►
From YouTube: IETF110-ICNRG-20210310-1200
Description
ICNRG meeting session at IETF110
2021/03/10 1200
https://datatracker.ietf.org/meeting/110/proceedings/
A
A
C
C
B
So
if
you're
using
macos
you,
you
can
hear.
D
B
D
D
Hi
everyone
can.
I
hear
me.
B
Still
waiting
for
dave
to
get
his
audio
working
just
a.
B
B
B
All
right,
yeah,
let's
get
started.
Actually
what
we
could
have
done
during
this
audio
test
was
to
think
about
a
note
taker.
You
still
have
some
time
to
think
about
it.
Okay,
yeah
welcome
to
icy
energy
at
itf,
110
yeah
under
kucha.
My
co-chair
dave
ram
is
currently
reconnecting,
yeah,
quick
information
about
the
online
participation.
B
All
the
slides
are
uploaded
in
the
data
tracker.
As
usual,
we
just
use
the
the
meat
echo
queueing
system.
B
So
if
you
want
to
say
something,
try
to
join
the
queue
and
also
feel
free
to
use
the
chat
to
do
your
presentations
so
it'd
be
maybe
collect
technical
questions
and
so
on.
B
I
just
noted
that
the
this
link
here
for
to
the
kodi
md
is
actually
not
correct.
It
seems,
I
think,
the
actually.
The
display
text
is
okay,
but
if
you
click
on
it,
you
are
arriving
at.
I
think
last
meetings
notes
click
just
like
it's
a
caveat
all
right.
We
are
operating
under
the
ietf
ipr
disclosure
rules,
so
in
essence
this
means,
if
you
aware
of
any
ipr,
related
contributions
contribution
in
drafts
presentations
discussions.
B
This
is
the
node
well
and
some
nodes
on
privacy
and
code
of
conduct,
so
the
irtf
has
a
privacy
policy.
The
itf
has
one.
We
advise
you
to
check
that.
B
B
C
B
E
Sometimes
you
see
rc,
published
and
so
on.
This
is
typically
for
documenting.
B
B
F
Meeting
again,
we
need
a
note
taker.
C
B
B
B
B
Great
thanks
a
lot
cedric
westphal
said
I
wanted
here
on
the
chat
we
appreciate
again.
This
is
just
for
taking
notes
for
the
questions
and
and
for
the
discussion.
Okay,.
B
Agenda
today
a
few
interesting
items,
so
we
can
tell
you
a
little
bit
of
about
what
what
happened
recently
on
icy
energy
draft
and
so
on.
Hitoshi
is
giving
us
an
update
on
ccn
info.
B
So
if
you
remember
there
was
a
discussion
about
how
to
deal
with
time
values
and
we
had
some
plans
to
kind
of
unify
the
time.
Tlv
specification,
maybe
also
updating
the
ccnx
specs.
B
So
this
turned
out
to
be
a
bit
difficult
and
yeah,
also
as
a
risk
in
terms
of
holding
holding
up
the
publication
of
icn
lopen.
So
the
authors
have
addressed
that
and
submitted
a
new
version
that
doesn't
have
dependencies
on
any
any
other
updates
to
to
season
x.
If
you
want
to
say
a
few
words
about
this,
let
me
know.
C
C
Well,
it
would
like
block
the
queue
when
finalizing
the
review,
so
we
instead
went
with
the
official
time
tlv
there's
another
rfc
for
the
number
now,
but
it's
called
time
tlv
and
reference
that
instead,
which
also
makes
more
sense-
and
now
it's
again
back
at
the
yeah.
I
think
colin
perkins
like
received
it
again
and
it's
not
progressing.
B
Yeah
great
thanks
for
explaining
that
right,
so
yeah
we
will
work
with
colin
and
the
irsg
to
move
this
forward.
B
Okay,
the
qos
architecture.
Yes,.
E
Yeah,
I
I
have
to
admit
I
I
haven't
had
chance
to
to
read
the
updates
here
in
detail.
I
suspect
it's
probably
ready
to
move
forward.
I
will
hope
to
get
to
it
in
a
couple
of
the
next
couple
of
weeks.
B
Yes,
yeah,
don't
worry
so
we
also.
B
I
didn't
push
this
as
much
as
we
as
we
should
have,
but
let's
get
it
out
now
so
yeah
we
so
we
we
read
it
and
we
think
it's
ready,
but
yeah.
Please
check
it.
B
Okay,
the
qos
architecture,
one!
This
is
a
informational
document
that
talks
about
the
new
opportunities
for
quality
of
service
in
ccnx
like
icn
protocols,
so
that
is
essentially
done
currently
in
the
rc
editor
queue.
B
B
Okay,
all
right,
so,
let's
start
with
our
presentation.
Hitoshi
do
you
want
to
share
your
slides?
If
not,
I
can
also
do
that.
Let
me
know.
H
H
Ccn94
is
implemented
on
routers
to
hold
on
the
respond
request,
repair
message
and
consumers,
which
means
the
end
user's
program
cc
info,
discovers
the
paths
and
the
content.
Caching,
information
in
ccns
such
as
reachability
hub
count,
rdt
on
passing
over
cache
conditions
and
so
on,
as
well
as
multi-pass
condition
ccna
for
define
the
configuration
for
information
disclosure.
H
H
But
actually
we
had
a
lot
of
comments
from
dave
and
paulo
when
we
have
been
working
on
version
4.,
so
from
version
4
to
version
6.
There
are
many
updates
and
the
changes.
For
example,
we
describe
the
differences
from
license
ping
and
traceroute.
H
Define
the
rlc8609
for
both
requests
and
reply.
We
didn't
mention
this
kind
of
specification
in
a
previous
version,
which
is
zero
four.
Now
we
support
these
optional
values
and
also
we
made
some
consideration
on
incompatible,
routers
and
information
provisioning.
This
one.
E
H
Detail
it
many
editorial
changes,
so
the
current
version,
I
believe,
is
really
a
good
shape.
Next
page,
please
so
I
picked
up
this
section.
Consideration
of
the
income
in
capable
routers
and
the
information
provisioning,
so
seasoning
for
itself
already
supports
something
like
administrative
profit,
access
control
and.
H
The
legal
income
in
capable
routers,
we
may
not
clear,
we
might
not
clearly
describe
the
situation
so
various
parts.
We
include
some
description
about
incapable,
routers
and
the
information
provisioning,
and
I
picked
up
sections
three
two
one
one
which
described
the
repressor
brought
here
with,
for
example,
not
reporting
the
following
values,
such
as
object,
size
of
the
account
and
so
on,
are
shown
in
figure
16
that
you
should
leave
for
a
figure
60
on
our
draft
or
do
not
report
these
budgets
due
to
their
policy.
H
In
that
case,
the
latter
substitute
field
to
buy
a
body
of
one.
This
means
actually,
when
allowed,
who
is
not
capable
of
reporting
some
values,
such
as
a
number
of
receivable
interests
or
of
these
values.
According
to
the
policy
receives
the
request.
The
router
sets
these
fields
to
all
one.
So
this
is
something
like
a
louder
hide
or
router.
H
Just
ignores
that
reporting
these
values,
the
value
of
each
field,
will
be
also
all
one
when
the
value
is
equal
to
or
bigger
than
the
mac,
that's
by
the
32-bit
field,
so
which
means
that
if,
for
example,
the
received
enter,
the
number
of
received
interest
is
more
than
40
billion,
then
we
may
have
some
conflict,
but
it's
really
their
case.
We
believe-
and
even
so,
if
the
ccn
info
user
program
invalid,
if
the
these
fields
received
are
set
to
the
value
of
one,
so
we
should
care
for
that
kind
of
situation.
H
Tlb
format
is
a
really
powerful
network
tool
providing
topology
and
in
our
cache
information
and
it
and
reviewed
by
ic,
energy
members
and
other
members,
and
we
also
provided
provide
a
reference
implementation
and
that
is
included
in
c4.
Sephora
is
a
software
platform,
including
folding
demon
csm
on
sample,
so
simple
is
not
just
for
ccna.
For
it's.
H
A
fully
full
set
of
ccna
communication
and
ccn
info
is
already
included
as
a
standard
as
a
default
and
standard
implementation
or
in
c4
and
c4
is
a
open
source
and
the
its
license
is
bsd3
cross
license
and
it's
fully
compatible
with
ccnxtv
message.
So
it's
really
easy
to
use,
and
it's
really
good
to
use
for
your
evaluation
or
enhancement
of
your.
H
B
Anything
on
the
chat,
no
okay,
quick
question
is
so
the
intended
status
for
this
would
be
experimental
right.
Yes,.
