►
From YouTube: IETF105-T2TRG-20190724-1330
Description
T2TRG meeting session at IETF105
2019/07/24 1330
https://datatracker.ietf.org/meeting/105/proceedings/
A
I
PR
guidelines
of
the
IETF
to
apply.
We
have
the
blue
color
pink
sheets
going
on.
Please
sign
them.
We
have
to
note
taker
sponsor,
thank
you
around
and
honor,
but
everyone
else
feel
free
to
chime
in
on
the
note-taking
we're
using
the
ether
pad.
You
have
the
link
of
the
indicated
webpage
and
in
the
official
agenda
page.
Is
there
anyone
using
jabber.
A
Anyone
who
can
use
jabber,
perfect
okay,
if
you
guys,
can
join
the
chapel
room
just
in
case
if
there's
something
to
relay
haven't
been
too
much
lately,
but
you
never
know.
As
usual,
we
have
our
github
repo,
where
you
have
the
latest
agenda
and
and
slide
material
available,
and
you
can
see
the
link
at
the
bottom
of
the
slides
our
agenda
for
today.
A
I
some
changes
done
this
morning,
we're
gonna
now
now,
starting
with
the
introduction
on
the
current
status
of
the
research
group,
upcoming
meaning
activities
and
going
on
then
holding
some
bit
more
detailed
reports
on
on.
We
see
creative
meetings,
oma,
etc.
After
that,
Michael
coaster
will
give
us
an
update
on
the
activity.
A
Little
convergence
and
also
w3c
community
group
on
schema
extensions
after
Michael
McCool
will
give
us
an
update
on
the
web
of
things,
and
then
we
have
two
short
presentations,
one
from
evallo
on
this
activity
that
we
were
already
doing
at
Aetna
Vichy
and
looking
into
how
we
can
use
young
or
object
Universal,
parsing
interface
and
Christian
I'm
just
will
give
us
an
update
on
that
new
transports
for
co-op
design
team.
After
that,
the
second
half
of
the
meeting
is
gonna
be
focused
on
the
edge,
so
we
will
have
Dirk
presenting
outrages
of
Indian
presentation.
A
A
Okay,
then,
let's
get
started
just
a
quick
reminder,
and
since
we
may
have
some
newcomers
here,
also
in
the
room,
their
scope
and
goals
of
fielding
research
group
is
that
we
are
looking
into
research
issues
that
we
can
turn
the
Internet
of
Things
into
reality,
really
focusing
on
those
low
resource
nodes,
how
they
can
communicate
among
themselves
and
also
with
the
wider
Internet.
We
are
looking
at
a
big
part
of
the
stack
but
really
focusing
on
the
opportunities
where
there's
potential
for
IETF
standardization.
A
One
thing
we're
often
asking
is:
what's
the
relationship
with
this
IOT
work
in
Tilden,
research
group
and
IETF
and
elsewhere?
There's
obviously
a
lot
of
overlap,
but
where
we
are
looking
at
these
research
issues,
whereas
the
IETF
side
is
looking,
for
example,
on
protocol
engineering
in
part
garden,
core
working
group,
a
lot
of
those
things
were
discussed
are
common
between
us,
but
the
standardization
aspects
are
done
in
the
core
working
group.
A
Another
idea
to
cutter's
is
for
security,
but
a
lot
of
the
research
topics
are
explored
here
and
then
there's
also,
for
example,
dalvik
working
group
in
an
IETF
side,
looking
into
guidance
aspects
on
the
same
area.
So
that's
how
these
are
roughly
working
together
and
there's
no
clear-cut
line.
But
it's
a
lot
of
collaboration
between
these
various
groups.
A
A
A
We
I
don't
have
anything
scheduled
right
now,
but
you
will
likely
see
something
something
show
up
and
then,
of
course,
in
the
next
thing,
apart
in
an
excited
meeting
in
Singapore
in
November
with
Muslims,
we
have
a
wish
hackathon
and
we're
also,
as
usual,
in
very
interesting
co-locating
academic
conferences
this
year
or
later.
So,
if
you
have
any
good
opportunities
in
mind,
please
to
let
us
know
but
bit
more
details
on
potential
for
the
Singapore
meeting.
A
So,
as
you
know,
we
are
using
these
meetings
also,
since
we
are
travelling
around
and
have
a
chance
to
interact
with
special
specific
communities,
we
want
to
use
those
opportunities
to
connect
with
those
communities
around.
So
in
the
meeting
in
Singapore,
there's
one
very
obvious
potential
point
of
contact
the
Singapore
smart
nation
project.
It's
a
big
smart
city
project
going
on
there
right
now
and
especially
since
they
do
have
some
requirements
on
standardisation,
its
kind
of
fits
nicely.
A
What
we
are
looking
at
so
one
potential
we
are
now
looking
into-
is
to
actually
have,
for
example,
before
the
IETF
meeting,
something
like
a
Friday
work
meeting
around
these
topics
and
just
to
give
you
a
bit
of
flavor
on
what
it
could
be,
and
there
was
a
good
presentation
in
the
web
of
things.
Work
so
come
on
a
month
ago
from
the
Singapore
smart
nation,
and
this
is
one
of
the
slides
that
they
use
that
I
loaned
with
permission,
what
they
are,
what
they're
really
aiming
for.
