the
racking
up
cap,
so
the
traditional
indicates
two
wins
here.
Is
the
pseudo
killer?
You
see
any
certificate
of
authority.
There's
a
public
theater
soda
wenting
is
an
incoming
packet.
The
Boutros
are
here
to
help
interact
with
silica
or
silicon
at
history
and
acquire
to
fire
the
necessary
silicate.
So
there
is
when
it's
a
porter
to
the
next
to
turn.
It's
also
here
to
perform
the
certificate
of
discovery
and
competition.
G
So
these
additional
messaging
for
the
authentication,
which
means
it's
copy,
invisible
theater
here-
are
actual
daily.
Just
obviously,
resources
cannot
be
a
problem
for
the
motor
is
a
highly
costly
for
each
troop
are
always
needs
to
modify
the
signature
every
day,
I've
never
admitted
and
a
full
of
consumer.
The
public
field
or
the
NGO
computers
should
have
be
prior
to
the
leader
of
beforehand
to
serve
these
promote,
propose
a
toast,
Papa
Papa
was
indicating
MacKinnon.
The
key
idea
here
is
the
ball.
G
Soup
concocted
circulation
between
the
consumer
in
recognition
and
a
consumer
between
particular
brand
of
coca-cola
and
all
the
past,
so
the
also
the
co-host
can
be
so
you
can
also
reach
independent
nature
also
can
be
partially
associate
mr.
Kasich
or
shorten
the
names
of
the
symposium,
because
the
Civic
in
the
names
of
the
silica
chain,
if
we
could
retreat,
know
where
the
catalytic
ways
so
we
use
the
Sierra
strategies
to
enhance
the
security.
Some
capicola
is
the
feminine
also
to
perform
with
such
national
facilitation
without
a
secure
messages
and
we
extended
the
entity
classic
Rocco.
G
G
The
Pope
is
the
customs
and
not
I,
used
to
see
a
barrister,
and
a
for
the
Nova
currently
is
the
order,
authorization,
seven
regions.
So
the
certificate
is
you
dose
based
on
this
kind
of
trust
for
the
widow
energy
castle?
They
are
food.
The
most
brutal
are
report
to
the
issue
secret,
which
are
the
energy
for
each
other
for
the
CIA
mr.
G
This
slide
shows
how
to
construct
our
proposals
circular
suspicion
change.
There
is
the
user
X
to
retrieve
the
data
this
they
are.
These
two
figures
show
that
Y
so
show
that
the
two
decade
of
non-aboriginal-
and
these
few
shows
record-
shows
that
there
are
four
copper
color
of
blueprint
illness,
either
Canada
interested
to
the
next
operator
and
their
Pandavas.
Typically,
the
next
Hobbit
the
certificate
of
the
next
of
the
pool
beach
in
there,
and
this
RK
also
offended
them
next
up.
Our
particular
safe
in
the
damnedest
of
this
stupid
are
any
hope
at
all.
G
And
finally,
when
these
limits
of
which
is
Commission
order,
the
Commission
Ladakh
am
verified.
The
increased
who
is
using
this
super
chin
and
the
Commissioner
also
has,
of
course,
the
silicate
of
drawn
its
to
the
logical
entity
of
new
administrator.
The
news
was
candies
and
enemies
and
their
Pamela,
the
Civic.
This
is
significant,
it
already
hoods
and
not
yourself
it
again
and
again,
you
see
another
silica
of
the
data
with
rather
data
which
leads
to
depreciate
and
the
city
Peter
Florrick's
how
to
eat.
G
Energy
L
is
certificated
work
way
of
war
zone,
connection
of
a
backhaul,
and
finally,
this
simply
this
Teja
Wilson
Creek
in
which
is
was
music
and
then
user
X
can
modify
the
certificated
I
am
from
user
X
to
the
logical
entity.
Who's
supposed
to
tennis
and
the
whole
solution
has
a
certificate
of
Rome
user
X
through
the
PA
and
through
the
Commission
order
and
purify
the
computer
and
also
purify
this
task.
So
from
this
pure
these
two
issues.
G
If
the
enemy's
attention
about
the
losers,
you
know,
if
so
also
the
similar
process,
you
got
extended,
slow
hope,
I,
oppose
the
connection
of
the
certificates
and
there
are
one
can
modify
this
circuit
here
and
end
with
an
interesting
case,
and
if
the
circular,
this
is
significant,
another
people
requiring
to
a
bleep
job
that
you've
ever
and
if
we
talk,
can
be
purified.
The
era
where
we
will
try-
this
is
also
the
part
of
the
circulation
has
been
cash.
G
Here
are
one
through
the
published,
the
Nordic
or
anything
official
and
because
of
the
certificate
of
channel
and
to
hear
anisette
hour-long
appended.
The
city
data
from
our
Tula
Tula
Arwa
in
Nizam
Papa
her
book
connected
the
certificate
and
also
use
series.
It
has
a
switch
open.
This
city
finish
hint
so
after
this
guitar,
which
is
the
you
lazy
legs,
can
verify.
This
certificate
can
purify
the
probably
server
part
of
public
order
and
I
verify
the
possible
just
three
minutes:
okay
and
this
lease
to
the
single
in
production
on
this
sympathetic
in
model.
G
This
is
a
the
name
of
research
has
stuff
for
the
certificated
chance
and
release
is
a
logical
to
the
finish
here
and
we
users
here
there's
a
trouser
to
shorten
the
cinema
champions
the
focus
on
my
feeler,
you
what
you
see
a
bit
bizarre
groans
indicator:
use
of
civic
para
RK
of
-1
R
to
replace
the
silicon
at
7
Sebastian
from
arcane
-1,
R,
4,
R
3
to
R
2,
and
this
is
our
speech
in
Cambodia.
Oh,
this
was
shows
all
secretly
what
we
are
invested:
three
security
neighbors
and
for
us,
that's
agreeable.
G
G
This
should
get
a
chair,
so
it
a
one
two
three
four
here
we
moved
into
the
silicon
aside,
my
repeat:
affected
the
performance
of
our
power,
our
poles.
So
we
considered
the
investigated
the
sector
size.
We
can
get
to
some
reference
for
keyd
transport
me
a
security
interest
container
environments
for
the
ECDSA.
G
H
G
H
A
Hi
Dave
I
ran
I.
Have
one
clarification
question
in
one
comment,
so
the
clarification
question
is:
what
do
you
lose?
If
not
every
hop
does
this,
in
other
words,
some
some
hops
have
the
capability
to
do
authorization
authentication
on
that
hop
and
some
don't
the
whole
thing
breaker.
Do
you
still
get
some
improvement,
I'm
thinking
about.
G
A
A
Alright,
so
there's
there's
two
aspects
of
that.
To
think
about
one
is
that
what
CC
and
X
says
is
that
if
you
haven't
authenticated
the
content,
you
can
still
store
it,
but
you're
not
allowed
to
return
it
to
a
user.
You
must
bypass
the
cache
if
you
haven't
yet
had
a
chance
to
check
the
signature,
so
you
can
store
it
in
the
cache,
but
you're
not
supposed
to
answer
an
interest
and
then
the
second
part
of
that
is.
A
You
want
to
be
very
careful
to
not
evict
authenticated
content
in
order
to
store
content.
You
have
not
yet
authenticated
because
that
won't
pollute
the
cache,
but
it
could.
It
could
cause
a
cache.
What's
the
right
come
on,
there's
cache
solution
and
there's
cache
flashing
yeah.
So
a
couple
things
to
think
about
on
the
cache
behavior.
F
E
I
Okay,
so
okay,
so
my
name
is
Jane
from
Hamburg
I
mean
I'll
actually
present.
Two
things
today,
first,
is
updates
to
the
to
the
ICN
open
draft,
and
second
is
some
extramental
results.
If
you
did
for
paper
and
I
will
start
with
the
with
the
updates
first,
so
we
are
at
version
2
now,
since
a
couple
of
weeks,
and
we
did
basically
two
major
changes
to
the
draft.
I
We
first
added
and
like
a
guidance
section
for
implementers
to
to
ease
the
efforts
of
implementing
and
the
more
or
the
better
thing
is
that
we
also
made
use
of
open
space,
efficient
compression
format
to
to
represent
time
periods,
and
this
is
the
time
tre
that
I
want
to
present
now
and
the
time
TV
is
actually
work
done
by
the
men
at
working
group
so
by
this
group,
and
they
especially
in
I'd,
use
in
the
broken
protocol.
For
this
now.
I
The
time
TV
is
actually
yeah
encoding
to
represent
time
periods
and
they
used
a
one
byte,
and
in
this
one
byte
they
use
five
bits
to
to
express
the
exponent
and
three
bits
to
express
the
mantissa
and,
as
you
can
see
whether
familiar
you
can
also
look
this
up
by
online
and
in
the
RFC
five
four
nine
seven
and
they
are
able
to
use
a
white
of
to
represent
with
this
one
byte
a
wide
range
of
time
periods,
with,
of
course,
less
precision.