F
B
Okay,
so
this
has
been
around
for
some
time
and
has
been
discussed
many
times
and
has
seen
quite
a
bit
of
review
already
and
hitoshi
and
team
already
did
some
updates,
reflecting
that
so
yeah.
We
also
think
this
is
ready
now
and
yeah.
I
would
follow
up
on
the
main
list
regarding
the
last
call.
E
Hey
yeah,
so
so
I
have
to
admit,
I
haven't
read
the
the
latest
version.
If
this
is
targeting
experimental,
I
just
make
the
the
comments
I
usually
make,
which
is
that
make
sure
the
draft
is
clear
what
the
experiment
is.
E
G
A
Would
it
be
if
oh
jesus
there's
that
echo
all
over
the
place,
given
that
this
isn't
a
management
and
instrumentation
tool?
Would
it
be
appropriate
just
to
say
that
you
know
this
is
the
tool
you
use
when
you're
running
experiments
to
do
the
measurements.
E
Yeah,
I
I
think
that
that's
reasonable.
You
just
need
to
type
back
to
you
know.
If
it's
an
experimental
document,
you
need
to
type
back
to
what
sort
of
experiments
people
are
doing
and
what
why
it's
useful
to
have
this
sort
of
tool.
B
E
Just
be
a
additional
document
that
that
describes,
you
know
this
is
a
tool
to
to
support
the
following
experiments.
B
So
I
understand
the
intention
I
mean
this
is
on
the
say
level
of
ping
and
tristart.
I
mean
that
does
a
bit
more,
but
so
it's
like
a
tool
that
enables
experiments.
So
I'm
not
sure,
maybe
that's
what
it
should
say.
Then.
H
I
have
very
basic
questions,
so
this
draft
requires
a
registry
registration
so
that,
because
it's
a
newly
added
several
type
values
and
so
on
for
the
ccnx
lfcs.
So
is
there
any
intention
or
is
any
any
relation
about
experimental
process
informational?
Even
though
we
have
a
request
for
the
ir
type
value
changes?
H
There
is
no
relation
so
okay,
so
my
question
is:
if
this
the
intentional
status
is
informational,
let's
say:
is
it
possible
to
request
to
change
the
new
type
value,
a
ccnx
type,
variable
packet
and
so
on
to
the
iron
registry.
H
So
this
is
a
really
well.
This
is
an
operational
tool
and
management
tool,
but
it's
really.
The
stage
is
really
similar
to
the
so
this
requires
such
kind
of
changes,
so
it
may
be
better
to
continue
as
an
experimental
document.
B
Yes,
absolutely
so
I
suggest
that
I
mean
dave
and
I
can
work
with
you
offline
to
to
to
find
a
good
formulation
for
for
the
research
for
the
experiments
that
this
one
enables.
E
Yes,
and
and
to
be
clear,
I'm
not
suggesting
you,
you
make
significant
changes
here,
no,
it's
not
already
in
there.
I
suspect
this
is
a
a
paragraph
for
most
to
to
highlight,
so
this
is
supporting
infrastructure
for
the
various
experiments.
Well,.
A
My
proposal
is
that
we
start
a
last
call
now.
Yes,
yes,
yes,.
A
B
Yes,
we'll
do
that
and
yeah,
please
everybody
have
a
look
and
let
us
know
what
you
think.
B
Thanks
again,
thank.
H
B
Let's
move
on,
I
think
we
have
janis
next
janis.
Do
you
want
to
run
your
slides
yourself.
F
D
So
how
does
this
work?
Do
I
have
to
okay
application.
D
F
B
F
Okay,
all
right
then
hi
everyone
so
welcome
to
this
talk,
which
is
much
less
kind
of
proposal
oriented
in
terms
of
proposing
a
new
solution
for
something,
and
rather
much
rather
on,
like
open
problem
statements
and
call
for
more
ideas
and
actually
proposals.
F
So
what
I'm
going
to
talk
about
today
is
decentralized,
what
we
call
decentralized
data,
delivery
markets
and
basically,
what
this
means
is
that
it's
a
like.
How
can
we
build
a
decentralized
cdn
based
on
a
peer-to-peer
network,
or
in
other
words,
how
can
we
place
incentives
for
fair
exchange
and
fast
delivery
on
top
of
permissionless
peer-to-peer
networks?
F
So
there
are
several
like
sort
of
sub
problems
into
that
and
I
would
like
to
like
quickly
go
through
some
of
them.
I
think
I
have
much
more
time
allocated,
but
it's
likely
that
it's
going
to
be
much
shorter,
shorter,
so
that
we
also
have
time
for
some
q
a
so
let's.
Let
me
try
to
kind
of
put
the
set
the
scene
first
and
so
that
everyone
is
on
the
same
page
but
feel
free
to
interrupt.
F
If
you
know
something
is
not
defined,
so
we
can
envision
a
network
where
we
have
storage
providers
like
cold
storage
providers
and
those
called
storage
providers
they
have
they.
They
store
like
encrypted
copies
of
data
that
someone
wants
to
publish.
F
Cold
storage
providers
can
also
provide
access,
but
it's
going
to
be
slow
because
they
have
to
decrypt
that
the
content
is
very
likely
only
stored
in
one
of
those
cold
storage
providers
which
might
be
far
away
and
so
on.
So
the
health
storage
providers
come
in
to
kind
of
solve
this
problem.
F
We
can
think
of
that
as
any
kind
of
an
entity
in
the
network
that
has
got
some
storage
and
can
contribute
to
content,
discovery
and
content
routing
system,
but
also
we
need
to
have
kind
of
algorithms
to
talk
about
content,
placement
and
content
copy
selection.
So
when
someone
requests
some
content,
how
can
we
get
it?
How
can
we
get
the
closest
copy?
F
A
website
could
be
anything,
so
that's
roughly
what
the
high-level
architecture
could
look
like
and
I'm
going
to
come
back
to
this
later,
but
I
would
like
to
stay
in
this
slide
for
a
little
bit
and
the
title
of
the
talk
is
decentralized
includes
the
the
component
of
decentralized,
which
means
which
probably
requires
some
more
kind
of
attention,
and
therefore
the
reason
that
I
thought
it's
a
good
idea
to
present
this
to
this
group.
F
Now,
if
we
want
to
have
something,
that's
a
system
where
any
user
can
actually
set
up
and
become
set
up
a
server
and
become
a
hot
storage
provider,
then
you
know
we're
dealing,
not
necessarily
with
trusted
parties.
So
what
is
important
then
here
is
to
have
some
sort
of
guarantees.
That's
when
we
receive
we,
we
make
a
request
to
the
network,
and
then
we
receive
some
content.
F
We
are
being
sensed
the
content
that
we
have
asked
for.
This
is
because
the
network,
you
know,
as
traditionally
in
most
cases,
b2b
networks
are
they.
There
is
no
kind
of
dns
like
authority
that
we
can
say
you
know
it's
going
to
provide
the
right
content.
E
F
And
therefore,
I
think
that
there
could
be
good
ideas
discussed
here
so
trying
to
break
this
bigger
problem
down
into
smaller
parts.
We
have
identified.
We've
done
this
study
a
couple
of
months
ago
within
protocol
labs,
and
we
have
come
up
with
these
different,
the
three
different
kind
of
sub
areas
of
the
problem.
F
One
of
them
is
the
data,
delivery
metering
and
the
fire
exchange,
which
means
basically,
how
can
we?
How
can
we
make
sure
that,
when
someone
is
receiving
some
data
they're
going
to
pay
for
what
they
have
received,
so
that
the
hot
storage
provider
is
going
to
pay
for
the
bandwidth
and
the
memory
and
the
cpu
that
they
have
used?
The
second
problem
is
the
distribution
graph
and
how
you
form
the
network.
F
Basically,
so
you
have
nodes
that
spin
up
a
server
they
want
to
have
they
have
some
storage,
they
contribute
the
network,
people
go
and
publish
stuff
there.
How
do
you
kind
of
do,
content,
placement,
content,
discovery,
content,
routing
and
and
so
on?
F
And
finally,
it's
the
economic
model,
which
is
very
likely
quite
different
to
what
we
have
today
because,
as
I
said
before,
the
these
like
different
nodes,
are
not
really
operated
by
a
single
entity,
but
as
in
a
peer-to-peer
network,
anyone
like
or
tor
like
network
anyone
can
spin
up
a
node
and
start
contributing
to
the
network.
So
there
needs
to
be
some
storage,
some
sorry,
some
economic
model
there
that
says
this
is
profitable
for
those
that
contribute
resources.
F
F
So
I'm
going
to
kind
of
go
into
these
three
a
little
bit
and
identify
talk
about
some
challenges
that
we
have
identified
and
basically
open
the
discussion
for
for
more
ideas
or
pointers
that
might
have
missed
and
so
on.