A
So
as
you
can
see,
the
smartness
and
platform
solutions
that
they
are
aiming
for
must
be
able
to
demonstrate
the
ability
to
standardize,
collect
and
aggregate.
Are
you
today
that
scale
sounds
of
course
very,
very
familiar
to
us
and
also
guaranteeing
or
cotton
carrot
and
water
resources,
so
have
security
there
secure
platform,
end-to-end
immutable
records
and
also
relevant
use
cases.
So
we
see
there's
not
a
potential
here
and
we
are
now
in
the
discussions
to
see
what
kind
of
things
we
could
be
doing.
A
A
Yes,
there's
a
lot
of
lot
of
history
in
various
forms,
but
you
know
security
security
is
not
easy.
Sometimes
take
takes
a
while
get
those
things
right,
but
of
course
the
work
doesn't
stop
there.
We
do
have
also
another
research
group
document
that
has
been
going
on
for
quite
some
time:
stressful
design
priority
it's
with
more
information
on
the
next
slide.
A
We
also
are
planning
for
a
whole
set
of
other
documents
that
are
currently
in
deed
or
submissions,
but
who
do
have
potential
of
becoming
a
research
group
documents
in
parts
go
on
H
and
IOT
will
be
presented
later
today,
also
secure
bootstrapping
for
IOT
story
that
has
been
going
going
up
and
down.
We
have
been
working
on
core
applications
and
in
part,
clock
or
interfaces
and
thinking
magus.
Maybe
some
of
that
information
should
be
turned
into
research
group
documents
and
also
last
IETF
and
decided.
A
We
had
quite
a
bit
of
discussions
on
layer,
3
considerations
on
connectivity
on
restricting
connection
enabling
connectivity.
We
may
want
to
work
on
some
research
group
documents
around
that
area
and
also,
in
that
context
of
wishi,
we've
been
discussing
distinct
concept
of
wishing
notes
so
short
notes
on
on
the
topics
that
we
have
been
discussing.
What
we
have
discovered,
what
we
have
learned
and
maybe
publishes
some
of
those
either
as
as
individual
documents
or
maybe
as
a
collection
of
documents.
A
So
restful
design
for
IOT
the
remaining
research
groups
we
have
right
now
there
was
a
small
update
that
head
well
as
a
lot
of
small
updates
on
on
this
document.
In
particular,
we
added
more
information
on
these.
Specifics
was
that
some
of
the
servers
are
constrained,
which
is
quite
different
from
the
big
web,
where
a
lot
of
use
of
the
servers
are
not
constrained,
and
also
on
the
dual
roles
of
servers
and
clients
that
they
are
usually
serving
both
roles.
A
Also,
one
of
the
point
of
this
document
is
to
provide
reference
is
to
for
more
information,
so
we
had
added
whole
bunch
of
better
and
more
references
across
the
whole
document
and
then
one
bigger
change
was
these
clarifications
on
the
push
mechanisms
and-
and
there
we
are
in
in
part,
while
working
on
to
alignment
with
a
core
dine
link
document
which
was
actually
updated
just
for
this
IETF.
So
we
also
need
to
work
on
some
updates
on
the
restful
design.
A
A
The
secure
boot
strapping
draft,
so
I
got
a
few
bullets
from
the
authors
of
that
document
and
they
what
they
wanted
to
highlight
that
in
the
IOT
state-of-the-art
security
document,
they
identified
secure
bootstrapping
as
one
of
the
key
challenges,
so
they
are
planning
to
work
on
that
topic
in
in
the
context
of
this
document,
in
particular
documenting
some
of
the
terminology
and
how
those
terms
relate
to
each
other,
onboarding,
commissioning
configuring,
etc.
It's
a
pretty
messy
field
when
it
comes
terminology
and
well
same
difference
in
terms
used
for
different
things
and
vice
versa.
A
But
this
is
thinking
Archie.
So
it's
a
secure,
bootstrapping,
adding
his
name
of
the
name
of
the
document.
Also,
the
goal
is
to
identify
common
design
assumptions
and
argit's
are
components
that
are
are
here
and
finally
investigating
the
benefits
and
challenges
of
EAP
for
IOT.
So
if
you're
interested
in
that
that
were
pleased
to
check
out
the
secure
boot
strapping
document
and
and
conduct
the
authors.
A
Only
she
what
we've
been
doing
since
the
last
two
is
that
was
meant.
We
had
two
online
meetings
where
we
were
focusing
on
these
online
meetings
is
planning
for
the
hackathon.
We
had
this
idea,
but
then
also
a
research
agenda.
So
what
are
the
topics
we
should
be
focusing
on
moving
forward
in
VC,
but
also
things
in
Archie
at
large,
so
we
have
identified
so
far
six
topics
or
topic
areas
that
seem
to
be
very
interesting
for
us
in
particle.
A
The
modeling
data
interaction
was
so
interesting
that
we
spent
most
of
the
calls
we
had
between
the
IETF,
focusing
on
that.
So
that's
clearly
something
we're
gonna
be
going
forward
with
and
Michael
will
address
some
of
those
aspects
in
his
presentation
and
overall,
this
whole
area
of
rest
based
high
fiber
media
and
how
it
works
together
with
interaction
and
data
modeling.
A
But
we
do
have
quite
a
bit
of
interest
also
on
the
connect
connectivity
area,
those
layer,
three
considerations
and
forward,
but
then
also
in
network
and
edge
computing,
and
this,
of
course,
is
something
together
with
the
co
Energy
we
have
been.