So
here
you
can
see,
for
example,
using
the
constant
per
thousand.
I
You
can
achieve
yeah.
Well,
you
do
from
nine
milliseconds
and
you
can
achieve
the
maximum
value
45
days
and,
as
you
can
see
in
the
picture
and
the
bottom
right,
that's
the
actual
distribution
of
the
values.
So
most
of
those
values
are
in
the
lower
range.
Then
you
have
to
cost
a
few
values
to
occurs
and
the
higher
ranges-
and
we
took
this
work
from
from
them
and
were
to
apply
this
to
the
SN
open.
And
what
we
basically
did
is
I
mean
we.
I
The
only
approach
was
to
to
just
cut
off
I
mean
the
CCO
Nixon
and
NBN
news,
actually
like
four
bytes
to
represent
these
these
time
values
and
before
bytes
you
can
achieve
the
same
range,
so
you
get
like
45
48
days.
We
have
four
bytes
representing
milliseconds
and
what
we
did
before
was
just
cutting
to
two
bytes,
which
of
course,
is
not
so
wait
because
then
also
only
have
64
seconds
to
represent.
I
So
when
the
or
initial
thought
that
was
well
noted,
sonority,
but
actually
it
makes
more
sense
to
have
a
wide
range
you
to
support
those
values.
So
we
just
took
those
two
basic
we
had
anyway,
and
now
we
use
them
and
use
a
time
care
we.
We
have
the
same
exponent,
so
we
have
too
big
a
mantissa.
So
we
have
four
position,
and
here
you
can
also
see
the
readiness
you
put
my
millisecond
for
minimum
and
48
days
for
maximum.
I
So
you
can
actually
we
use
this
time
TLB
for
the
interest,
lifetime
and
content
freshness,
because
they
are
time
periods
and
CCA
makes
you
also
have
absolute
time,
which
is
used
for
the
content,
store
and
content
caches.
We
can
represent
this,
of
course,
this
approach
and
that's
why
I
mean
let's
say
an
issue
here
using
this
formula,
you
can't
express
zero
because
the
minimum
value
is
zero
milliseconds.
So
what
we
did
we
defined?
I
Okay,
if
the
time
code,
if
a
and
B
are
zero,
then
we
want
to
represent
zero,
because
some
protocols
or
specifications
may
use
the
the
zero
value,
for
example
the
systemic
semantics
or
messages
from
this
using
the
zero
say
Emile's
or
the
left
interest
after
one
zero
to
say:
okay,
don't
open
pit
stays
on
one?
Don't
use,
reading
data
and
stuff
like
this,
and
so
we
defined
this
and
you
can
see
the
minimum
maximum
ends.
I
Yes,
taste
the
same
and
that's
another
problem,
you
might
say
the
problem
and
because
we
don't
know
what
the
application
is
going
to
use
this
and
as
a
as
a
time
value
it
might
not
fit
in
this
or
it
might
not
use
the
by
the
time
value
and
that
we
can
represent
it
in
the
time
code.
So
there
are
some
solution
to
this
and
that
we
try
to
outline
in
the
rough.
I
You
can,
of
course,
wound
up
the
value
to
a
value
type
like
a
time
value
and
which
always
work,
especially
if
you're
saying
later
then
yeah.
You
don't
want
to
change
if
I
could,
if
the
time
period
is
inside
the
security
envelope,
which
is
true
in
the
ndn
vote
for
the
freshness
time,
and
they
don't
foresee
a
mix,
I
think
it's
the
help
site
which
is
not
in
this
illicit.
You
Joe
can.
A
A
It's
interesting,
the
Crypt
should
think
about
it
because
again,
these
are
experimental
RFC's
and
we
can
change
them.
We
don't
have
to
worry
McChord
quebec,
where
compatibility
is
not
nearly
as
strong
and
consideration,
and
my
intuition
here
is
that
this
is
universally
better.
There
are
no
downsides
to
adopting
this
as
the
as
the
way
to
represent
interest
lifetime
and
and
other
time
scale
fields.
Yeah.
I
A
I
So
I
want
to
move
on
to
the
see
the
results
on
depicted
for
a
paper
for
the
networking
conference
and
obviously
of
this
year,
and
here
we
have
a
little
simple
set
up
or
three
IOT
devices,
plus
two
devices
limited
resources,
and
we
are
comparing
three
stacks.
So
we
have
fun
as
core
of
UDP
a
TP
6/6
OpenStack,
the
typical
80s
stacked,
and
we
use
NBN
over
50
forum
views
envy
and
vice
and
lopen
over
54,
and
we
have
consumer
for
a
producer
and
we
have
to
name
configurations.
I
The
first
name
is
this
express
its
name
short
and
you
can
see
all
the
example
and
stuff
and
the
red
part
of
our
example
is
the
thing
that
we
used
in
the
stateful
compression
to
light
the
name.
So
it's
not
there
when
we
want
to
be
ascending,
the
name
and
the
name
long.
You
can
see
that
we
use
yeah,
bigger
prefix
for
the
stateful
compression,
so
we
compress
more.
So
we
have
these
two
configurations
and
we
go
on,
and
here
you
can
see
the
average
processing
time
for
each
role
in
each
stack.
I
I
This
is
the
actual
processing
in
the
lo
pan
and
then
we
have
the
wet.
This
is
just
the
applications,
everything
the
pop
sticks,
open
or
I
or
I
send
opened.
So
what
we
see
here
is
basically
the
we
head
of
course,
processing
time
yeah.
We
are
doing
it
on
convergence
layer,
but
we
also
can
see
that
on
the
link
layer
we
reduce
time
by
sending
less
bytes
or
or
handing
less
bytes
over
to
the
buffer
transceiver.
I
So
in
total
we
actually
don't
add
ya,
know
significantly
more
pausing
times
in
these
scenarios,
and
then
we
look
at
the
message
sizes
and
then
message
sizes
here
over
the
air
and
again
for
all
those
worlds
and
and
and
configurations
is
just
zooming
into
the
Indian
part.
And
here
we
can
see
that,
depending
on
the
configuration
we
even
reduce
or
reduce
messages,
we
have
been
compression
to
reduce
messages
by
for
the
former,
even
by
almost
more
than
50%,
and
we
also
did
some
measurements
energy
measurements.
I
So
we
gain
like
100
or
reduce
the
energy
consumption
100
to
300
micro
Joule
for
one
interest,
data
or
transaction.
So
we
reduce
also
the
energy
consumption
and
as
an
outlook,
we
add,
of
course,
minimum
complexity.
This
is,
of
course,
always
true
if
the
converges
layer,
but
we
have
I,
mean
all
those
benefits,
of
course
outweigh
the
things
that
I
just
said,
and
the
next
question
actually
would
be,
and
also
to
the
chests
and
how
we
should
proceed
with
this.
With
this
draft,
I
mean
it
is
out,
there
should
be
yeah.
This
is
basically
yeah.
E
I
E
E
Okay,
so
that's
a
good
indication
that
we
need
more
people
looking
at
this,
because
I
think
it's
yeah.
We
think
it's
really
really
important
work.
It
really
also
clearly
demonstrates
very
nice
properties
and
we
want
to
parish
it
at
some
point.
As
the
next
mental
hour,
see
in
general,
I
mean
normally
work
is
the
author's
cannot
have
to
give
us
a
signal
when
they
think
it's
it's
ready
and
but
I'm
guessing
that
what
I'm
hearing
is
that
you're?
Not
quite
sure
you
would?
We
could
like
to
get
some
of
the
tech
at
this
point.
I'm.
K
And
there's
also
the
question
about
this
time.
What
Dave
was
rising
about
the
time
development
I
mean
first
of
all,
I
guess
is
a
discussion
on
the
mailing
list.
We
should
we
can
in
initiate
this
and
but
I
mean,
given
that
there
is
a
consensus
on
that,
the
time
tier
we
should
be
changed
or
the
old
one
should
be.
Deprecated
question
is
well
that's
what
I
was
proposing,
yeah
yeah
sure,
but
other
questions
out
how
to
do
this.
Should
this
be
a
separate
document
to
deprecate
I
think.
A
We'd
reissue
the
messages
draft.
The
CCO
next
message
is
draft
and
deprecated
the
old
CEO.
They
and
say
people
tell
people
to
use
the
new
TLV.
Okay,
I
mean
procedurally,
that's
what
one
would
do
for
CC
annex
for
cc.
Nx
I
can't,
unfortunately
tell
the
Indian
people
how
to
change
their
specs.