F
So
there
are
some
desired
properties
here
which
pretty
much
I
have
actually
talked
about.
You
know
we
we
want
to
have
things
like
we
want
to
have
anyone
being
able
to
get
in
the
system
and
start
contributing
in
terms
of
the
metering
and
the
first.
F
The
first
problem,
as
I
said,
the
the
main
like
topic
here,
is
how
can
I
make
sure
that
I
that's
that
I
am
going
to
pay
for
the
content
that
I
have
received
from
the
network
and
therefore
there
is
a
payment
there,
which
must
only
happen
when
the
delivery
happens,
must
only
happen
if
the
sla
that
should
be
formed
somehow
is
met,
and
then
how
can
you
basically
make
sure
that
no
one
misbehaves
there
are
several
like
the
the
challenges
and
requirements
here?
F
Kind
of
the
attack
vectors
if
we
want
to
put
in
this
way,
is
that
we
that
there
is
fairness,
so
we
don't
want
users
to
like
receive
content
without
paying.
We
don't
want
users
to
put
like
hot
storage
providers
to
put
resources
into
the
network
without
getting
paid.
How
do
we
avoid
kind
of
spam
attacks
where
someone
is
going
to
spin
up
a
bot
and
just
request,
like
a
million
times,
one
document
so
that
the
resources
of
the
other
node
is
depleted
and
so
on?
F
So
this
has
been
like
generally.
Fair
exchange
has
been,
of
course,
a
very
well
studied
topic
in
the
past,
and
it
starts
from
the
like
late
90s
with
e-commerce.
So
it's
not
a
very
new
thing
that
is
being
identified
here.
Of
course,
experience
like
what
the
user
sees
like
has
changed
from
what
very
old
studies
have
like
talked
about
and
have
assumed
to
what
is
being
seen
more
more
recently.
F
So
we
want,
we
don't
want
to
have
like
you,
non-user
friendly,
like
setups
the
the
transfer
of
data,
whether
it
is
real-time
like
video
demand
or
whether
it
is
a
large
file,
transfer
the
order,
the
sorry
the
transfer
should
start
immediately.
Ideally,
it
should
be
private
as
well,
so
that
you
know
we
don't
have
leaks
into
the
network
about
information
and
things
that
the
user
is
consuming
as
content
on
their
browser
and
so
on,
and
of
course,
performance.
F
Now,
we've
done
a
literature
review
and
we
have
broken
down
the
the
papers
that
we
have
read
into
these
different
areas.
So
there
are
different
ways
that
someone
can
can
deal
with
that
there
could
be.
You
know,
paper
packets,
kind
of
approaches
where
every
time
someone
receives
some
content,
they
have
to
pay
back
and
only
then
the
next.
F
The
next
piece
of
content
is
is,
is
received,
so
you
can
think
of
that
as
a
kind
of
tcp
packet
arc
thing
where
you
know
we
only
send
the
payment
and
only
then
the
next
chunk
of
data
is
going
to
be
released.
There
is
look
unlock
approaches,
which
means
that
you
know
there
is
a
request
to
the
network.
F
E
F
There
is
always
here
the
fair
exchange
problem
of
you
know
who
is
going
to
make
the
first
step,
and
this
is
something
difficult
to
solve.
Even
you
know,
in
this
day
and
age
like
someone
will
have
to
pay
first
and
if
the
other
party
is
willing
to
screw
them
up,
they
can
still
do
now,
then.
Finally,
of
course,
there
are
like
advancements
and
more
sophisticated
ideas
that
have
come
up
recently.
F
These,
I
think,
the
the
approaches,
such
as
zero
knowledge
contingent
payments.
That's
you
know
where
you
can
overcome
many
of
the
problems.
That's
where
they're
in
the
past,
then
there
is
optimistic
for
exchange,
which
means
that
there
is
like
the
two
parties.
Transacting
are
not
alone
in
doing
that.
F
There
is
somewhere
a
kind
of
third
party
which
both
both
of
the
transacting
parties
have
to
kind
of
trust,
but
at
the
same
time
the
third
party
is
not
there
always
is
there
only
if
some
dispute
happens
and
only
then
they're
going
to
come
in
to
resolve
the
situation.
Perhaps
they
there
is
some
kind
of
collateral
that
is
locked
together
in
this
third
party,
and
you
know
that
the
the
penalties
in
case
of
misbehavior
and
finally,
there
are
well
known
reputation.
F
Systems
where
you
know
the
kind
of
hope,
storage
provider
or
the
other
party
that
you
want
to
transact
with
is
only
chosen
like
is
chosen
based
on
their
reputation
and
therefore
someone
who
is
misbehaving
or
not,
meeting
slas
or
things
like
that,
they're,
basically
never
chosen,
and
therefore
only
those
that
behave
properly
participate
in
the
network.
In
the
end.
F
Now
these
kind
of
summarizes
the
the
first
thing,
though
there
are,
of
course,
many
more
questions
and
much
more
depth
to
go
in
there,
but
perhaps
it's
more
interesting
to
go
to
the
second
topic,
which
is
the
distribution
graph
like?
How
do
you
build
the
network
and
what
are
the
different
like
strategies
that
are
built
in
the
network
in
order
to
make
this
work
from
the
network
layer
network
level
perspective,
so
how's
the
network
for
formed?
How
are
requests
sufficiently
rooted?
How?
F
Where
do
you
place
content
items
and
how
do
you
replicate
them?
So
this
is
very
important
because
you
know
here
we're
talking
about
cdn
and
therefore
it's
not
quite
enough
to
say.
Okay,
there
is
a
copy
of
the
network
somewhere
at
the
other
end
of
the
planet,
because
you
know
the
delay
to
go
then
and
come
back
is
going
to
be
very,
very
high
and
that's
not
what
we're
used
to
nowadays.
F
So
there
needs
to
be
kind
of
a
very
interesting
like
replication
and
like
smart
caching,
mechanisms
behind
that
can,
in
a
sense,
autonomously,
replicate
content
and
put
it.
You
know
where
demand
comes
in,
so
these
are
the
types
of
problems
that
are
involved
here.
F
Of
course,
again
there
are
several
desired
properties,
yeah,
the
content,
replication
and
caching,
as
I
said,
there
needs
to
be
an
economic
model
of
the
economic
model
that
I
mentioned
in
the
beginning,
and
I'm
going
to
go
into
in
a
little
bit
needs
to
support
that.
So
the
all
parties
need
to
be
in
agreement
when
replication
happens
and
extra
storage
space
is
included,
including
the
network.
F
But
of
course
again
the
system
should
be
kind
of
permissionless
in
the
sense
you
know,
as
must
be
peer
systems,
that
you
know
you
can
get
in
the
network
and
start
participating
without
kind
of
asking
someone
to
kind
of.
Let
you
in
so
in
terms
of
the
literature
review.
We've
seen
lots
of
lots
of
past
works
here
there
are
many
peer-to-peer
cdns
that
are
actually
in
operation
today,
and
you
know
the
they
do
deliver
content
they.
You
know
they're
like
from
passwords
from
that
session.
F
That
was,
you
know
the
first.
I
think
the
first
and
very
prominent
kind
of
work
that
puts
peer-to-peer
as
a
kind
of
complement
to
the
more
centralized
infrastructure
that
are
coming
is
running,
and
there
are
many
there
are
many
others
as
well.
So
this
has
been
a
good
source
of
inspiration,
but
the
what
we
didn't
find
there
is
kind
of.
F
How
do
you,
like
all
these
systems,
always
assume
some
centralized
control
over
the
network
like
what
is
being
published
where
it's
being
stored?
How
like,
if
someone
wants
to
get
in
the
system?
You
know
there
needs
to
be
someone
that
is
going
to
let
them
in
in
a.
F
And
so
on,
there
is
always
appear
to
be
like
video
and
demand
literature
which
what
they
did
is.
Basically
it
took
things
to
the
extreme
from
the
point
of
view
of
having
a
video
on
demand,
which
has
got
very
strict
requirements
and
tried
to
find
you
know,
investigate
several
different
algorithms
for
to
make
it
work,
and
you
know
if
you
make
something
work
for
video
and
demand.
F
You
know
this
good
group,
of
course,
is
very
well
aware
of
because
it
takes
many
boxes
on
what
are
the
requirements
here
and
what
you
know
what
we
want
to
achieve
and,
as
I
said
in
the
beginning,
you
know
having
named
data-
and
you
know,
self-certified
data
is
something
that
is
very
least
in
my
opinion-
to
building
these
sorts
of
networks.
F
So
several
different,
well-known
designs
around
that
we
went
into
more
detail.
Of
course,
the
the
most
straightforward
thing
to
do
is
you
know:
there's
a
peer-to-peer
network.
Okay,
there
can
be
a
dht.