We
have
been
exploring
and
never
never
forgetting
security
when
it
comes
to
IOT
and
and
finally
terminology.
That's
something
what
we
have
repeatedly
identified
is
very
valuable
to
get
some
more
clarity
and
more
contribution,
so
we're
likely
to
produce
some
terminology
documents
going
forward.
A
Then
quick
report
on
the
vichy
hackathon,
so
this
was
the
sixth
time
we
had
wishing
hackathon
at
this
time
we
had
roughly
nine
participants
depends
how
many?
How
do
you
compete
all
the
people
in
the
table?
Couple
of
participants
also
remotely.
We
had
two
key
focus
areas
in
the
hackathon
one
was
this
IOT
data
model
convergence
and
the
other
one
was
hypermedia
for
IOT
and
coffee
making.
A
So
briefly,
more
on
the
data
model
convergence.
What
we're
working
there
is
how
we
can
use
the
one
data
model,
simple
definition,
format
to
enable
data
and
moral
model
interchange.
So
one
data
model
is
new
activity.
Looking
on
consolidation
of
data
models,
but
Michael
will
be
covering
more
details
on
that
in
a
moment,
but
what
we
did
in
the
hackathon.
We
were
working
on,
for
example,
this
tool
that
we
can
automatically
convert
if
so
like
with
them.
A
The
models
from
Ole
Miss
back
works
the
models
expressed
using
the
STF
language,
and
so
that
tool
is
currently
available
online.
You,
its
web
based
service,
but
also
the
code,
is
open
source.
You
can
find
that
in
in
github,
also
based
on
this
work,
we
did
find
some
good
improvement
suggestions
for
the
STF
language,
in
particular
on
the
data
type
schema
and
constraints
that
we
are
now
implementing
on
on
the
STF
language
itself.
Also,
one
contribution
of
the
hackathon
was
this
tool
for
generating
schemas
in
CDL.
A
As
you
know,
CD
deal
with
what
we
are
widely
using
instead
of
JSON
schema
native
context,
so
we
can
turn
the
chasing
schemas
of
of
SDF,
also
into
CD
deal
and
use
the
CD
deal
tools
for
that
and
as
a
side
product.
We
did
end
up
also
designing
at
least
a
first
proposal
for
JSON
format
for
coral,
so
that
enables
you
to
use
your
chasing
cooling
in
in
context
of
coral
applications.
A
One
activity
related
test
was
this
binary
data
extraction,
so
avala
was
working
on
on
on
tooling.
That
could
extract
a
binary
data
using
young
and
there
was
a
problem
statement
written
on
that
and
also
evolved
and
deployed
a
playground
deployment,
as
we
will
hear
bit
more
bit
more
about
that
and
the
whole
young
object
Universal
parsing
interface
in
develops
presentation.
It
was
also
a
good
activity
at
the
IETF
hackathon
and
the
other
main
area
was
disproving
coffee
with
hyper
media.
A
So
you,
as
you
all
know,
I
mean
making
making
coffee
has
been
the
grand
goal
of
IOT
since
the
1990s
and
the
time
of
the
hypertext
coffee
pod
control
protocol.
So
what
we
are
doing
here
in
the
hackathon,
we
were
taking
a
look
at
the
same
topic
area
with
more
modern
tools
and
techniques
and
part
about
what
can
coral
and
coop
do
here.
So
we
have
a
reference
scenario.
That
is
what
we'll
be
working
on.
It's
called
carrier-grade
coffee
machine,
also
known
as
Karstens
coffee
machine.
A
What
we
were
doing
with
with
that
the
reference
in
our
is
that
you
or
describing
the
coffee
machine
using
hypermedia
and
make
it
discoverable
and
you
we
had
hypermedia
clients
that
were
automatically
will
to
discover
different
menu
options,
but
you
wanna
have
a
lot
there
or
a
cappuccino,
maybe
with
a
that's
a
whiskey
on
top.
I
know
it
made
me
be
able
to
make
the
coffee
selections
and
then
execute
on
on
the
brewing
part.
A
So
we
have
now
two
open-source
implementations
using
Quraan
Khoa
being
able
to
do
all
of
that.
The
brewing
part
is
maybe
something
a
bit
work-in-progress.
You
actually
have
to
plug
in
a
real
coffee
machine
there,
but
the
whole
application
logic
is
already
there
they're
running
both
on
riot
OS
on
an
actual
IOT
port
and
then
also
a
python
implementation
that
the
christian
was
working
on
and
they're
the
presentations
you
have
links
for
both
implementations.
A
Then
finally,
we
had
this
last
Friday.
The
meeting
with
OMA
spec
works.
So
this
is
a
part
of
our
long-term
collaboration
and
coordination
work
with
OMA
that
we
are
exchanging
information
and
making
sure
that
our
protocols
and
data
models
are
used
in
a
good
way.
You
know
in
Maine,
but
also
that
the
requirement
that
OMA
has
are
addressed
by
the
IETF.
So
we
had
a
very
good
luck
within
the
tutorial
by
Hannes.
By
the
way
all
these
materials
are
available
in
in
the
meeting
github.