Maybe
maybe
you
figured
that
out
after
eight
years,
all
right.
C
A
L
So
this
draft
is
related
to
an
existing
draft,
which
is
the
ICN
on
LTE
and
4g
and
which
is
which
describes
how
to
enable
IC
an
over
4G
system,
and
this
current
draft
describes
how
to
enable
ICN
over
5g
system
but
I
think
what
it
does.
It
leverages
very
similar
design
principles
from
the
other
draft
and
with
some
extensions
which
are
very
unique
to
5g
system.
L
So
here
is
a
kind
of
an
overview
of
the
updates
that
are
there
in
this
draft
compared
to
the
previous
version,
which
was
0-2
so
here,
a
new
section,
5.2
2.3
has
been
added
which
again
borrows
the
similar
design
principles.
From
section
four
of
the
LTE
4G
track
up
to
0-3,
we
will
I
think
we
have
a
slide
which
describes
some
details
about
this
section
and
also
the
section
6
which
is
new
and
which
replaces
the
previous
section
6
in
version
0
2
and
some
of
the
sections
are
left
open
for
future
updates.
L
Now
a
few
slides
on
the
recap.
So
this
is
the
5g
architecture
and
I
think
the
important
point
here
is:
this
draft
basically
uses
some
of
the
5g
design
principles
listed
below
like
a
service
centric
design,
distributed
land
support,
control
and
user,
pin,
split
decoupling
of
flat
and
core
network
and,
of
course,
this
non
IP
PDU
sessions,
support
and
also
this
is
the
ICN
X
control,
plane,
extensions
or
5g
system
that
will
also
describe
earlier.
D
L
Is
just
presented
here
for
a
reference
now
in
this,
this
new
section
that
is
5.2
dot
3
is
which
is
the
new
section
here,
which
kind
of
borrows
the
concept
from
the
LTE
draft,
which
shows
our
dual
stack
and
where
it
hide.
It
basically
shows
how
ICN
and
IP
dual
stack
is
deployed
over
the
five
disks,
and
there
is
a
transport
convergence
layer
that
has
been
described,
so
it
can
support
specifically
IP,
as
well
as
ICN
services,.
L
Then
the
new
section
6,
that
is
in
version
0-3,
describes
how
ICN
can
be
deployed
over
a
5g
land
type
service
and
in
that
some
of
the
details
that
are
described
there.
Some
of
the
key
features
are
something
which
is
termed
as
a
path
based
forwarding
over
the
NX
interface
and
to
give
you
a
very
high-level
overview
of
this
method
is
the
path
between
the
source
and
destination.
L
Now,
as
a
name
service
based
transaction,
where
Internet
services
are
now
interpreted
as
the
main
unit
of
transfer
to
be
routed
over
a
Nicean
layer,
then
there
is
also
a
four
layer
flat
protocol
stack
that
is
described
and
mapping
of
the
Internet
services.
For
example,
HTTP
to
ICN
layer
is
described
so
this
it
describes
how
the
icin
layer
uses
an
interaction
with
something
called
a
named
resolver
to
register
and
discover
HTTP
based
services,
and
there
is
the
last
one.
L
Is
this
Salvy
service
proxy
devices
that
interfaces
to
legacy
devices
and
peering
networks,
so
in
this
section,
is
also
shows
and
how
these
IP
based
services
over
ICN
and
is
enabled
and
it
it
describes
the
control
plane,
as
well
as
the
data
plane,
and
it
assumes
an
end
to
end
land
connectivity
to
be
provided
by
solutions
such
as
v
chillin.
I
think
this
is
the
last
slide
or
conclusion
the
next
steps.
Of
course
the
author
also
wants
to
collect
feedback
from
the
from
the
group
and
some
of
the
proposed
next
steps.
L
Are
we
of
course
address
some
of
the
comments
that
are
received
in
this
meeting
and
then
update
section
6
on
ICN
over
5
GN
architecture,
for
example,
ICN
flow
management,
mobility
handling
and
ICN
over
5
GL
n,
and
we
plan
to
have
a
demo
in
the
next
ITF
105
I.
Think
that's
it
from
presentation.
Any
comments,
questions
I'll
be
happy
to
at
least
try
to
answer
or,
if
not
I
will
note
it
down
and
pass
it
on
to
the
authors.
A
K
A
So
we
pared
it
down
to
Indian
and
CCN,
which
also
caused
a
bunch
of
problems
in
terms
of
use
of
terminology
that
wasn't
identical
or
even
sometimes
in
conflict
between
the
two
and
what's
happened
since
is
that
much
of
the
things
that
caused
in
the
end
to
have
different
terminology
from
CCN
have
gone
away
because
Indians
changed
their
architecture
and
now
is
sufficiently
similar
to
CCM
that
the
terms
actually
mean
the
same
thing.
So
here's
some
some
some
document
history,
it's
been
through
a
number
address.
A
It's
actually
had
fairly
wide
review,
so
you
can
see
how
it
went
through
here.
So
there's
been
a
sort
of
a
set
of
comments
from
a
few
people
of
the
last
version
of
the
draft.
There's
now
a
security
consideration,
section
yay.
So
those
chairs
don't
believe
there
are
any
issues
left
and
our
plan
is
what
we
would
like
to
do
is
to
do
a
IRS,
GPL
IRS,
a
review
of
the
draft
and
send
it
on
toward
toward
publication
and
unless
I
will
issue
a
message
on
the
mailing
list
for
any
basically
to
see.
A
A
We
have
about
just
under
an
hour,
which
is
great,
because
today,
there's
been
very
little,
work
done
to
figure
out
how
to
do
quality
of
service
differentiation
for
ICN
traffic,
reflect
about
it.
A
bunch
I've
given
a
number
of
presentations
at
former
interims,
and
we
now
have
at
least
two
drafts-
are
proposing
solutions
to
several
aspects
of
this.
A
So
this
is
a
this
I
think
is
our
first
opportunity
for
a
with
enough
time
to
have
a
nice
high
bandwidth
discussion
to
get
some
advice
from
working,
our
G
participants
and
stuff
is:
how
important
is
this?
Should
we
bother?
Are
we
wasting
our
time
and
if
not,
if
we
aren't
wasting
our
time?
What
direction
we'd
like
to
take
the
work
in
so
I
think
this
is
a
kickoff
final.
Kick
officers
for
some
more
concrete
work
in.
D
D
D
So
we
presented
an
idea
about
us
in
last
IETF
meeting
in
Bangkok.
Prakash
did
that
and
today
we
are
where
we
have
put
up
to
the
this
document,
or
the
draft
containing
this
outline
here.
So
roughly
the
three
important
sections
it
is
like
divided
in
the
name,
base,
cures,
design
and
essentially
the
cure,
smarter
design
and
how
the
encoding
will
happen
in
the
message.
The
second
thing
is
about
the
network
procedures
and
and
how
the
various
I
scan
it
for
elements
like
consumer
forward
are
going
to
react
to
this
particular.
D
So
as
a
motivation
for
the
Q
nucleus
design
is
the
need
for
defining
the
consumer-driven
q
s
mechanism,
essentially
the
when,
when
users
are
or
requesting
for
the
content,
it
is
the
quality
of
the
service
which
is
user
is
going
to
demand
depending
on
the
the
type
of
service
e
is.
Actually
you
know,
executing
of
the
type
of
the
application
is
executing
the
kind
of
service
plan
or
the
service
he
is
subscribed
with
the
customer,
the
service
provider
and
so
on,
and
from
that
perspective
the
second
motivation
factor
is
about
the
design.
D
D
Now
the
awareness
of
the
Q
s
context
is
also
with
the
application
in
the
sense,
as
I
said
earlier
like
if
application
is,
is
trying
to
do
a
simple
video
download
or
he
is
trying
to
participate
into
a
real-time
video
conference
so
depending
on
the
type
of
service,
the
the
quality
of
service
requirement
would
change
and
then,
in
that
context,
is
only
available
with
application
and
that's
reason.
The
EQs
mechanism
has
is
basically
won
by
the
cont
now
aspect.
D
It
is
use
it
to
the
to
the
us
treat
me
in
the
sense
identification
of
flow
is
certainly,
you
know,
is
important,
but
as
to
once
the
flow
is
identified.
What
kind
of
frequent
is
is
to
be
know
assigned
to
the
particular
flow
is
so
that
perspective
the
flow
classification
draw
mechanism
that
is
already
power.
You
know
available
for
the
review,
as
well
as
this
Q
s
treatment
mechanism.
D
So
what
we
propose
from
a
name
based
us
treatment
perspective
is
that
we
propose
a
new
type
of
a
name
segment,
leveraging
TLB
based
design.