You
know
like
very,
very
straightforward
approach.
Of
course
there
are
pros
and
cons
there.
The
dhd
is
very
difficult
to
scale
to
very
big
numbers,
it's
very
difficult
to
make
as
efficient
as
possible
and
actually
meet
the
slas
that
we
want
today.
The
same
is
for
other
systems
such
as
pub
sub,
which
is
potentially
even
slower
again.
F
These
things
work,
but
you
know,
have
to
take
into
account
the
performance
as
well.
Then
there
are
name-based
routing
approaches
that
have
been
investigated
in
the
icn
area
and
they
seem
to
be,
like
you
know,
addressing
many
of
the
problems.
F
So
if
you
have
some
kind
of
hierarchical
type
of
naming
and
you
can
like
do
name-based
forwarding,
then
many
of
the
problems
of
you
know
that
I
mentioned
before
and
many
others
can
be
solved,
so
you
can
do
smart
dashing
across
along
the
path
you
can
have,
multicast
and
and
so
on.
So
this,
together
with
the
data
being
addressed,
is
very
seems
like
a
very
promising
approach.
F
For
that,
and
of
course,
then
there
are
other
types
of
approaches
such
as
dns,
like
system,
or
you
know
the
kind
of
name
resolver
that
cloudflare
has
been
building
recently
and
has
released
recently.
So
so
that's
another
very
promising
thing.
Of
course,
the
the
last
two
are
the
downsides
kind
of
are,
in
some
cases,
they're
less
tested
out
there
in
the
wild
and
therefore
there
needs
to
be
lots
of
testing
before
someone
could
build
on
them
with
high
confidence
yeah.
F
So
there
are
several
different
kind
of
things
like
on
the
technical
side
that
we
went
into.
One
extra
thing
to
to
keep
in
mind.
Is
that
we're
talking
here
about
a
paired
beer
network
and
therefore
I
hope
it
was
clear
it's
like
an
overlay
network.
So
it's
not
necessarily
at
the
network
router
level,
and
therefore
you
know
how
you
manage
that
in
terms
of
the
traffic
that
is
included
in
the
network
is
is
something
that
needs
to
be
taken
into
account.
F
So
if
you
have
name
based
routing,
for
example,
where
things
are
rooted
from
one
like
like
router
in
the
network
overlay
router
in
the
network
across
a
path,
then
this
means
that
these
when,
in
reality
the
direct
underlay
path
would
not
be
the
same,
then
obviously
you're
putting
lots
of
like
extra
overheads
on
top
of
those
overlay
routers.
So
there
is
a
tipping
point
there.
I
believe,
where
you
know
for
say,
unpopular
content.
F
You
know
it
could
be
like
resolution
based
approaches
such
as
dhd-based
or
dns,
based
or
whatever
could
get
better
results
in
terms
of
like
hitting
the
trade-off
between
performance
and
and
overhead
in
the
network.
But
then,
from
a
point
onwards,
things
are
probably
different
for
popular
content.
It
is
worth
doing
by
going
the
name-based
routing
approach
and
being
able
to
like
aggregate
requests
and
do
all
that
great
stuff,
so
yeah.
F
So
that
summarizes
the
second
part
on
the
distribution
graph
and
finally,
a
few
words
about
the
economic
model,
which
are
not
necessarily
directly
related
with
what
this
group
does.
But
if
we
wanted
to
build
this
kind
of
permissionless
ctn,
then
someone
would
have
to
take
this
into
account
and
what
I
believe
is
actually
that
you
know
through
some
like
metadata
that
could
leave
together
with
and
like
the
content
name.
Somehow
these
things
could
be
awesome
manifests
that
could
be
part
of
the
system.
F
These
things,
the
economic
kind
of
requirements,
could
be
integrated
into
into
the
network
model
itself.
F
So
very
briefly,
the
economic
model
includes
the
kind
of
three
different
parts:
the
content
providing
economy,
which
is
the
content
provider
here
who
wants
to
store
content
into
together
with
some
hot
storage
provider.
F
Then
what
you
have
is
the
graph
forming
economy
and
basically
hope,
like
as
a
hot
storage
provider
as
a
router,
an
overlay
router
in
the
network,
who
are
the
other
peers
that
you're
connecting
to-
and
this
has
got
several
different
extensions
there-
that
I
don't
have
the
time
to
go
to,
but
you
can
imagine
the
case,
for
example,
where
there
is
a
content
provider
down
here
that
stores
the
content
with
this
storage
provider.
F
Hey,
I
don't
know
if
you
can
see
my
mouse,
but
I'm
moving
it
anyway
and
then
suddenly
the
content
becomes
available
in
location,
c,
the
hot
storage
provider
at
location
c
because
of
some
demands
from
from
down
here.
F
So
what
do
you
do
now
that
the
the
store,
the
hot
storage
provider
has
got
a
deal
with
the
contents
like
in
the
content
providing
economy
with
the
content
provider,
to
say
that
look
I'll,
be
storing
your
content
and
therefore
you
can
be
kind
of
paying
me
the
what
happens
when
we,
when
we
make
the
network
be
so
that
it
would
be
better
to
replicate
that
to
location
c,
because
it's
closer
to
where
demand
is
so
who
is
getting
the
kind
of
payment
now
for
the
services
that
they
provide
in
the
network?
F
And
you
know
acting
rationally
of
course,
location.
C
storage
provider
doesn't
really
care
about
the
network
about
the
performance
of
the
network.
They
just
don't
want
to
replicate
content
because
they
can
use
that
storage
memory
for
something
else.
At
the
same
time,
location,
a
hot
storage
provider
again
doesn't
really
care
about
the
performance
in
the
network
and
says:
okay.
F
I
can
save
all
the
content,
because
this
way
I'm
going
to
get
all
of
the
money
from
the
like
all
of
that
payment
from
that
content
providing
economy,
I'm
sorry,
the
content
provider
and
therefore
you
know
that's
in
my
best
interest
to
do.
But
at
the
same
time
you
know
if
you
want
to
build
an
efficient
like
cdn.
F
In
this
way,
they
should
not
be
like
that,
so
there
should
be
an
economic
principle
behind
it
that
says,
content
should
be
replicated
where
demand
comes
up,
and
when
this
happens,
perhaps
the
reward
or
the
payment
from
the
network
is,
should
be
split
between
these
two
just
as
an
example
here
and
then,
finally,
you
have
the
content
exchange
economy,
which
is
the
clients
and
the
the
storage
provider
that
they
interact
with
and
how
they
pay.
F
So
again,
a
large
body
of
literature
here,
and
we
did
you
know
quite
a
bit.
We
split
like
the
papers
that
we've
read
in
different
categories.
F
These
do
not
really,
you
know,
they're
in
a
different
area,
not
really
linked
with
icn
or
content,
addressing
necessarily,
but
the
problem
is
that
what
we
need
to
do
is,
as
I
said
before,
we
need
to
link
these
two,
and
you
know
what
at
least
we're
looking
to
do
is
to
link
these
two
and
have
some
sort
of
perhaps
metadata
or
manifest
files
to
live
together
with
with
content
addressed
data,
so
that
we
can
and
like
embed
indirectly
embed
economic
principles
into
the
naming
structure
or
the
metadata
itself,
and
you
know
like
having
authenticated
likes
or
self-certified
data
helps
a
lot
with.
F
This
helps
a
lot
with
going
away
from
necessarily
location
based
addressing
as
we've
seen
in
the
past.
There
are
a
number
of
attacks.
Only
a
few
of
them
are
listed
here
that
we
can
discuss
further.
If
you
want
and
yeah,
I
think
this
concludes
my
talk.
There
is
this
github
repository
there,
where
you
know
this
was
a
very
context,
condensed
summary
of
what
exists
in
that
website
there.
F
So
yeah,
please
go
there
and
contribute
to
the
discussion.
There
is,
let
me
see
if
I
can
go
yeah,
so
this
is
the
link.
It's
like
a
pretty
long
document,
yeah
around,
like
20-25
pages,
so
yeah,
but
all
this
is
like
subject
to
others
getting
involved
and
having
more
collaboration.
B
Okay,
great
thanks
janice,
so
we
planned
some
time
for
discussing
this.
So
don't
be
afraid
to
ask
your
questions.
So
we
have
some
questions
in
the
chat
already.
I
think
I'd
like
to
moderate
this
a
little
bit.
So
maybe
let
me
first
ask:
are
there
any
you
know,
questions
for
clarification
or
anything
that
you
didn't
understand
and
would
like
to.
B
Clarify
so
you
can
use
the
queueing
system
or
add
your
questions
to
the
chat
window,
so
maybe
one
question
to
start
with,
so
I
think
I
will
just
start
reading
the
questions
here
and
other
comments.