A
If
you
want
to
go
check
them
out,
we
had
also
a
good
presentation
on
the
whole
object
registry
model,
if
so
like
within
the
object
and
like
with
them
doing
requirements
going
forward,
but
too
OMA
will
be
working
on
for
the
next
version
of
the
protocol
and
data
models.
We
did
also
discuss
document
dependencies
between
OMA
and
IETF
in
particle
research,
directly
dining
corpora,
SMS
and
cinema
registry.
Some
of
those
will
be
discussed
tomorrow
in
the
core
meeting.
A
We
also
had
a
dis,
unconference
style
discussion,
so
they
were
a
bunch
of
topics
that
could
be
proposed
any
are
of
interest
of
all
of
these
communities,
and
then
we
will
discuss
those
in
on
former
fashion.
We
had
a
whole
long
list
of
ideas,
but
unfortunately
we
had
a
chance
to
only
tackle
three
of
them
cause
so
long
and
interesting
discussions
on
all
of
them,
but
again
data
model
convergence
between
live
with
m2
m1
DM.
A
In
particular,
we
were
discussing
and
also
the
role
of
the
the
modern
hybrid
media
formats,
such
as
a
score
corinne
format
and
an
internal
choral
when
it
when
it
comes
to
like
with
them
to
him
and
also
briefly
touch
the
access
control,
modeling
aspects.
Unfortunately,
I
don't
have
more
time
on
that
topic
today,
but
you
can
go
and
check
out
the
meeting
minutes
on
the
github
page.
B
Okay,
you
know
the
topic
is
IOT
data
model
convergence,
I'm
going
to
talk
about
two
activities
that
are
going
on.
One
is
extensions
to
schema.org
for
IOT
and
the
other
is
one
data
model,
as
Ari
just
mentioned,
so
for
IOT
extensions.
What
we
want
to
do
is
extend
what
we're
working
on
is
extending
the
schema.org
model
to
accommodate
IOT,
semantics
and
schema.org.
If
you
don't
know,
is
a
public
online
resource
that
provides
semantic.
Interoperability
for
for
web
pages
basically
is
currently
being
used.
B
So
we're
we're
also
connecting
those
semantics
to
what
we
call
features
of
interest
and
I
have
a
little
more
description
of
what
that
is
a
little
later.
These
definitions
are
currently
based
on
an
emerging
meta
model,
that's
pretty
popular
across
different
device,
vendors
and
different
service
providers,
and
it's
a
sort
of
an
affordance
model
consisting
of
properties,
actions
and
events,
and
those
things
are
rolled
up
into
what
we
call
capabilities
and
I
have
some
more
detail
about
that
as
well,
and
we're
building
a
community
contribution
process.
B
So
the
meta
model
is
basically
a
semantic
model
for
the
affordances
that
a
device
or
something
offers
for
software
to
interact
with
it.
We
call
those
interaction,
affordances,
there's
property,
which
is
some
readable
and
writeable
piece
of
state,
there's
an
action
which
is
how
software
induces
a
state
change
through
some
parameterization,
so
things
that
are
more
sophisticated
than
just
updating
some
resource.
B
The
meta
model
has
this
concept
of
a
capability
that
basically
rolls
at
the
set
of
events,
actions
and
properties
together
to
do
some
common
function,
like
you
know,
temperature
measurement
or
controlling
temperature,
or
turning
something
on
and
off.
So
we
try
to
keep
these.
You
know
small
and
granular,
so
we
have
a
system
where
you
can
compose
things,
but
they
could
be
larger.
B
Aggregations
like
you,
could
describe
an
air
conditioner
as
a
capability
or
a
pump
in
an
industrial
setting
as
a
capability
as
well,
and
we
also
define
datatypes
and
the
main
purpose
that
there
is
to
associate
semantic
meaning
with
a
piece
of
data
that
you're
sending
back
and
forth
as
an
event
or
or
action,
and
also
provides
some
data.
Constraints
like
you
might
have
to
say.
Something
is
a
string,
type
or
number
type
or
boolean
type.
B
So
that's
basically
a
diagram
of
the
model.
You
know
if
capabilities
that
provide
interaction,
affordances
of
types,
events,
actions
and
properties
and
they
exchange
data
items
that
are
data
typed
and
that's
really
our
whole
our
whole
model.
It's
really
pretty
simple
and
it
overlaps
with
a
lot
of
what
other
organizations
and
vendors
are
already
using
in
terms
of
semantic
definitions.
B
Additionally,
you
know
you
have
the
problem
of
quantities
in
engineering
units
and
things
like
that,
and
there
are
existing
ontology
that
already
described
those.
So
our
intention
is
to
connect
to
those.
In
addition,
there
are
some
more
base,
basically
Tala
G's
such
as
SSN
and
so
si
ref
that
that
have
concepts
that
are
useful
as
well.
So
we
want
to
be
able
to
use
those
to
extend
the
definition,
and
this
is
all
being
done
in
various
RDF
languages,
a
JSON
schema
and
turtle.
Things
like
that.
B
So
the
this
integration
allows
us
to
take
a
feature
of
interest
like
something
in
a
building
or
a
room
or
a
floor,
or
you
know,
part
of
a
automobile
apply
an
IOT
schema
definition
that
tells
you
how
to
sense
it
or
actuate
it,
and
then
also
combine
that
with
the
parameters
of
data
that
you
exchange
with
it.