So
essentially
we
have
the
tagging
value
where
the
type
of
TLP
we
can
you
know
live,
is
the
type
and
then
introduce
a
a
new
metadata
field
or
into
the
into
the
name
container
name
now,
and
then
it
would
allow
for
allow
for
then
in
place
processing
of
the
interest.
D
So
what
I
mean
by
this
in
place
processing
is
that
went
along
with
the
content
name
and
we
have
the
metadata
associated
with
it.
So
you,
the
forwarder
or
the
the
given
network
router,
is,
is
having
the
whole
information
within
the
content
name
itself,
and
the
other
aspect
is
that
the
content
name
is
the
mandatory
field.
D
Unlike
the
the
optional
hop
to
operatives,
that
I
mean
that's
another
possibility
where
we
can
put
the
qsr
treatment
information
of
the
cues
marker
information,
but
in
that
from
that
sense,
from
at
least
our
trap
prospective
we
are
proposing
to
have
at
in
in
the
content
name
itself
and
the
surface.
A
very
similar
mechanism
is
already
widely
accepted
and
used
in
HTTP
based
you
are
elsewhere.
D
There
are
certain
things
like
you
know,
called
path
parameter,
so
the
query
parameter.
So
that
mechanism
is
oddly,
it's
proven
from
that
perspective,
and
there
are
two
aspects
here:
cell
name
base
once
we
encode
the
qsr
marker
into
the
name
then,
depending
on
whether
the
mutability
of
the
cuse
marker
is
is
is,
is
the
should
is
required
or
not,
depending
on
that
we
can
either
input
it
into
the
content
name
or
encode
it
into
the
into
the
hub
hop-by-hop
idaho,
on
extractives.
D
So
here
is
the
reference
model
of
how
the
name
is
encoding
of
this
few
smattering
will
be
so
we
are
drinking
up
being
the
content
name
into
a
routable
content,
name
component
or
the
part
and
around
non-radical
component
name
now.
The
notable
is
is
the
the
prefix
part
in
which
we
don't
really
disturb
anything.
As
far
as
the
country
fix
procedure,
or
you
know,
processing
procedures
are
there,
so
only
in
the
end
of
the
whole
content,
name
that
it
is,
you
know
that
may
be
provided
by
the
consumer
application.
D
D
Now,
as
far
as
the
design
of
the
cue
s
market
itself
is
concerned,
that
is
still
under
discussion
or
you
know,
enter
with
you
on
our
site
that,
depending
on
whether
you
know
it
should
be
a
flat
naming
scheme
or
or
hierarchical
naming
scheme
in
the
sense
whether
we
will
need
one
name
segment
or
two
next
segment
that
we
are
still
working
on
and
the
design
or
the
actual
encoding
of
it
will
also
be
be
primarily
be
influenced
by
the
type
of
services
and
then
type
of
the
applications.
That
customer
is.
D
D
So
we
will
see
quickly,
some
of
them
a
no
network
element
procedures
years
next
slide.
Please
so
I
will
just
quickly
run
through
this
three
things,
because
I
want
to
spend
more
time
on
the
the
forward
rate
design,
but
but
and
then
I
will
request
you,
you
know,
look
at
the
draft
so
that
we
can
have
more
further
discussion
offline
so
and
then
the
case
of
consumer
procedure.
It
is
aware
of
user,
the
application
and
the
it's
excuse
context.
D
So
it
is,
let
us
say,
assume
the
LPN
process
is:
is
returns
like
couple
of
two
forwarding
interfaces
or
three
forwarding
interfaces
so
based
on
the
QoS
marker
of
the
individual
phases
or
or
the
you
know,
the
US
market.
That
is
there
in
the
content.
The
actual
decision
will
be
done,
based
on
the
Q
s
marker
as
to
which
forwarding
interface,
the
the
interest
rate
to
be
forwarded,
and
then
essentially,
the
selection
of
the
forwarding
interfaces
is
nothing
but
the
actual
selection.
D
Final
selection
of
the
the
Q
that
that
the
interest,
the
FIR
the
packet
will
be
forwarded
or
put
into.
But
the
further
processing
and
and
the
choice
of
the
forwarding
path
essentially
then
becomes
the
problem
of
selection
of
the
forwarding
key
resistance
so
based
on
the
Q
s
marking
in
the
interest.
So
this
is
a
potential
case
of
bid
scaling
that
we
will
see
in
the
next
next
slide.
Any
please
go
to
connection.
D
Yeah,
okay
and
finally,
the
producer
procedure
is
like
producer
fees
will
be
aware
of
the
disaggregation
between
the
out
table
name
and
its
us
marker
or
that
non-routable
component,
and
so
that
it
will
look
up
the
content
into
the
into
the
cash
or
or
its
storage.
Only
looking
at
the
routable
content
name,
so
that
that
that
this
aggregation
is
send
kind
of
identified
and
if
the
content
is
is
found
in
the
you
know,
let's
say,
for
example,
and
the
intermediate
router
finds
the
content
in
the
content
cache.
It
will
also
react
in
the
same
way.
D
D
So
this
is
again,
this
is
the
not
a
final
cure.
You
know
pit
design.
So
essentially,
what
we
are
saying
here
is
that
in
the
existing
design,
where
we
have
interface,
ID
mapping
and
the
content
name,
we
are
adding
a
third
column
called
Kiva's
marker
now,
for
the
sake
of
the
discussion.
We
have
taken
this
illustration
where
we
have
shown
three,
or
rather
two
Caesars
scenarios
where
the
two
or
more
interest
are
received
with
the
same
content
name,
but
with
a
different
Q
s.
D
Marker
on
on
two
different
interfaces,
so
take
a
case
of
interest
number
one
and
interest
number:
you
know
they
are
received
on
this
two
different
faces,
but
they,
the
Q's
marker,
is
different
and
for
the
sake
of
discussion
here
we
are
calling
it
the
Q
s,
level,
one
and
PS
level
two.
So
essentially,
these
u.s.
markers
are
hierarchical
in
nature.
That's
why
the
order
is
inclined.
If
we
consider
them
as
a
flat
naming
scheme,
then
it
means
that
Q
s
master
are,
are
kind
of
independent
having
independent
values.
D
Now
and
in
the
second
case,
where
there
are
two
different
interfaces.
Sorry
interest
with
one
and
three
with
the
I
am
content
with
received
as
a
different
us
marker
on
on
the
same
interface.
So
in
this
case,
phase
1
and
phase
sorry
interest,
1
and
3
I
received
on
the
same
phase,
but
they
are
received
with
the
different
cues
marker,
which
is
level
1
and
level
3.
Now
this
is
possible,
in
the
sense
the-
and
this
is
where
the
potentials
the
scaling
issue
is.
D
Is
that
when
we
receive
the
will
say
receive
the
interest
with
the
cues
lower
us
marker
as
a
first
interest
and
if
you
receive
another
request
from
a
different
user,
but
with
a
higher
QSR
markup.
Now,
in
this
case,
since
the
lower
us
marker
is,
is
would
have
been
forwarded
already
and
now.
The
second
interest
comes
with
a
higher
q-s
marker.
Now
that
cannot
be
aggregated
in
the
sense.
If
we
have
to
give
the
priority
treatment
to
that
but
club,
you
know
interest
request,
it
has
to
be
forwarded
as
well.
D
Otherwise,
then
we
are
not
meeting
the
givers
requirement,
so
in
that
case
the
higher
level
qsr
marker
interest
will
will
also
be
forwarded,
and
that's
where
the
little
bit
of
loosening
of
that
bit
scaling
or
sorry
the
bit
aggregation
principle
that
we
have
today.
That
will
be
loosened
a
bit
and
that's
where
we
are
calling
it
as
a
speed
scaling
issue
go
to
the
next
slide.
Please.
D
So
essentially,
what
we
are
saying
is
that,
with
multiple
interests
for
the
same
content
with
a
different
us
marker
are
received,
and
since
the
qsr
markers
are
different
forward,
a
forward
both
the
interest
upstream
and
then
essentially
creating
two
bigger
increase
into
the
into
the
interest
table
now,
the
net
result
is
that
for
given
content
with
aggregation
has
to
be
loosened
up.
As
I
said
earlier,
proportional
the
number
of
unique
us
markers
and
how
we
can
address
this
potential.
D
You
know
losing
of
the
scaling
issue
is
by
keeping
the
number
of
use
markers
limited
now
again,
as
I
said
earlier,
number,
the
the
type
of
us
markers
and
all
the
the
and
their
treatment
philosophy.
She
is
still
yeah
still
not
done,
but
that
this
is
one
way
of
looking
at
it
in
the
the.
The
other
assumption
is
that
in
real
time
cases
we
may
not
hit
all
this
upper
bound
of
the
Q's
markers
all
the
times.