We
can
also
hear
them
so
maybe
just
on
the
on
the
fundamental
motivations
on
your
very
first
or
second
slide.
You
said
well,
if
you,
you
know
want
to
do
content
delivery
on
this
commissionless
p2p
network.
You
have
all
kinds
of
challenges,
so
I
mean
considering
that.
B
So
you
know
cdn
is
you
know
widely
deployed
today
and
also
the
economics
are,
I
think,
quite
well
understood
you
could
still
argue.
There
are
some
efficiency
problems,
especially
like
with
multicast
like
traffic,
but
so
what
is
the
fundamental
problem
that
you
would
like
to
address
with
the
system.
F
As
compared
to
them
like
the
economics
of
cdns
and
and
content
delivery
more
in
general
yeah,
so
I
mean
why
do
I
say:
why
would
people
need
this.
F
Right,
so,
okay,
so
the
economic
okay
starting
from
the
economic
model
is
different
because
you
know,
as
I
mentioned,
towards
the
end,
you
don't
have
a
single
entity.
That
necessarily
kind
of
is
on
top
of
that
now.
Why
would
people
need
this?
It's
a
different
kind
of
like
doing
things,
a
different
way
of
doing
things.
Instead
of
you
know,
storing
data
on,
like
you
can
think
of
it
as
a
dropbox
thing.
F
B
So
maybe,
let's
talk
a
bit
about
the
say,
security
and
and
trust
architecture
here,
so
you
you
mentioned
things
like
sales,
certified
names,
for
example.
So
what
is
the
actual
assumption
on
the
say,
trust
architecture?
I
could
say
who
is
certifying
these
names,
so
I
mean
sometimes
we
use
self-certified.
B
In
a
say,
this
kind
of
sloppy
way.
I
mean
someone
still
needs
to.
I
mean
you
want
to
map
this
to
some
real
world
identity,
perhaps
or
there
needs
to
be
some
other
trust
fruit.
So
what's
what's
the
idea
here.
F
F
The
the
I
mean
yeah,
as
I
said.
First
of
all,
I
don't
really
have
answers
to
many
of
these,
so
any.
I
believe
that,
with
all
the
different
applications
that
could
be,
you
know
could
be
applicable
on
top
of
such
a
system.
I
think
there
wouldn't
be
a
one
size
fits
all.
So
in
some
cases
you
know
there
could
be
some
identity
like
like
the
name
of
some
sort
through
a
public
key
or
or
something
like
that.
That
would
be
the
identifier
of
someone
or
like
differently
in
terms
of
data.
F
It
could
be.
You
know
like
what
we've
seen
like
hashing
through
hashing
the
content
itself
and
using
that,
like
as
part
of
the
name
itself,
means
that
someone
can
basically
later
verify
that
the
data
they
receive
these.
You
know
the
data
they
asked
for.
So
if
you
have
some
part
of
a
hash
as
part
of
the
name,
then
this
kind
of
is
an
option
to
go
forward.
F
B
Yeah,
let's
look
what
else
we
have
in
the
chat
so
dave.
Do
you
want
to
ask
me
questions
yourself?
Should
I
read
them?
Is
your
mic
working.
A
Let's
see
if
I'm
causing
echo
am
I
causing
echo
sounds
good.
A
Okay,
I
think
I
may
have
resolved
all
my
audio
problems,
so
there's
a
general
theme
to
to
the
questions
which
is
the
connection
of
security
to
the
economics.
A
There's
some
extreme
points
like
going
all
the
way
down
the
blockchain
rabbit
hole,
and
then
there
are
other
points
which
are
what
are
generally
called
permissions
systems
where
you
know
you
sign
up,
and
you
have,
you
know,
have
an
identity.
That
is
that
is
understood
in
the
context
of
the
permissioning
system
that
you
use,
but
you
seem
to
be
advocating
something.
That's
in
the
middle
there
and
I'll
it'll
be
really
interesting
to
get
your
views
on
whether
you
think
there's
some
kind
of
stable
architectural
point
in
between
those
extremes.
A
A
The
the
the
system
has
no
idea
really
what's
going
on
other
than
you
know,
somebody's
either
doing
a
proof
of
stake
or
proof
of
work
type
of
thing
in
order
to
to
do
a
transaction
with
the
system,
which
is
either
to
store
something
or
retrieve
it.
The
other
extreme
is
is
a
permission
system
where
you
sign
up,
and
you
have
an
identity
in
that
system
and
other
and
and
there's
a
there's,
some
third
party
or
central
there's
quasi-centralized
authority
that
keeps
track
of
who
is
permissioned
and
who
is
not.
A
You
know
current
cdns
work
that
way
there
are.
There
are
obviously
other
types
of
econo
entities
that
can
do
this,
but
you
seem
to
be
somewhere
in
between
those
advocating
something
in
between
those
two
things.
F
Right,
so
that's
right,
there
isn't
kind
of
one
size
fits
all
in
my
opinion
here
there
certainly
are
applications
where
you
know
could
benefit
from
a
blockchain
type
of
thing
where
some
sort
of
identifier
or
you
know,
transaction
identifier-
is
kind
of
reported
there
and
look
there.
But
again,
I
don't
think
that
this
would
be.
You
know
like
sustainable
to
do
for
everything
all
all
the
content,
a
user
wants
to
request
on
the
other
side.
F
F
Now
this,
although,
although
this
goes
to
the
other
side,
it
kind
of
builds
on
top
of
the
base
network
set
up.
F
F
You
can
start
an
isp
today,
and
you
know
you
don't
need
to
kind
of
get
permission
well
like
simplistically
speaking
like
at
the
kind
of
those
who
are
transporting
the
beats
of
the
overlay
should
be
able
to
kind
of
get
their
gear
up
and
and
start
doing
stuff,
and
you
know
then
building
applications
on
top
of
that
could
stretch
the
system
to
the
other
to
the
other
part,
but
these
two
should
be
able
to
kind
of
interoperate
both
as
regards
you
know,
how
do
you
request,
resolve,
transfer
content
and
who
is
paying
who?
F
How
are
transactions
kind
of
aggregated
somewhere
or
reported
somewhere?
A
A
F
Yeah
yeah,
I
mean
this
like
you
know,
this
is
kind
of
more
of
a
things
that
you
know.
You
realize
that
modern
applications
and
users
would
require
from
the
network
and
therefore
you
know
when
you're
trying
to
design
something
like
that.
You
should
take
these
into
account,
whether
they're
possible
or
whether
you
know
they
are.
F
You
know
they
are
performance
like
damage
there.
Then
that's
fine,
you
know
like
thor,
has
got
performance
that
much,
but
you
know
many
people
use
it
and
basically
I
I'm
not
aware
of
any
great
system
that
has
like
providing
privacy
at
the
you
know,
ideal
level
of
the
desired
level
and
at
the
same
time
have
the
same
performance.
So
yes,.
A
You
can
take
a
couple
of
the
really
detailed
things
offline
with
giannis
because
he
made
it.
He
made
a
claim
that
pits
introduced
additional
attack,
vectors
and
I'm
just
curious
what
those
might
be,
because
I
think
all
of
them
are
already
there
with
a
fib
in
netflow.
A
D
Okay,
in
terms
of
title,
I
didn't
know
that
so
yeah
I'll
have
a
look.
B
Probably
good
for
for
the
minutes,
so
it
seems
to
me
so
there
are
kind
of
two
schools
of
thinking.
So
when
you
know
approaching
this,
like
quantum
distribution
in
say,
I
see
n
ways
so
say
like
recent
work
in
the
say:
icn
environment.
So,
for
example,
like
the
icdn
paper
at
last
year's
icm
conference.
B
So
this
seems
to
start
from
say
the
like
the
fundamental
icn
assumption
so
that
you
have
secured
named
access
to
data
and
then
you
think
about
okay.
What
do
I
need
to
provide
some
kind
of
approximation
to
what
cdn
does
today
so
but
in
a
say,
nicer
way,
so
self-organized
say
less
manual,
traffic
steering
and
so
on,
so
basically
leveraging
icn
principles.
B
So
you
seem
to
assume
that
there
there
is
a
like
peer-to-peer
like
network
in
the
first
place
and
so
you're
sorting
out
the
economics,
and
so
then
you're
trying
to
you
know
plug
in
some
centric
or
content
addressing
networking
concepts
they
have
to
to
solve
the
the
data
transport
between
caches
or
between
nodes
in
the
system.
Is
that
the
correct
way
to
characterize
it,
or
what
do
you
think.
F
Yeah,
the
acdn
paper
is
great.
Actually
it's
a
great
paper
actually
and
was
there
was
lots
of
inspiration
that
at
least
I
got
when
doing
this
literature
review.
So
it's
very
much
along
the
lines
that
the
paper
talks
about.