So
the
software
knows
sort
of
what
is
what
it's
interacting
with
and
what
influence
is
expected
to
have
in
the
physical
world
and
also
sort
of
what
kind
of
data
to
send
it.
B
So
this
is
getting
pretty
close
to
an
end-to-end,
bare-bones
semantic
system,
and
these
look
like
like
this:
that's
the
same
basic
interaction
model
with
events,
actions,
properties,
data's
data
schemas
as
data
types
connected
through
various
RDF
property
types.
Two
features
of
interest,
like
has
feature
of
interest-
you
could
just
say
this.
This
has
a
feature
of
interest.
You
can
say
this
is
a
property
of,
or
this
is
associated
with.
There's
a
set
of
relation
types
or
property
types
in
RDF
that
we're
creating
to
to
enable
this,
and,
of
course,
these
are
all
extensible.
B
We
have
some
high-level
interoperability
demonstrations
using
node-red,
where
you
essentially
feed
in
the
semantic
terms
and
the
API
is,
can
be
generated
through
through
some
automation
and
we're
currently
working
on
a
submission
process.
So
if
you
want
to
make
some
new
semantic
definitions
fill
out
a
form
and
submit
it,
and
we
have
some
automation
behind
that
that
that
creates
the
proposed
RDF
instances
for
that
we
have
a
w3c
community
group
schema
extensions
for
IOT.
That
is
still
ramping
up.
B
B
That's
that's
really
the
first
challenge
and-
and
you
know
to
put
a
point
on
it:
it's
not
an
API
description,
it's
not
something
like
swagger
or
anything
like
that.
It's
it's
semantics,
only
I
have
a
couple
of
examples
and
it
depends
on
having
a
way
of
binding
a
protocol
to
this
to
actually
use
it
in
the
system.
So
something
like
a
w3c
web
of
things
thing
description
can
contain
a
protocol
binding.
We
can
use
swagger
as
a
protocol
binding
or
open
API
and
I.
B
So
we've
built
a
little
DSL
domain-specific
JSON
format
for
this,
and
this
is
really
this
is
an
example
of
the
JSON.
But
if
you
strip
out
some
of
the
point,
Shu
a
ssin,
it's
easier
to
see
what's
going
on
here
and
essentially
it's
it's
making
the
same
kind
of
definitions
you
would
make
in
RDF,
but
just
using
JSON.
So
you
essentially
have
here's
a
definition
that
I'm
making-
and
here
are
its
terms,
and
here
are
the
definitions
of
the
terms.
B
The
status
of
this
is
we've
been
working
on
it
for
about
six
months.
It
was
suggested
from
one
of
the
ZB
hive
meetings
where
they
bring
in
people
from
across
industry
to
talk
about
what
some
of
the
big
problems
are,
and
the
consensus
was.
One
of
the
big
problems
was
lack
of
a
common
data
model.
So
we
went
to
create
this
group
to
address
it.
It's
not
an.
D
Okay,
so
I'm
gonna
give
an
update
on
the
w3c
web
of
Things
project,
which
is
actually
just
wrapping
up
its
first
two,
your
charter
and
actually
there's
an
interest
group
and
there's
a
working
group
inside
w3c
and
overall,
is
about
200
participants,
different
organizations
and
people,
universities,
for
example.
We
just
had
a
workshop,
which
is
mentioned
briefly,
where
we
discussed
even
charter
for
the
next
round
of
standardization,
but
we
also
discussed
various
applications
and
use
of
semantics,
for
example,
in
the
in
the
in
the
IOT.
We
also
have
two
documents.
D
D
We
also
been
working
on
the
scripting
API.
However,
we
decided
not
to
make
this
normative,
because
we're
still
working
on
the
use
case
of
the
deployment
models
for
it,
but
this
is
still
under
development
and
it's
meant
to
make
it
easier
to
write
things.
I
owe
to
York
raishin
layers
or
program
IOT
devices.
Behavior
is
also
a
document,
the
buying
templates.
D
So
again
we
have
two
rec
track
items
and
we
have
a
note-
and
this
is
also
going
to
be
a
note-
as
are
the
security
guidelines,
so
yeah
there's
different
levels,
but
basically
rec
is
a
full
standard.
It's
a
recommended
standard.
Our
recommended
use
and
a
note
is
just
an
informative
document.
So
rec
is
normative
and
a
note
is
informative.
D
And
just
submit
more
detail
use
these
documents,
so
the
architecture
is
just
I
mean
it's
technically
normative,
but
it's
it's
a
fairly
high
level
document
that
describes
the
overall
use
cases
and
applications,
but
it
also
talks
about
I,
guess
the
philosophical
model
or
the
semantic
model
of
IOT,
including
our
high
level
affordances,
which
I'm
gonna
talk
about
in
a
minute
and
in
the
thing
description
is
a
more
detailed
document
that
gives
the
exact
information
model
for
this
document.
It
also
gives
a
json-ld
1.1
serialization
of
the
information
model.
D
So
there's
a
bit
more
information
here,
so
so
I
think
I
just
want
to
mention
start
over
here.
So
there's
a
certain
set
of
constraints
about
you
know
what
has
a
TD
and
what
does
a
TD?
What
roles
play,
what
kinds
of
devices
can
have
TVs?
Can
you
have
devices
both
produce
and
consume
TDS?
What
functionally
serve
in
a
network
of
devices
and
so
forth?