D
I
mean
it
says
when
there
are
five
levels
of
Q
s,
mark
they're,
not
theoretically,
it's
possible,
but
maybe
practically
in
real
time
case.
Not
all
five
u.s.
marking
level
would
be
well
would
be
received
for
four
same
given
context
so
that
that,
but
that's
an
assumption
and
we
can,
by
using
a
hierarchical
or
partial
order
of
Q
s,
treatment
design
we
can.
We
can
reduce
the
the
scaling
requirement.
D
And
this
is
the
way
the
the
higher
the
power
the
her
articles,
the
u.s.
marker
or
the
ordered
us
marker
base
design
will
help
us
to
avoid
this
bit
aggregation
to
an
extent.
So
what
the
algorithm
here
is
that
we
will
forward
the
second
or
the
duplicate
interest
only
if
it
receives
with
the
highest
us
he
was
marking.
D
So
if
any
interest
that
we
see
is
with
the
lowest
us
marking,
for
which
an
interest
is
already
pending
with
the
highest
us
marker,
then
we
will
not
forward
that
in
particular
interest
and
and
the
similarly,
when
we
receive
the
content
for
that
particular
interest.
In
that
case,
it
will
satisfy
all
the
interest
which
which
are
pending
below
below
that
u.s.
marker
level
or
or
we
can.
D
Alternatively,
we
can
also
say
that
any
any
interest,
the
sorry
any
content
that
we
receive
with
us
marking
will
will
satisfy
all
the
interests
that
are
pending
ultimate
because,
ultimately,
the
purpose
is
to
serve
the
content
to
the
user,
but,
depending
on
the
key,
was
marking
our
next
slide.
Please.
D
The
forwarded
behavior
on
the
data
arrival
is
there
are
two
the
arrival
of
the
data
packet
to
satisfy
all
the
pretend
trees
against
the
content
name
in
addition
to
the
cues
markers
in
the
data
packet,
and
there
are
two
possibilities,
depending
on
the
type
of
us
mark
marker
aggregation
that
we
do
in
the
pit.
So
as
I
said
earlier,
so
if
the
application
is
done
on
with
the
highest
us
marker
level,
then
we
can
status
satisfy
all
the
pending
interests
with
respect
to
Y
of
the
received
QSR
marker,
or
we
can
aggregate
the
e.
D
We
were
smoking,
but
these
these
are.
You
have
to
be
evaluated,
so
this
quick
or
you
know,
comments
on
whether
the
US
marker
design,
whether
it
should
be
hierarchical,
all
flat
and
then
I,
have
already
covered
some
of
these
as
points
that
declared
naming
scheme,
design
of
a
cue
smarter
may
not
help
the
optimize
the
loss
of
the
interest
aggregation
at
bit
because
now,
from
the
flag
perspective,
each
HQs
marker
is
independent
of
each
other,
so
there
is
no
higher
or
lower
asset.
D
So
at
from
the
aggregation
perspective,
let's
say
if
we
have
five
US
marker
define
so
there
and
if
we
receive
five
different
cubes
must
better
be
having
five
interest
in
the
network
rather
than
one
which
is
like,
which
is
today
with
the
cue
swap
and
from
a
hierarchical
post.
Depending
on
what
end
of
era
we
come
up
with,
we
will
have
the
relative,
less
relatively
less
number
of,
in
fact,
off
yonder
aggregation
and
following
factors
will
govern.
D
The
selection
of
the
type
of
qsr
markers
here
is
the
type
of
service
the
user
is
subscribed
with
the
service
for
order
and
in
the
class
of
services
or
the
whether
the
user
is
doing
the
Lea
conferencing
conferencing
or
he
is
just
doing
the
progressive
download
and
so
on,
and
then
RFC
four
five
nine
four.
They
list
down
the
class
for
service
classes
that
we
can
leverage
or
we
can
refer
to
next
slide.
Please.
D
So
this
is,
these
are
the
advantages
that
we
are
seeing
here:
the
flexible
scheme
by
virtue
of
disaggregation
of
the
audible
main
component
and
non
notable
name
component,
and
so
that
both
the
information
is
available
to
the
router
rack
when
it
is
processed
in
the
P
content,
name
or
the
name.
Graphics.
Both
are
available
at
the
same
time,
so
that
it
doesn't
have
to
look
up
elsewhere.
D
D
D
The
independence
of
routable
name
component
from
the
cue
s
marker
is,
we
are
saying
that
is
not
going
to
be
at
the
flip
scaling
asset
because
we
will
not
be
any
change
into
the
field
table
as
we
have.
We
are
proposing
the
change
in
the
pit
table
and
then
there
is
an
opportunity
also
increment.
The
cue
is
the
were
forwarding
decisions
as
now,
since
the
the
content
name
is
going
to
have
the
Q
s
mark
inside
it
next
slide.
Please.
D
So
this
is
the
last
leg:
oh
I'll
just
cover
the
future
work
items
here
is
about
the
evaluate
the
performance
of
this
name
based
us
marking
scheme
on
on
the
pit
scaling
or
various
the
scaling
issues
that
we
have.
We
have
to
study
the
empirical.
You
know
performance
and
evaluate
the
impact
of
on
the
pit
aggregation
and
the
advantage
of
the
the
optimize.
Some
of
the
optimization
that
we
talked
about
and
the
design
of
the
naming
scheme
of
the
q
is
marking
itself
on
our
markers
itself.
D
That
is
something
pending
or
not
near
done
and
explore.
If
house
mapping
scheme
can
also
be
used
for
classification
now,
this
is
just
an
exploration
part.
We
don't
have
any
evidence
of
it
and
explore
the
use
case
of
in
network
nutrition
in
the
q
s
marker
and
depending
on
whether
we
need
to
do
the
mutation
or
not
and
depending
on
what
kind
of
values
we
can
get
from
that
particular
feature,
then
the
decision
or
the
design
of
the
Q's
marker
will
also
be
will
be,
will
be
derived
or
will
be
influenced
based
on
that.
D
So
this
is
it
for
now
in
this
version
of
the
truck,
and
we
will
be
working
more
closely
with
the
community
and
then
we
will.
We
will
already.
The
email
has
some
of
the
interactions
between
us
and
then
Dave.
He
has
posted
his
comment
and
there
is
some
field
event
that
we
also
added,
and
we
request
all
the
community
here
to
read
the
draft.
It
is
still
work-in-progress.
This
is
just
first
version
and
we
will.
D
E
I
That's
one:
that's
so
Jenny
handbook.
So
could
you
bring
up
a
slightly
creeping
up
the
slide
again
that
the
Pitt
issue
so
where
we
have
this
table
and
the
table
yeah?
So
I
have
a
question
here,
and
so
we
are
this:
a
good
rating,
the
pitch
entries
and
we
have
a
new
new
column
that
the
chorus
marker
is
there
actually
a
table
that
the
maintaining
with
that
course
marker.
So
can
we
just
use
any
any
course
markovia.
I
D
Sorry,
so
this
table
is
just
a
notional
here
right
now,
so
this
is
not
the
real
design
and
then,
as
far
as
the
cues
level
mark
use
markers
are
concerned
again,
as
I
said
earlier,
the
the
the
design
of
the
cue
s
market
itself
is
not
yet
complete.
So
the
the
numbers
are
just
numbers
here.
They
don't
really
mean
anything
at
this
point
of
time
to.
I
D
A
Let
me
jump
in
here,
I
think
I've,
already
communique
this
the
drafts
off
authors.
This
is
a
completely
unreasonable
bit
design,
but
there
is
there
is
a
reasonable
fit
design
that
doesn't
do
it.
This
way
that
doesn't
have
scaling
problems
and
I
think
I
partly
explained
it
in
my
comments
and
their
draft.
Oh
yes,.
E
I
I
What
is
the
picture
that
we
come
from,
so
we
have,
of
course,
this
constrained
IOT
concert
net
black
background
and
in
our
use
cases
notes
always
have
limited
resources,
limited
resources
in
form
of
memory
in
front
of
CPU
power
and,
of
course,
low
energy
capacities
and
hourly
players.
If
we
use
typically
a
very
high
delays,
agencies
restrict
them
to
use,
and
we
have
you
shared
wireless
media,
so
we
can't
use
the
media
all
the
time
we
want.
We
have
to
share
it
with
other
people
other
nodes.
I
So
in
this
environment
we
have
resources,
but
these
resources
are
no
relation
to
with
the
participants
and
other
participants
that
are
actually
there.
So
we
have
to
show
the
limited
resources
with
other
nodes
and
all
thoughts
on
this
is
the
first
well
categorize
in
the
draft.
What
do
we
actually
have
for
resources?