I
mean
it's
been
a
couple
of
months.
I
read
it
but
yeah,
it's
certainly
very
relevant.
So
in
terms
of
I'm
not
sure
I
get
it
I'm
getting.
If
there
is
already
a
peer-to-peer
network
and
you're
trying,
so
you
can
assume
that
you
know
there
is
you.
F
B
B
A
Can
you
hear
me
you're
back
okay.
F
Okay,
so
it
was
thrown
out
yeah
okay,
so
I
was
talking
right.
F
D
F
Yeah,
so
I
don't
know
up
to
which
point
you've
heard,
but
I'll
try
to
summarize
again,
so
there
isn't
necessarily
the
assumption
that
a
php
network
necessarily
exist
already.
What
has
to
be
there
is
that
you
know
we
can
assume,
since
this
is
a
kind
of
theoretical
work
at
the
moment.
F
We
can
assume
that
you
know
at
some
points
once
these
things
are
sorted
out,
people
would
be,
you
know,
get
interested
and
say:
okay,
I
I
can
participate
in
this
network
and
then
you
know
there
could
be
users
and
like
servers
and
therefore
overlay
nodes
like
popping
up
to
participate
in
the
network.
So
if
your
question
was,
does
this
network
already
exist
not
necessarily
at
least
like
this
study
in
literature?
Review
was
not
done
based
on
some
existing?
I
don't
know
topology
or
other
characteristics.
B
Yeah
kind
of
I
was
more
thinking
like
more
like
in,
let's
say
the
in
the
the
theoretic
approach,
so
this
seems
to
be
based
on,
say,
primarily
p2p
con
concepts
so,
like
maybe
the
earlier
icn
systems
like
net
in,
for
example,
also
yeah
right.
So
I
I
was
trying
to
make
the
point
that
I
think
there
are
these
like
two
different
schools
of
you
know,
thinking
about
the
content,
distribution
problem,
and
so
I'm
not.
B
Sites,
just
maybe
trying
to.
F
Organize
yes,
so
in
the
in
the
in
the
second
topic
that
I
talked
about
in
the
graph
forming
one
that
that's
what
I
talked
about
when
I
meant
when
I
talked
about
the
tipping
point,
so
you
know
like,
if
you
assume
the
icn
like,
like
the
network
level
icn
approaches,
then
you
can
have
a
very
nice
kind
of
name-based
forwarding
scheme
that
kind
of
gets
all
the
benefits
of
information,
centricity
and
content
centricity.
F
Now,
if
you,
if
you
say
that
okay,
this
system
works
on
the
layer
above
as
an
overlay,
then
doing
like
name
based
forwarding
between
overlaid
nodes
is
going
to
inevitably
increase
the
traffic
that
those
nodes
will
have
to
transport
through
the
machine
right
because
they're
not
on
the
path
of
you,
know
the
sender
and
the
receiver.
So.
F
Unpopular
content,
I
think
a
kind
of
resolution
based
approach
similar
to
past
approaches
like
net
infant
and
like
similar
things,
would
be
more
performance
efficient.
But
when
popularity,
for
example,
goes
from
a
point
on,
even
though
you
know
there
is
some
extra
performance
overhead,
the
popularity
and
the
number
of
times
that
this
will
need
to
be
transported
in
the
overlay
network
would
make
it
so
that
you
know
basically
doing
some
name
based
forwarding
would
be
beneficial
there
too.
So,
but
I
would
be
very
interested
to
hear
opinions
about
that.
F
That's
only
my
opinion.
I
haven't
done
any
like
you
know,
simulation
or
pen
and
paper
study,
so
yeah
sure
it's
intuition,
yeah.
B
Yeah,
it's
it's
interesting!
So
what
I
mean
when
say
looking
at
so
what's
happening
in
cdn
today.
I
think
what
you,
what
you
see
is
so
like
yeah,
I
mean
more
hyperscalers
moving
out
to
the
far
edge,
and
so
one
of
the
interesting
challenges
is
seems
to
be
marrying
like
cdn
overlays,
with
like
efficient
kind
of
multicast,
like
distribution,
so
for
life.
E
B
Like
video
and
and
so
on-
and
so
you
could
say-
I
mean
so
maybe
the
overlay
approach
is
really
inefficient
here.
You
could
do
it
in
a
different
way.
Do
you
also
think
that
so
your
approach
would
be
able
to
address
this
type
of
cdn
usage,
so,
like
multicast
like
near
live
content.
D
F
You
know
if
yeah,
if
there
were
solutions
to
the
like
problem
areas
that
I
talked
about,
then
I
think
that
would
be
a
very
good
use
case.
Actually,
that's
very
correct.
B
Okay,
let
me
check
the
chat.
Yeah
see
no
other
questions.
Yeah
then
thanks.
A
lot.
Janice
was
interesting
discussion.
B
Okay,
okay,
cool
all
right,
let's
move
on
so
next
we
have
malicious
talking
about
new
ideas
for
icn
and
iot.
Do
you
want
to
run.
I
I'll
try
I'll
try
to
share
my
slides.
I
will
not
share
my
video
because
I
have
a
low
bandwidth
here,
so
I
will
just
share
my
slides
because
if
not
everything
will
freeze
so
let's
draw
a
screen.
I
B
A
I
E
I
B
I
So
maybe
I
will
introduce
why
I
wanted
to
do
this.
You
gave
me
way
more.
You
gave
me
way
more
time
than
I
wanted.
I'm
involved
more
and
more
into
what
I
would
call
not
your
grandfather's
iot
systems,
and
I
think
there
was
I
think,
shank
was
on
the
call
and
shank
had
written
something
for
icnrg
some
time
ago
about
icn
for
iot,
and
I
think
this
is
like
an
old
idea
becoming
new
just
because
of
what
iot
systems
are
becoming.
I
So
you
can
go
to
the
next
slide,
because
this
is
where
I
define
what
what's
happening:
okay
and
the
world
of
iot,
as
at
first
a
lot
of
the
iot
systems.
It
was
very
easy.
You
were
putting
a
lot
of
of
sensors
and
you
were
just
you
know
from
your
web
app
or
your
web
center.
I
You
would
send
a
command
and
the
command
would
take
a
temperature
or
would
take
a
single
measure
and-
and
it
was
very
simple-
and
it
was
really
care-
you
know
it
and
it
was
doing
it.
However,
the
whole
move
to,
I
would
say,
autonomous
systems,
but
not
only
those
have
shown
that
cloud.
I
The
iot
is
not
for
the
cloud
anymore,
but
it's
a
cloud
edge,
so
this
is
actually
something
we've
discussed
in
in
coin,
but
it
is
also
something
that's
very
much
related
to
information-centric
approaches
in
the
sense
that
it's
all
now
very
much
data-driven
with
a
lot
of
ai,
so
you're
not
asking
just
for
a
measure
you're
asking
for
a
bunch
of
measures
that
are
following
a
certain
pattern
and
will
be
used
to
improve
on
that
pattern
or
increase
the
the
value
of
that
data
set.
I
There's
a
lot
of
discovery.
There's
a
lot
of
distribution
of
function,
there's
a
lot
of
orchestration.
These
are
this
intermediation
and
functional
decomposition.
Actually,
the
previous
talk
mentioned
some
of
that,
because
it's
related
and
it
owes
the
security
and
it's
embrace.
I
call
it
the
power
of
the
and
and
where
it
is
not
anymore.
Also,
you
don't
want
to
know
that
temperature.
I
I
Like
I
need
to
take
these
things
once
a
month,
because
this
is
what
my
industrial
engineer
recommends-
obviously
there's
a
ton
of
new
applications,
this
obviously
industry
and
agriculture,
5-0
the
whole
health
care
and
education,
a
lot
of
raising
application
and
robotic
and
optics
and
and
all
the
big
data
applications
are
running
everywhere
and
again.
That
means
that
when
you
send
a
command
to
your
iot
system,
you
may
not
get
an
answer.
I
You
may
get
a
bunch
of
stuff,
it
may
be
a
a
picture.
It
may
be
a
link
to
another
element
that
could
give
you
a
better
answer
than
the
one
that
you
asked.
So
this
thing
is
becoming
a
lot
of
like
a
small
information
center
network
in
the
sense
that
and
there's
a
lot
you
know
and
it
it
leads
into
data
discovery,
data
filtering
so
much
more
again,
a
a
information
system
than
just
what
the
original
iot
was,
and
obviously
with
this
emergence
of
whatever
6g
is.
I
But
a
lot
of
people
seem
to
be
putting
in
6g
everything
that
the
telecom
industry
forgot
about
5g,
and
it
includes
all
these
computing,
a
cache
that
are
distributed
in
the
network
and
and
you
do
that
to
save
bandwidth.