D
We
also
have
a
definition
which
I
think
is
are
practically
important
in
this
in
this
talk
about
our
interactions.
What
is
an
interaction
and
we
use
this
concept
of
an
affordance.
So
as
a
difference,
this
is
coming
a
user
interface.
You
know
our
research
idea,
so
a
door
handle
is
a
good
example
of
an
affordance,
so
you
have
a
door
and
you
want
to
operate
on
the
door
and
you
got
to
figure
out
how
do
I
operate
on
this
door?
D
How
I
open
it,
and
so
you
want
to
do
a
certain
thing
at
a
high
level,
you
and
open
the
door,
but
you
got
to
look
at
the
door
and
figure
out
how
do
I
use
the
door
to
open
it?
There
might
be
different
kinds
of
handles,
but
B
you
have
some
information
about
how
to
operate,
reach
those
handles
to
cause
the
door
to
open
and
so
there's
a
way
to
aspects
to
the
finger
scription.
One
of
the
semantics.
D
You
know:
what
does
this
thing
do
and
what
is
it
I
mean:
there's
a
how
do
I
use
it?
You
know
what
protocol
vary,
what
bits
do
I
put
on
the
wire?
What
protocol
do
I
use
to
make
it
do
that
thing,
and
so
the
things
efficient
describes
both
of
these
and
in
practice
it's
this
json-ld
1.1,
doc,
JSON
document,
and
because
it's
json-ld
now
it's
json-ld
1.1,
which
is
still
a
spec
in
progress.
If
people
have
had
previous
experience
with
Jason
LD
1.0
I
really
feel
sorry
for
you.
D
Ok
1.1
is
a
vast
improvement
in
particular
edition.
Ld
1.0
was
meant
to
be
a
serialization
format
for
RDF
databases,
.
.
and
it
was
a
horrible
JSON
document
had
all
kinds
of
weird
conventions:
json-ld
1.1.
The
goal
is
to
allow
people
to
semantically
annotate
idiomatic
JSON,
as
used
in
practice.
So
so
really
we're
going
to
do
is
take
jason
as
using
web
services
or
for
data
data
flows
and
be
able
to
annotate
it
with
semantics.
And
so
this
is
the
json-ld
1.1
document,
which
means
is
much
more
idiomatic
json
and
much
closer.
D
So,
for
example,
if
you
can
use
the
IOT
schema.org
terminology,
you
can
just
pull
it
in
with
an
external
reference
like
this
and
then
use
those
terms
with
an
IOT
prefix,
for
example,
iot
light
and
became
a
type
of
isis's,
so
I
think
one
of
the
interesting
things
going
forward
is
how
things
like
SDF-
and
this
you
know,
work
together.
So
I
think
in
practice
will
happen.
You'll
have
an
SDF
definition
for
a
bunch
of
data
models
in
various
ecosystems.
D
Well,
the
compiler
that
compiles
down
to
RDF
based
definitions,
right,
maybe
Jason
LD,
but
we
can
then
pull
in
and
then
we
can
annotate
this
with
those
definitions
and
then,
when
this
gets
pulled
into
a
database,
you
know
you
pick.
Please
pull
it
into
an
RDF
database,
you'll
be
able
to
pull
in
so
definitions
and
have
all
the
semantics
come
along
for
the
ride.
So
that's
kind
of
how
this
fits
together.
The
SDF-
and
this
are
kind
of
complementary
mechanisms.
D
Sdf
defines
data
models,
lets
you
attach
data
models
to
protocols
and
and
and
also
data
schemas,
so
anyways.
These
are
additional
notes,
but
the
most
important
one
here
is
that
the
buying
templates,
which
describe
how
the
protocol
bindings
work
for
particular
ecosystems
and
protocols-
and
this
is
meant
to
be
a
living
document.
You
know,
as
we
uncover
new
things
may
be
modeled
we'll
describe
what
is
the
typical
way
to
use
those
those
ecosystems
with
with
a
thing
scription.
D
Now
our
current
status,
we
quite
late
in
the
process,
decide
to
use
Jason
ld1.
Previously
we
used
1.0,
but
it
was
had
too
much.
You
know
resistance
because
of
the
syntactic
difficulties.
Well,
1.1
has
been
a
great
improvement.
We've
been
working
with
the
json-ld
1.1
working
group
in
WPC
to
make
sure
it
has
all
the
right
stuff
for
us
to
do
things
in
a
natural
way
and-
and
we
basically
also
have
a
bunch
of
default
values
in
our
systems.
You
don't
specify
everything.
You
know.
D
A
lot
of
things
have
defaults
and
you
can
just
leave
them.
Leave
them
out
and
they'll
do
something
reasonable.
We
also
have
security
metadata,
so
you
can
declare
what
kind
of
security
is
required
to
access
a
thing.
This
is
an
extensible
mechanism,
so
we
have
a
few
basic
things
and
they're
all
ready,
but
the
idea
is,
you
can
pull
an
external
vocabulary's.
So
if
someone
defines
a
new
format
or
a
new
security
mechanism
for
controlling
access
to
a
thing,
you
can
declare
the
public
metadata
about
that.
D
So
you
can
say
you
know
what
is
needed
to
get
authorization
to
get
access.
This
thing
it's
not
going
to
put
you
know
private
information.