I
What
resource
so
very
important
for
for
us
and
for
us
I
mean
and
also
in
the
IOT
and
moment,
and
of
course
the
first
thing
is:
you
have
to
think
about
the
link
layer,
resources
I
mean
here,
I
mean
like
the
media
access,
and
this
is,
but
you
have
office
space
for
forwarding
juice,
which
is
independent
of
the
limpio
technology
itself,
and
here
we
can
do
some
optimizations
at
the
zoo
later,
and
we
also
have
pick
resources
and
resources.
You
can't
decide
should
apparently
be
the
sort.
I
I
Decisions
for
short-term
action
and
placement
and
the
third
categories,
the
contents
or
resources
we
can
also
we
have
maps
we
can
control
about
where
we
want
to
deploy
the
content.
So
we
have
the
first
decision,
the
cache
decision,
what
we
deploy,
but
we
place
the
content
number
of
decision,
which
I
mean
separated
environments
which
actually
place
if
he's
swimming
into
the
nuclear
resource
management.
First,
so
it's
like
I
said
we
have
like.
I
So
we
can
configure
the
number
of
time
slots
if
you
have
to
network
a
number
of
channels.
If
you
have
a
network,
we
have
control
the
latency
of
packets
a
fee
if
you
do
like
a
priority
based
forwarding
and
using
the
40
queues
and
something
that
we
might
want
to
work
on
later
management,
where
we
can
based
on
priorities,
for
example,.
I
If
we
go
to
the
the
pit
resource,
which
is
more
interesting.
This
is
the
higher
level
part,
and
here
we
have
actually
we
can
decide
only
whether
the
traffic
should
be
like
latency
should
have
like
a
low
latency
over
the
trip
to
the
latency,
of
course,
the
reliability.
So
you
have
to
get
dimensions
here
that
we
can
control
with
bit.
I
can
evict
a
pit
entry
based
on
on
those
priority,
so
we
can.
I
Yeah,
we
can
just
drop
the
incoming
interest
if
there,
if
the
priorities
are
to
over
this
and
key,
if
you,
of
course,
we
have
Kanpur
coverage
measurement,
management's
the
ideas,
but
this
is
for
future
work.
This
is
whether
we
want
to
propagate
this
information
on
the
path
or
not,
thoughts
on
the
content,
store
itself
and
I
mean
there's
already
a
lot
of
caching
strategies
and
cache
decision
strategies.
I
But
if
we
have
a
prioritized
traffic,
then
we
can
also
decide
on
which
dimension
we
want
to.
We
want
to
optimize
for
latency
right
ability,
or
both
maybe-
and
this
is
also
them
yeah.
This
requires,
of
course,
something
yeah
I
mean
we
have
to
decide
on
a
placement
of
replacement
for
this
in
separated
environments,
so
there
are
actually
I
mean
if
we
have
the
recipe
for
building
a
course
and
probably
like
this,
we
have
a
traffic
flow
classification.
I
I
The
classic
creation
itself
is
quite
simple,
and
this
is
what
we
want
in
the
IOT.
It
should
be
simple
and
we
don't
want
to
put
more
information
into
packets
than
we
need,
and
here
just
we
are
actually
maintaining
a
list.
This
is
why
I
asked
the
question
before
we
are
maintaining
a
list
of
valid
classes,
and
then
we
have,
of
course
entering
the
incoming
interests
in
data
tickets.
So
we
can
obtain
obtain
a
name
TLV
from
post
packets
and
we
can
do
a
long
as
prefix
match
all
those.
A
I
If
you
do
it
this
way-
and
this
is
what
we
want
from
here
once
we
have
the
class
we
are
defining
in
the
in
the
in
the
draft
that
came
out
yesterday,
we
are
defining
at
these
two
dimensions
that
are
important,
especially
in
the
IOT,
which
is
latency
and
reliability,
and
here
we
have
two
different
priority
levels,
which
is
for
latency,
regular
and
ex
parroted
for
reliability.
We
have
Agra
unreliable
and
those
categories
are
those
priorities
actually
decide
on
how
we
evict
pattern?
Trees
have
the
evict
contents
or
entries.
I
K
Don't
have
a
question.
I
would
like
to
make
a
remark,
because
I
mean
the
submitting
that
drafts
are
late,
as
yesterday
has
had
some
reason
of
lengthy
discussions
with
shank
and
other
and
the
other
office,
and
the
discussions
about
the
subject
itself,
and
what
we
found
is
that
actually
the
QSI
mentioned
in
the
ICN
is
interesting
because
you
have
more
interesting
than
in
the
IP
world,
because
you
have
these
different
mechanisms
which
you
have
and
what
is
even
more
interesting.
K
Is
that
you
actually,
the
big
effects
you
can,
at
least
in
a
constraint
environment
produce,
are
from
the
correlating
these
things,
so
one
one
example:
I
mean
one
old
example.
We
are
as
an
issue
we
already
arose
in
the
early
days
of
the
IC
energy.
Is
the
static
T
correlation
between
bids?
So
if
you
have
three
nodes
and
each
node
evicts
a
bit
entry
that
actually
and
each
vote
takes
another
one,
then
you
have
like
D
correlated
pits
that
are
useless,
and
the
same
is
also
true
on
the
node.
K
If
you
are,
if
you
have
the
cash
and
the
pits
I
mean
you
have
an
interplay.
So
if
you,
for
instance,
evict
something
in
the
pit,
then
you
should
think
what
does
it
mean
for
your
cash
and
so
on?
That
makes
it
actually
very
very
interesting
and
I.
Guess
it's
a
rich
of
here
then
at
least
the
drafts
we
have
written
or
have
read.
A
If
you're
willing
to
see
some
interest
data
exchanges,
I
would
I
would
point
out
that
the
ec3
scheme
can
actually
instantiate
that
state
dynamically
at
very
low
cost
in
in
overhead
on
the
packets.
So,
if
you're
willing
to
not
demand
that
the
QoS
classification
and
enforcement
happened
on
the
very
first
packet,
you
see
from
a
consumer,
you
can
look
at
the
data
packet.
Coming
back,
see
an
EC
three
more.
You
see
three
count
in
there
and
then
build
your
your
your
your
table
or
prefix
and
class
right.
A
So
just
something
to
think
about.
We
may
we
may
be
able
to
sort
of
like
get
the
best
of
both
worlds.
You
might
pre-configure
some
things
that
you
could
also
learn
a
very
low
cost,
as
opposed
to
using
a
CNT,
which
is
the
maybe
I
should
have
given
the
the
classification
talked
first,
but
but
people
have
heard
that
before
so
maybe
something
to
think
about
that
we
might
be
able
to
get
the
best
of
both
worlds
here.
Just.
I
A
I
I
A
A
A
A
C
A
A
I
know
when
go
work
for
Cisco,
but
I
can
I
think
safely
claim
that,
because
of
the
way
cisco
has
donated
or
the
IPR
on
CCN
to
be
royalty-free,
blah
blah
blah
blah
all
that,
based
on
normal
IETF
rules,
I
think
it's
now
they're,
not
long
no
longer
representing
Cisco
out
they'd
say
for
me
to
tell
people
that
you
know
I,
don't
think
this
is
gonna.
This
is
a
problem
necessarily,
but
it
has
to
be
disclosed.
A
The
second
is:
how
do
you
group
traffic
into
equivalence
classes
so
that
you
can
manage
the
QoS
state
first
until
which
package
should
get
similar
treatments
and
when
you
need
to
do
rate
shapers
weight
limiters
and
other
forms
of
of
ensuring
that
track
sources
of
traffic?
Don't
overrun
their
allocations
in
the
resource
manager?
How
do
you
decide
whose
packets
are
in
which
equivalence
class
to
do
that?
So,
as
I
mentioned,
what
the
traffic
class
treatment
should
be
is
covered
in
this
drift?
A
We
just
heard
about-
and
this
riff
I'm
talking
about
here,
addresses
strictly
the
flow
classification
problem,
so
there's
some
more
fundamental,
deep
problems
in
flow
classification.
These
have
been
around
since
the
beginning
of
time
and
based
on
various
engineering
trade-offs
in.
In
some
cases
there
was
a
physics
they
are
probably
unlikely
to
ever
change.
A
So
if
the
classifications
are
too
finely
granular
either
in
terms
of
how
many
you
have
at
the
same
time
or
how
quickly
they
turn
over,
there's
serious
scalability
problems
in
the
actual
resource
management
algorithms
I
mean
this
basically
has
to
come
down
to
somebody's
writing.
An
input
policer
/
shaper
and
a
scheduler
write
for
the
for
the
resources
right,
and
we
only
have
certain
ways
to
do
that.
You
can't
wave
your
hands
and
say:
oh
you
mean
I,
have
I
have
to
simultaneously
track
75
million
flows,
gee
that
might
not
work
in
a
court.