I
You
know
you're
not
going
to
send
everything
at
the
same
time,
because
you
would
kill
the
network,
so
you
store
a
lot
of
things
locally,
but
then
you
need
to
find
them
and
and
all
the
new
satellite
networks
which
are
becoming
a
little
bit
more
like
a
way
to
transfer
that
information
from
one
point
to
another.
But
again,
if
you
think
about
this,
these
things
are
iots
in
themselves.
I
So
the
target
of
all
this
research,
I
think,
because
there's
a
lot
of
it,
that
is,
research,
it's
not
something
that
has
been
resolved
because
the
current
standard,
which
you
know
everybody
that
does
iot,
knows
about
mqtt,
which
is
a
very
simple
pub
sub
and
even
what
has
been
developed
in
in
and
more
recently
in
co-op,
which
is
slightly
more
sophisticated.
I
But
it's
still
not
what
you
know
we
would
like
to
have
so
it's
a
better
edge
that
we
need.
We
need
this
better
edge
to
be
intelligent
and
to
be
able
to
deal
with
the
information
you
want
to
have
distributed
computation.
But
for
this
group
it's
not
a
computation,
that's
interesting,
but
it's
this
decentralized
orchestration
and
the
security
models,
and
that
is
actually
something
well.
It
relates
to
the
presentation
before,
but
the
security
models
are
very
evolved
in
icn
and
they're,
not
very
involved
in
current
iot
systems.
I
So
it's
it's
also.
This
integrated
edge
network
in
cloud,
it's
more
like
a
computer
board
where
you
have
all
these
little
iot
systems
that
communicate
with
one
another.
It's
not
a
telephony
network
and
it's
not
a
a
cloud,
a
pure
cloud.
So
it's
this
cloud
edge
and
this
edge
that
supports
the
network
and
and
the
cloud
and
the
network
supporting
the
edge
again
for
the
in
the
that
case,
for
the
the
better,
I
would
say,
handling
of
all
the
information
that
is
being
generated
in
the
iot
system.
I
So
what
do
I
mean?
Maybe
you
can
put
the
next
slide,
of
which
I
could
speak
about
for
about
three
hours,
but
this
is
actually
an
iot
system
and
it
fits
to
a
bunch.
It's
very
small
and
I
know
dave,
you
will
complain
about
it
and
I'm
sorry.
I
yes
and
the
question
is
computational
orchestrating
data
fusion.
Yes,
actually
what
this
thing
is
for
people
who
don't
want
to
go
into
my
very
small
characters,
the
top
one
actually
has
has
this
icm
thingy.
I
I
Well,
you
can
think
about
the
autonomous
cars
because
they're
the
ones
who
are
like
the
the
worst
of
that,
because
they
generate
an
incredible
amount
of
data,
but
even
much
smaller
ones.
If
you
start
taking
pictures,
if
you
start
taking
more
advanced
measures
again
that
correlate
more
than
one
sensor.
I
You're,
not
interested
in
everything
you're
interested.
When
something
happens,
filtering
and
computation-
I
think
filtering-
does
decides
what
data
you
want
to
keep
and
the
computation
actually
works
on
it
and
may
do
some
data
discrimination
and
reduce
the
data
by
processing
and
not
just
by
deciding
which
data
you
want
to
keep.
So
the
features
of
these
architectures
are
obviously
intelligent
controllers.
An
intelligent
controller
is
something
that
is
again
data
driven.
I
You
want
to
have
information
based
queries
again,
because
you
would
like
to
not
keep
everything
or
just
have
the
data
when
you
need
it
and
not
all
the
time
you
want
to
they
do
your
data
filtering
on
metadata.
That
goes
a
little
bit
to
the
question
about
filtering
a
lot
of
times.
Data
filtering
is
made
on.
You
know
the
it's
it's
based
on
the
header
or
it's
based
on
the
destination
or
the
source,
but
you
would
like
it
to
be
on
metadatas.
I
That
means,
when
that
thing
was,
was
measured,
how
it
was
measured,
which
data
set
it
it
it.
It
relates
to
does
obviously
orchestration
computation
against
for
again
for
the
the
data.
How
do
you
call
it?
You
know
you
want
to
go
from
an
image
that
is
megabytes
into
a
decision.
I
So
is
this
image
the
one
I
want
to
keep
based
on
when
I
get
it.
I
I'll
come
back
to
your
question,
dave
and,
and
then
there's
obviously
local
and
remote
processing,
and
I
think
this
is
related.
Okay,
I
I
get
the
question
you
have
dave,
throwing
data
away
is
a
computation,
but
for
me
maybe
I'm
I'm
too
much
connected
to
processing
an
image
in
this.
I
can
decide
that
I
want
to
keep
this
image
because
it's
it's
it
was.
I
So
why
icn?
Because
what
needs
improvement?
It's
this
inter
operability,
that's
needed
between
the
edge
devices
themselves
and
you
should
always.
You
should
not
always
rely
on
distant
cloud
to
make
these
decisions
you
want
to
do
them
based
on
the
information
that
is
there
and
on
the
processing.
I
That
is
there
and,
under
the
result
of
your
data
discovery,
I
see
dave
frowning,
I
don't
like
it,
but
that's
fine
and
I
think
the
the
main
reason
is
that
we
need
better
pub
sub
and
data
discovery
and
because
the
devices
are
much
more
capable
than
they
were
before,
and
we
can
argue
which
functions
we
absolutely
need.
But
we
cannot
disagree
on
the
fact
that
you
want
to
have
queries
and
results
of
these
queries
that
are
going
to
now.
I
We
actually,
at
least
in
one
of
the
projects
that
I'm
working
on
we're,
probably
going
to
do
our
own,
but
maybe
there's
a
there's,
a
there's,
a
role
for
icm,
which
has
a
lot
of
the
functionalities
we
need
and
that's
it.
I
I
think,
there's
need
for
research
there.
I
may
put
my
other
hat
at
one
point
and
and
try
to
start
that
research,
especially
again
as
it
relates
to
a
lot
of
stuff
that
people
are
talking
about
in
haptics
and
robotics,
which
are
also
connected,
but
I
think
there's
there's.
I
There
was
work
in
iot
in
icn
and
I
think,
maybe
again
maybe
it's
time
to
to
rethink
about
it
and
restart
some
of
the
research,
because
what
I
can
see
is
that
the
the
iot
systems
are
not
getting
simpler,
they're
getting
more
powerful
and
even
more
data
driven.
So
that's
it!
I
I
didn't.
I
you
gave
me
way
too
much
time.
I
just
wanted
to
throw
an
idea.
B
I
A
The
first,
the
the
fundamental
objects
are
computations
and
the
data
is
just
something
that
flows
through
the
computation
yeah
yeah.
If
you
take
a
data
centric
view
of
the
world,
the
data
is
the
the
first
order
thing
and
computing
is
some
second
order
thing
that
happens
on
the
data,
and
neither
neither
of
those
extremes
is
quite
right,
so
the
model
I
always
have
in
my
head
is
what
happens
at
the
lhc
right,
where
it
isn't
a
simple:
oh,
let's
just
throw
most
of
the
data
away
because
it's
useless
it's.
A
We
have
to
do
a
bunch
of
computation
to
figure
out
what
data
to
throw
away
yeah,
and
this
is
a
staged
thing.
So
so
what
happens
very
close
to
the
sensors
in
the
in
the
detectors
is,
is
very
different
from
what
happens.
One
stage
back
two
stages
back
three
stages
back
and
all
of
that
it
has
to
do
with
with
what
computation
can
be
done
where,
as
opposed
to
what
data
is
where,
because
the
data
is
going
to
be
where
the
data
is.
I
A
A
Many
cases
like
industrial
process
control-
and
you
know,
physics
and
stuff
like
that
right,
so
you
don't
need
to
discover
the
data.
You
know
exactly
where
it's,
where
it's
going
to
be
because
you
designed
the
system.
I
I
Conversation
yeah
because
this
is
actually,
if
you
I
I
actually
looked
at
this,
maybe
from
a
different
point
of
view.
I
looked
at
it
more
as
a
criticism
of
existing
pub
sub
in
iot,
and
the
existing
pub
sub
in
iot
is
about
asking
a
question
and
getting
an
answer
and
it's
a
simple
question
that
comes
back
with
a
single,
a
simple
answer
but
and
the
goal
there
is
that
you're
going
to
you
know
it's
it's
your
nest.
I
It's
your
nest,
you
know
thermostat
or
it's
you
ask
alexa
to
play
a
song,
that's
fine!
It
works.
If
I'm
an
industrial
system,
I
will
measure
the
temperature.
I
will
measure.
A
A
A
I
Okay-
okay,
it's
not
it's!
Not
it's
not
the
well.