In
the
thing,
it's
just
going
to
say,
here's
what
you
need
to
know
to
get
in
anyway.
These
extension
points
in
addition
to
the
screen
metadata,
you
can
potentially
have
additional
protocols
and
header
options.
Look
at
this
and
you
can
also
have
a
semantic
annotation
and
actually
the
semantic
annotation
can
take
place
at
the
entire
thing
level.
D
So
I
think
that
one
thing
that
I
presented
oma
but
I
don't
have
presented
here
is
we've
noticed
a
few
places
where
we
need
some
more
work
on
aligning
our
standards.
So
one
thing
is
a
lot
of
standards
now
use
JSON
schema
embedded
inside
them
like
open
API
uses
it
SDF
uses
it.
We
use
it,
but
JSON
schema
itself
doesn't
really
have
a
single
unified.
You
know
formal
standard
associated
with
it
there's
a
draft.
This
is
something
we
feel
strongly
about.
I
feel
strongly
about.
It
needs
to
be
needs
to
be
moved
along,
I.
D
Think
addition,
the
JavaScript
and
scripting
API
we
to
more
formally
understand
the
deployment
model
for
that
and
also
figure
out
how
it
works
with
edge
computing,
is
an
edge
computing
service,
a
thing
or
is
it
a
web
service?
You
know,
do
I
open
the
API
for
it.
Could
you
is
your
thing
description?
Is
there
some
common
thing?
We
can
do
that,
describes
both
things
and
web
services
and
is
an
edge
service,
a
thing
or
a
web
service.
D
And
yeah
so
I
in
in
the
presentation.
These
are
all
live
links.
If
you
want
to
have
some
reading
material
in
the
plane,
you
have
nothing
else
to
read:
here's
here's
all
the
specs
are
and
the
important
one
here
is
going
to
be
the
candidate
recommendations
or
the
architecture
and
the
things
portion
itself.
There's
many
examples
by
the
way,
as
well
in
the
thing
scription
I
think
busily
last
slide
so
questions
they.
C
E
D
D
There's
also
an
IETF
draft
which
has
expired,
because
it
was
an
effort
to
move
JSON
schema
forward
as
no
IETF
draft
and
I
think
that
that
organization,
people
working
on
JSON
schema
will
be
very
interested
in
moving
that
forward
again
as
an
IETF
draft,
and
there
are
several
organizations
that
need
it,
and
so
this
is
something
that
I'd
like
to
see
happen
myself.
Okay,
I
need
more
questions.
F
Just
as
an
anecdote
on
this,
this
ITF
draft-
you
are
talking,
that's
the
draft
zip
schema,
oh
for
that
is
actually
used
as
the
basis
for
an
appendix
of
the
document.
That
is
right
now
getting
done
in
in
the
security
incident
handling
energy
here
Jason
I
would've,
but
fortunately
they
also
have
a
CDL
appendix
so
you
can
find
out
what
it
really
means.
So
there
are
various
versions,
I
think
the
newest
versions
of
JSON
schema
Olga
also
being
submitted
to
the
IDF,
so
you
can
also
find
them
as
internet
draft
and
ok.
D
D
D
Yeah
so
I
use
one
challenge
here
is
that
we
see
JSON
schema
being
embedded
in
several
reasons
and
there's
a
real
chance
for
these
to
diverge,
especially
we
talk,
even
if
we
have
common
requirements
like
sea
bore
extensions.
So
we
really
need
to
I
think
look
at
this
problem
and
see
if
we
can
take
that
little
piece
and
get
that
more
standardized,
because
there's
there
really
is
a
need
for
kind
of
a
generic
data.
Schema
and
JSON
schema
right
now
is
used.
G
Okay,
so
hello,
everyone,
my
name,
is
Val
Petrov
and
I
will
be
presenting
you
P,
which
is
I,
will
start
by
defining
the
problem
statement
and
then
I
would
describe
how
we
try
to
solve
this
problem.
So
during
the
which
act
on,
we
were
discussing
that
all
the
applications
and
interoperability
that
we
have
been
trying
to
do
or
great
if
you
have
fresh
new
IPs,
but
once
you
have
some
very
constrained
networks
like
Laura,
SiC
box
and
reality
and
others
and
or
very
constrained
devices.
G
So
in
order
to
try
to
make
interoperability
with
such
kind
of
devices
possible,
it's
important
to
be
able
to
understand
the
data
that
they
are
sending
and
be
able
to
format,
data
that
come
from
other
applications
so
that
they
can
be
received
by
those
devices
and
after
some
considering
well
in
the
IDF.
We
know
how
to
do
data
modelling,
who
use
yunk,
and
we
decided
to
try
to
use
the
same
thing
here
as
well.
And
of
course
we
have
the
added
benefit
that
I
want.
G
G
We
can
add
some
restrictions
of
the
final
results.
We
can
add
even
units
so
that
the
data
can
be
semantically
annotated.
And
yes,
then
we
have
some
specific
extensions
to
the
yonk
model
so
that
in
the
end
we
have
the
parsing
collage
ik.
For
example.
Here
you
can
see
we
can
have
some
information.
What
is
which
are
the
bits
that
need
to
be
passed
from
the
binary
representation
in
the
position
attribute
and
using
such
kind
of
information?
G
We
are
able
to
read
in
the
binary
data
and
then
obtain
some
usable
values
that
could
be
sent
as
JSON
or
something
else
as
a
receipt.