A
If
they're
too
coarse,
you
can't
separate
traffic
at
a
fine
enough
level
to
have
meaningful
fairness.
So
a
good
example
of
that
in
the
IP
world
is
in
order
to
avoid
tracking
flows
and
one
of
the
popular
algorithms
is
stochastic,
fair
queuing
where
you
just
patch
things
into
a
number
of
buckets,
and
you
ensure
that
if
the
hashing
works
reasonably
well,
one
person
one
person
can
inject
traffic
and
overload
a
single
bucket
that'll,
be
spread
across
the
buckets
and
then
you'll
get
some
amount
of
fairness
and
that
works
relatively
well.
A
If
there
aren't
really
sophisticated
attackers-
and
you
have
enough
stochastic,
your
queueing
buckets
sfq
does
not
actually
provide
you
with
with
good
treatment.
It
just
says
within
a
given
traffic
treatment,
if
I
can't
classify
the
floors
granularly
enough,
I
can
still
do
something
in
a
mediate
between
no
bitching,
simple
class
based,
queuing
and
and
slow
clear
flow
Clos
and
killing
like
full
wfq
priority
queuing.
A
If
it's
not
securely
encoded,
in
which
privileged
class
trust
across
any
kind
of
domain
boundary
is
problematical
and
we've
seen
this
with
IP
dip
surf
right
in
the
sense
that,
because
there
isn't
a
separate
way
to
classify
flows
and
it
isn't,
it
can't
be
managed
the
boundaries
whether
the
service
providers
do.
When
the
traffic
arrives
from
somebody
else's
network,
they
just
stop
learning
right.
They
just
throw
it
away
right
because
they
don't
know
what
to
do
with
it
and
they
don't
want
traffic
coming
in.
A
That
will
take
their
resources
at
a
higher
level
of
usage
than
what
they
locally
want
to
use.
On
the
other
hand,
if
you
securely
encode
this,
it
makes
aggregation
somewhere
between
difficult
or
impossible.
I
I,
don't
have
a
proof
that
it's
impossible
with
my
intuition
is
that
they're
there
does
not
exist,
a
feasible
algorithm
that
take
us
securely
encoded
low
classification
mechanism
and
aggregated
reasonably
so
differences
from
IP.
We
don't
have
a
nice
five
tuple
and
specifically
there
no
source
addresses
right.
A
So
we
can't
use
a
flow
description
matter
like
IP
does
and,
however,
symmetric
routing
allows
the
four-state
to
mirror
the
FID
state,
which
is
very,
very
nice.
You
don't
need
this
totally
independent
data
structure
and
that
tracks
you
know
every
possible
flow
that
could
be
entering
a
router.
You
only
need
to
track
the
things
that
are
active
at
the
moment
and
the
probe
based
interaction
as
cleanly
separates.
A
What
the
producers
desire
to
be
the
specify
of
traffic
equivalence
classes
from
the
consumers
desire
to
control
the
actual
traffic
treatment
in
the
network,
and
there
still
I
think
some
controversy
about
how
absolute
that
separation
ought
to
be
and
I
think
we
should
discuss
it
further
on
I'm
on
record,
advocating
an
absolute
separation
that
only
producers
are
in
charge
of
half
flows
are
classified
and
only
consumers
are
in
charge
of
How.
Tos
treatments
are
specified
so
someone
constraints
here.
A
The
equivalence
classes
have
to
be
bound
somehow
with
the
names
and
the
corresponding
interest
in
data
packets,
so
that
you
can
maintain
stable,
State
across
multiple
intrastate
exchanges
and
be
stable
across
a
set
of
names
sharing
some
common
handle.
So
this
argues
that
it
be
prefix
based
in
some
way
and
that
on
the
classification,
be
communicated
from
the
producer
to
the
consumer,
so
the
consumer
knows
what
the
state
of
affairs
is
on.
A
You
would
like
to
have
you
know
a
routing
prefix
that
looks
like
CNN
or
or
France
gov
right,
and
you
know
an
oil
traffic
going
to
CNN
or
well
traffic
going
to
France
in
a
given
part
of
the
namespace
that,
if
that's,
what
your
fib
has
it's
not
sufficiently
granular
to
do
any
useful
for
classification
and
practical
scale
of
the
routing
needs
to
do
some
kind
of
route
aggregation
right
which
further
blurs
the
discrimination
of
these
equivalence
classes.
So
you
need
something
that
relates
to
the
name.
A
So
so,
we'd
like
a
mechanism
in
the
flow
orders,
that
is
at
least
as
granular,
isn't
a
weldable
main
prefix
and
hopefully
more
granular
without
scalability
becoming
intractable,
the
lightweight
encoding
and
some
decent
security
trade-offs,
because
there's
no
there's
no
scheme
that
we
I
could
come
up
with
with
earlier
the
other
people.
I
talked
to
that
on
both
very,
very
highly
secure
and
as
the
other
properties,
you
want
so
some
thoughts
on
scaling
this.
What
what
state
do
you
have
to
keep
on
a
per
flow
basis
when
the
flow
count
is
very
high?
A
We
know
that
you
can't
build
routers,
that
power
flow
state
when
you're
running
millions
of
flows
through
them
for
producers
and
consumers,
as
opposed
to
the
routers.
The
state.
The
state
scale
is
naturally
with
a
number
of
applications
and
application
interactions
that
are
going
on
simultaneously.
So
we
would
argue
that
you
don't
really
need
to
worry
about
how
much
state
needs
to
be
kept
around
flow
classification
for
consumers
and
producers.
A
That's
not
the
limiting
factor
on
if
an
ICN
forwarder
can't
keep
all
the
state,
and
it
faces
this
common
problem
of
which
flows
to
remember
and
which
flows
to
forget
or
on
how
to
take
a
set
of
flows
and
without
keeping
poor
flow
State
to
be
able
to
do
better
than
just
throwing
the
state
away
like
doing
stochastic
fair
queuing
of
some
sort.
So
we
don't
solve
these
problems
do
fundamental,
but
hopefully
the
mechanisms
that
have
been
proposed
have
sent
some
value.
A
So
we
propose
two
different
mechanisms
in
the
draft
and
we
do
not
make
any
recommendations
or
or
statements
about
which
one
we
should
adopt,
or
should
we
adopt
both
one
is
a
main
component
count
in
interest
in
data
packets
and
the
other
is
an
equivalence
class
named
component
that
binds
the
equivalence
classes
directly
into
the
namespace.
So
let's
take
a
look
at
the
gülen's
class
component.
How
am
I
doing
on
time?
I
don't
want
to
take
too
much
more
time.
I'll
go
through
this
quick,
pretty
quickly.
A
Part
of
the
name
will
be
like
the
name
of
the
video
and
then
a
low
and
then
it'll
be
said.
Oh,
this
is
the
audio,
and
this
is
the
video
and
if
you
count,
if
you
set
your
your
count
to
be
as
long
as
the
discriminator
discriminates
audio
and
video,
you
have
now
separated
audio
and
video
into
two
separate
equivalence
classes,
which
means
you
can
give
them
different.
A
Alternatively,
we
can
actually
put
the
equivalence
class
markers
I'm,
gonna
use,
but
we
can
encode
the
desire
to
Berlin's
class
directly
in
the
name
by
adding
a
name
component
to
the
names
of
the
content
objects,
and
so
this
is.
This
has
some
interesting
properties.
The
first
is
it's
a
minima
for
the
lifetime
of
the
Associated
data,
which
we
could
talk
about,
but
seems
to
be
an
attractive
property
for
when
somebody
generates
data
they
know
what
that
equivalence
class
it
ought
to
be
in,
and
it
can
stay
that
way.
D
A
A
The
other
thing
is
that,
because
this
is
in
the
name
and
and
controlled
by
the
producer,
or
you
may
be
able
to
use
this
for
load
sharing
and
QoS
driven
demultiplexing
of
interests
of
traffic
arriving
at
at
producers
on
one
other
thing,
I
mentioned,
that
is
in
the
draft
with
isn't
in
the
slides.
Is
it's
nice
to
be
able
to
use
these
equivalence
classes
to
control
caching
on
algorithms
as
well
and
discriminate?
So
a
good
example
is:
you
could
have
a
cache
resource
manager
that
doesn't
just
evict
content
randomly
based
on
LF.
A
You
are,
or
LRU
or
whatever
algorithm
is
used,
but
if
you
can
say
that
if
I
have
just
evicted
a
packet
in
this
equivalence
class
for
a
user,
I
might
as
well
evict
all
of
them
or
evict.
You
know
if
it
concentrate
my
eviction
up
for
cache,
based
management
or
an
equivalence
class
basis
right
so
I
don't
spread
my
cache
evictions
randomly
across
all
of
the
traffic,
all
the
content
that
I'm
that
I'm
caching
right,
oh
and
hence
have
reduced
cache
behavior
across
all
of
the
all
of
the
users.