Maybe
the
boson
is
not
the
the
the
the
generic
application,
but
I
will
tell
you
that
what
you
described
first,
I
think
finding
bosons
is
not
an
iot
system,
but
that's
fine,
but
I
I
follow
the
thought.
What
I
will
tell
you
is
that
right
now
in
what
is
available,
if
you
do
an
industrial
iot
system,
first,
there's
no
processing
close
to
where
the
information
is.
I
You
also
have
a
lot
of
need
because
of
connectivity
issues
in
that
case
to
accumulate
the
data
in
the
system.
So
you
know
where
it
is,
but
you
don't
know
you
may
not
want
to
download
the
whole
database
just
to
have
access
to
one
piece
of
information
and
again
in
systems
that
are
in
outside.
I
would
say
where
there's
a
lot
of
bandwidth,
you
may
want
to
have
a
way
to
do.
I
Well,
it's
I
don't
even
call
it
data
reduction
because
at
one
point
you
may
want
to
have
access
to
the
whole
thing,
but
you
may
want
also
to
interrogate
the
system
based
on
a
much
richer
set
of
information,
queries
and
creation,
and
to
create
a
very
much
more
rich
data
set
than
just
what
you
would
get
by
just
asking
a
simple
query:
you
may
not
want
to
know
if
it's
a
boson
or
not.
You
may
want
to
know
how
the
boson
was
created,
I'm
not
in
physics
anymore.
I
don't.
I
My
original
degree
was
in
physics,
but
I
anyway,
if
you
think
I
will
just
I
my
question
to
the
group-
was
if
there
was
a
need
to
revisit
icn
and
iot.
You
seem
to
indicate
that
no,
I
had
a
thing
to
indicate.
Yes,
so
maybe
somebody
else
has.
A
I
I
What
what
I'm
I'm
staying
here
and
again,
there's
echo?
What
I'm
saying
here
is
that.
I
I
I
If
data
y,
which
is
also
from
another
forum
of
data
gathering,
has
it
as
a
I
should
you
should
be
able
to
publish
your
computation,
and
so
what
I'm
saying
is
that
you
need
a
much
richer
set
of
queries
into
both
your
data
and
your
processing
at
the
edge
than
what
is
currently
being
offered
and
the
sub
could
be
a
request
to
suit
the
computation.
Yes,
and
in
the
case
of
the
system,
I
have
in
my
mind
it.
It
actually
is.
I
Discovery
problem.
Well,
of
course
it's
it's
it's.
It
is
beyond
iot,
and
if
you
look
look
at
at
the
in
my
other
group
and
coin,
all
the
data
all
the
discovery
drafts
they,
you
know
they're
looking
at
a
whole,
much
a
whole
bunch
of
other
ways
of
doing
it,
and
there's
also
work
being
done
at
at
ericsson
in
sweden,
on
how
discovery
is
also
based
on
on
semantic
discovery,
which
is
another
interesting
topic
and
did
from
the
question
about
an
actuator.
I
The
actuators
are
now
pretty
dumb,
so
I
would
say
that
they
also
can
do
the
the
sophisticated
stuff
that
we
want.
What
we
want
to
do
actually
we're
not
and
actuators,
are
very
simple
processors.
Maybe
you
want
to
have
something
much
more.
I
I
I
have
in
my
mind
that
you
know
if
I
want
to
have
an
image
processing
response.
B
So
so
I
think
there
are
really
different
perspectives
on
this,
and
so
I
think
I
mean
your
presentation
today
is
essentially
yeah.
It's
it's
a
kind
of
start
of
a
longer
discussion,
yeah.
E
B
And
so,
as
dave
pointed
out,
so
you
you
can,
you
know,
conceive
these
iot
systems
in
a
way
that
you
say.
Oh,
you
know
I'm
trying
to
fulfill
some
purpose,
I'm
looking
for
data
and
the
environment,
and
I
need
to
find
it
and
ways
to
access
it.
That's
one
way
of
seeing
it.
Another
way
is
so,
for
example,
if
you
think
about
like
a
data
streaming
framework,
so
maybe
like
apache
flink
or
something
where
you
have.
B
You
know
functions
in
a
in
a
graph,
essentially
so
directly
graph,
and
so
these
functions
know
what
they
are,
what
they
do.
Somebody
is
kind
of
connecting
those,
and
so
they
are
kind
of
also
told
which
input
data,
for
example,
to
subscribe,
to
or
or
to
access.
So
that
system,
you
wouldn't
discover
things,
but
still
you
would
like
to
benefit
from
the
ability
to
put
functions
at
convenient
places,
and
then
you
know
let
the
system
figure
out
the
connectivity
itself
and
yes,.
I
And
actually,
this
is
something
that
also
you
know:
iot
systems
go
into
industrial
or
agriculture,
or
things
like
that
and
a
lot
of
times.
You
would
like
not
to
replicate
all
functionality
everywhere
and
what
you're
saying
is
is
interesting
because
you
know
maybe
one
one
node
has
it.
I
Sadly,
I'm
working
on
it
on
another
proposal
for
a
big
iot
system,
and
once
I'm
done
with
it,
I
will
have
a
very,
very
nice
use
case,
but
I
had
I
had
hardly
time
to
breathe
in
the
past
two
weeks.
So
I
will
it's
it's
my
plan
of
of
showing
how
a
a
current
system
is
limited
compared
to.
I
B
Yeah,
let
okay
so
okay,
phil
remove
himself
from
the
queue
all
right
yeah,
so
I
would
say:
let's
follow
up
on.
This
would
probably
be
good
to
talk
more
about.
You
know
specific
scenarios
so
also
alluding
to
the
recent
comments
in
the
chat
and
well,
I
still
think
there's
there's
some
really
interesting
potential,
also
really
new
perspectives
that
are
emerging.
So,
let's,
let's
do
that
thanks
again
for
this
discussion,
starter.
I
Thank
you
and
thank
you
for
everyone
for
allowing
me
to
do
this
and
again
sorry
that
I
didn't
have
much
time
to
to
make
it
better,
but
I
will
get
to
it.
Thank
you.
B
Okay,
thanks
do
we
have
a
jerk
on
the
call
or.
B
In
that
case,
so
dave
as
the
the
co-chair
of
the
icn
conference,
would
you
like
to
talk
to.
A
Yeah
hi,
so
I
just
want
to
publicize
the
fact
that
our
annual
conference
is
coming
up,
we're
now
on
the
eighth,
like
eighth
iteration
of
this
and
for
the
research
community.
This
is
kind
of
the
premier
venue
for
getting
your
work
on
icn
related
topics
through
the
review
process
and
published
and
presented.
A
So
the
conference
is
the
end
of
september
22-24,
as
it
is
every
year
towards
the
end
of
september,
and
please
note
carefully
that
the
paper
registration
deadline
is
at
the
beginning
of
may.
So
you
have
two
months
to
put
together
a
super
paper.
If
you
haven't
already-
and
I
could-
it
could
be
the
next
slide,
because
this
year
we've
expanded
the
call
for
papers
with
some
suggested
areas
that
we
haven't
necessarily
emphasized
so
much
in
the
past.
A
Recovery,
that's
been
a
topic,
that's
been
increasing
over
the
last
few
years.
That's
that
that
we're
continuing
to
emphasize-
and
I
think,
there's
a
lot
of
opportunity
to
do
new
work.
There
measurement
and
management
and
again
another
topic
that
we
we
want
to
really
encourage.
People
to
think
about.
Now
that
we
have
a
solid
technical
base
is
what
are
the
more
economic,
soft
political,
sciencey
and
economic
aspects
of
what
icn
can
mean
to
not
just
the
networking
community
but
to
the
comp
to
the
computer
science
community
as
a
whole.
A
So
please
consider
submitting
your
work
use
hot
crap
for
the
conference
management.
If
you
take
a
look
at
the
program
committee,
it's
it's
not
just
the
usual
suspects.
We
have
a
pretty
broad
participation,
not
just
in
the
icn
community
but
broadly
in
computing.
A
So
so
please,
let's,
let's
make
a
really
good
conference
this
year.
It
will
be
virtual,
but
we're
hosting
it
out
of
the
equal
reporting
technique
and
the
cisco
systems,
paris,
innovation,
research
lab
in
paris.
So
thanks
very
much.
Let's
move
on.
B
Great
thanks:
we
are
looking
forward
to
a
really
nice
program
again
this
year
and
yeah.
This
brings
us
to
the
end
by
the
way
thanks
everybody
for
attending
yeah,
we'll
follow
up.
Regarding
the
the
last
caller,
please
help
us
getting
this
ready
yeah.
Please
keep
on
doing
all
this
great
work
and
presenting
new
ideas
to
the
group
and
writing
drafts.