A
So
we
can
use
this
for
cache
control,
as
well
as
for
resource
control,
as
well
as
for
link
resource
control,
so
consumers
can
associate
arrived
in
data
packets
with
the
correct
equivalence
class
and
then
use
that
to
manage
subsequent
interest
data
exchanges.
Now
the
set
of
consequence
of
this
is
if
the
consumer
doesn't
doesn't
know
ahead
of
time.
What
this
is,
he
has
to
do
some
interested
exchange
in
order
to
learn
the
information.
A
The
other
thing
is
associated
measurements
of
things,
coming
back
like
RT,
T's
and
marginal
delay.
They
can
do
help
perform
flow
and
congestion
management
for
the
equivalence
class
as
a
whole
in
the
consumer.
So
the
consumer
is
not
just
on
metering
his
traffic
on
a
per
object
per
interest
data
basis,
but
can
design
his
flow
control
metering
across
the
entire
equivalence
class
photos.
They
need
flow
instance,
granularity
data
structure
in
order
to
support
it
on.
A
Typically,
these
are
more
granular
than
the
prefixes
in
the
field,
but
less
granular
than
the
names
in
the
pic
which
sort
of
says,
oh
god,
my
router.
It
needs
another
data
structure
that
doesn't
exist
today,
so
it
needs.
You
know
you
need
to
actually
design
this
and,
as
I've
said
a
number
of
times,
there's
no
magic
pixie
dust
to
sprinkle
on
the
flow
count.
Scalability
problem
right.
So
what
do
you
do
with
all
this?
A
You
can
enforce
a
rate
control
on
the
equivalence
classes
as
a
whole
by
dropping
packets
or
queuing
them
in
two
different
ways.
You
can
estimate
the
number
of
simultaneous
flows,
traversing
bottleneck
links
that
can
improve
performance
in
congestion,
and
you
can
make
more
intelligent
selections
of
which
packets
to
cache
in
an
ICN
forward
or
prefer
cash.
Many
packets
in
the
same
equivalence
class
already
sort
of
alluded
to
that.
So
that's
I'm
done
so.
A
C
A
I'm
not
sure
that
the
the
equivalence
class
thing
by
itself
has
this
property,
but
certainly
introducing
QoS
markers
in
the
absence
of
any
of
some
very
carefully
constructed
limitations
on
how
you
can
use
them
and
what
the
what
the
effect
on
resource
managers
is
could
very
easily.
You
know
have
a
deleterious
effect.
Yeah.
A
C
I
I
think
book
to
get
still
time
for
questions
and
so
turn
back
to
3
D
to
the
e
ce
NCT.
If
I
think
I
had
a
comment
on
on
the
mailing
list
before
so,
basically
what
he
proposes
to
to
get
to
have
a
new
name
component
type
correct
for
anemia
so
and
in
CC
mix,
for
example,
you
have
a
plethora
of
different
type
name
component
IDs
in
the
draft,
so
you
can
I,
don't
know,
I
think
it's
two
bytes
of
unused
name
component
type
space.
But
what
if
I
mean?
I
If
you
choose
to
have
the
chorus
in
the
indicator
and
then
component,
then
application
won't
be
able
to
use
any
other
name
component
type
I
mean
I,
never
use
those
types.
Actually,
I,
don't
know
what,
if
the
divet
the
sensor
bill,
not
to
restrict
applications
from
using
other
types
and
just
generic
name
component.
Yes,.
I
A
A
Say
oh
gee
this
this
this
name
that
says
video
on
captures,
what
I
mean
by
grouping
things
together,
so
I'll
use
the
string,
video
and
but
make
the
name
component
type
equivalence,
class
or
or
if
I
have
an
existing
name.
Space
and
I
want
to
cut
put
a
cut
between
the
part
that
should
be
treated
together
in
the
part
that
you'd
be
treated
separately.
I
insert
a
name
component
type
in
so
it's
in
the
middle
right.
A
A
Excuse
me
doesn't
have
to
be
I,
don't
I,
don't
know
a
good
alternative
other
than
making
it
part
of
the
name,
space
design.
And
if
it's,
why
the
name
space
design
it's
a
mechanical
process
to
either
use
that
name
component
type
and
in
the
existing
name
structure,
name,
cinta,
name
syntax
or
you
know,
insert
it
in
the
middle
hmmm.
A
You
see
implications
and
yeah
I
see
applications,
because
it's
bad
idea
to
request
data
by
prefix,
okay,
things
so
that
but
I
think
you
raise
a
good
point,
which
is
that
how
this
explicit
you
CNT
style
flow
classification,
interacts
with
prefix
fetch,
but
in
an
Indian
environment
is
something
I
haven't
thought
about
all
right,
and
if
we
want
to
support
that
generally
in
the
architecture,
we
have
to
think
about
it.
Okay,
thanks,
if
you
want
to
think
about
it,
be
great.
Alright,
ok!
So.
E
A
E
J
B
N
Okay,
so
thanks
for
staying
around
to
the
end
of
ITF,
this
is
a
call
not
for
work
necessarily
in
this
work
in
this
research
group.
It's
actually
something
called
from
the
European
Commission.
So
it's
to
call
on
those
that
are
doing
research
in
the
space
to
reach
out
to
the
next
generation
and
in
that
initiative
and
seek
funding.
So
we
all
know,
hopefully
the
minute
but
we're
proposing
to
build
the
next
generation
of
the
internet,
somehow
encompassing
European
values
as
safe
for
humans,
and
you
know
if
it
was
a
washing
detergent,
even
bigger
and
brighter.
N
So
this
is
there's
currently
four
open
cause.
These
are
the
grant
agreement,
numbers
that
are
running
for
these
projects
and
I
worked
for
an
organization
called
giant.
So
there
was
this
larger
ecosystem
of
people
consulted
a
challenge
as
the
association
of
research
networks,
the
ripe,
NCC
and
Center
were
also
consulted,
including
nerd
Society
and
even
people
of
the
ITF,
and
this
was
conducted
by
Internet,
Foundation
and
Gartner.
They
did
this
consultation
whittled
down
some
topic,
and
this
has
resulted
in
a
pre
nji,
a
funding
call
which
will
soon
return
to
a
flagship
funding.
N
Call
of
the
European
Commission,
and
so
they
created
this
vision
document
you
can
find
it
if
you
go
to
ng
agony,
you
slash
vision,
and
this
is
to
try
and
build
the
next
iteration
of
the
internet.
You
know
with
centered
on
human
values,
and
so
these
are
the
pillars
that
I
think
that
the
future
internet
will
be
built
as
part
of
that,
and
it
starts
obviously
just
with
infrastructure,
and
that's
you
know
to
be
decided.
N
Is
this
going
to
be
a
parallel
coexistence
on
current
infrastructure,
new
layers
added
in
order
to
use
the
current
infrastructure
that
we
all
know
and
love
or
sometimes
don't
love,
or
will
it
be
entirely
new
protocols
built
to
to
support
that
network?
So
at
the
moment,
funding
comes
theirs
for
funding
calls
in
operation.
N
Obviously,
this
deadline
is
very,
very
close.
So
if
you
were
interested
in
in
j00
discovery
or
the
privacy
enhancing
technology
open
coin,
that
don't
weigh
these
open
calls
cycle
very
quickly,
so
they
close
on
April,
1st
I,
also
open
a
game
there,
April
1st
and
so
the
the
project
that
I'm
involved
in
is
NGO
trust.
N
N
Execution
is
a
two-thirds
one-third
financing,
so
you
need
to
find
90,000
euros
of
your
own
to
be
matched
with
the
hundred
eighty
thousand
euros
that
we
would
offer
and
the
commercialization
is
a
50/50
co-chair
that
that
doesn't
take
a
an
equity
stake
in
your
in
your
business.
If
you're
going
to
go
the
commercialization
equally,
your
internet
foundation
coordinate
the
other
project,
very
small
amounts
of
money,
5,000
to
50,000
euros,
and
these
projects
cycle
very
very
quickly.
So
every
two
months
in
effect
basically
they're.
Always
they
always
have
an
open
call
open.
N
N
Another
foundation
also
do
one
on
a
search
and
discovery
under
the
same
conditions
and
there's
a
group
called
ledger
that
are
doing
one
on
distributed
data
and
distributed
ledger
technologies.
I
may
offer
up
to
200,000
euros
or
so
before
a
bear
or
a
project
stream.
You
can
find
out
more
on
enjoyed
early.
You
slash
open
course,
if
you're
in
the
u.s.,
since
these
are
European
funded
projects,
you
need
to
have
fat
I
focus
on
Europe
and
Europe.