We
have
working
example
and
that
people
can
play
with
I
mean
we
have
like
playground
where
people
can
submit
their
own
models
and
their
data
and
see
what
will
be
the
result
and
the
link
for
this
is
in
the
Moishe
hackathon
wiki
page,
which
I
can
make
sure
to
put
it
in
the
ether
pad.
G
Not
so
so
that
would
be
useful,
and
so
in
the
extensions
we
have
some
other
useful
things
like
being
able
to
do
multiplication
and
addition
of
additional
I'm,
some
arithmetic
of
operations
with
the
values
to
attend.
We
can
do
mapping
between
battles
using
announce.
We
can
have
choice
between
different
possibilities
if
we
know
that
some
part
of
the
format
depends
on
some
part
of
some
other
part
of
the
payloads,
such
kind
of
things,
error,
I
currently
also
supported,
and
that
is
to
see
if
there
is
any
interest
for
such
kind
of
work.
G
F
H
F
F
Mechanism
for
translating
between
two
formats
at
the
bit
level,
in
this
case
the
uncompressed
at
the
compressed
form
of
record.
So
it's
a
little
bit
different,
but
I
think
many
of
the
questions
that
come
up
when
you
do
something
like
this
are
similar.
So
if
somebody
wants
to
do
an
interesting
paper,
compare
for.
H
K
So
the
history
of
this
is
that,
quite
for
quite
some
time,
the
core
working
group
has
promised
to
make
coop
available
over
transport
with
a
single
your
eyes
scheme
over
all
the
transports
and
a
group
of
people
that
is
build
house
in
SME
have
taken
up
work
on
this
again
around
last
ITF
and
got
together
and
first
try
to
cite
a
few
goals
of
what
we're
actually
trying
to
do
in
in
this
kind
of
small
sub
working
group
with
what
we
call
our
mini
Charter.
So
this
is
nothing.
K
K
And
we
are
you
looking
into
coop
over
SMS,
particularly
for
the
reason
that
it
is
an
example
of
a
co-op
transport
that
is
not
using
the
rest
of
the
IP
stack,
but
using
something
different.
There
are
other
up.
There
are
other
transports
in
the
works
that
have
similar
properties,
but
by
dealing
with
co-op
over
SMS,
we
hope
to
rule
out
any
any
roadblocks
for
those
other
transports
for
the
for
the
single
scheme
we
are,
and
especially
the
uriah
lasting
problem.
We
are
drawing
a
lot
from
HTTPS
alternative
services
for
the
for
announcing
the
active
transports.
K
We
are
building
on
Bill's
previous
work
in
the
protocol
negotiation
document
that
is
so
far
still
being
worked
on
under
a
core
name,
but
most
of
this
is
experimental
enough
that
we
are
doing
it
in
in
t2
TRG,
because
it
is
a
research
topic.
We
don't
really
have
a
clear
path
on
on
how
to
get
there.
Only
the
coop
over
SMS
work
is
aiming
to
be
to
start
in
call
ready,
because
that's
that
should
be
a
simple.
K
That
should
be
a
simple
and
nothing
we'd
appreciate
if
more
people
provided
input
for
this,
so
we
are
doing
we're
tracking
the
goals,
the
individual
problems
to
solve.
In
an
initial
tracker,
there's
the
your
I
we'd
appreciate
look
eyes
on
the
bully,
the
approaches
that
we
are
discussing
so
far
and
the
end
yet,
and
the
overall
mini
chato.
F
C
C
C
And
now
for
something
completely
different,
so
the
chest
asked
me
to
not
cut
with
some
outrageous
thoughts
about
IOT
and
edge
computing,
which
I'm
also
guilty
of
having
introduced
to
you
here
in
this
group
some
meetings
ago
and
I
did
about
so
these
things.
These
thoughts
here
are
actually
less
defined.
Then
you
would
maybe
think
right
now.
So
just
please,
but
so,
if
you
ever
found
yourself
sitting
in
a
management
meeting
was
a
piece
of
paper
and
a
pencil,
and
you
know
you
making
checks
for
every
buzz
word
that
comes
up
IOT
edge
computing.
C
This
is
the
ideal
password
bingo
and
the
problem
is
not
test.
These
are
not
relevant
areas
or
important
topics.
It's
just
these
terms
are
so
broad,
it's
almost
not
possible
to
talk
meaningful
about
them
without
being
really
specific.
What
you
mean
so,
for
example,
in
there
all
these
different,
you
know
vertical
scenarios
and
that's
reality.
C
Whole
networks
they
have
you
can
all
you
can
do
computing
in
all
of
those,
but
it's
going
to
be
completely
different
so
depending
on,
for
example,
you're
talking
about
virtualized
servers,
maybe
lambda
functions,
different
detection
models,
and
so
on
so
I'm
not
saying
there
is
no
compute
in
these
scenarios,
because
quite
the
opposite,
they're
interesting
things
you
would
do.
For
example,
in
industrial
see
there
there
are
different
opportunities
and
relevant
use
cases
for
computing.
C
They
have
different
programs
where
we're
computing
so
like
programmable
functions
would
be
interesting,
for
example,
close
to
the
deterministic
networking
you
want
to
have
maybe
a
low
latency
control
function.
Some
people
talk
about
virtualizing
PLC,
so
putting
them
in
a
data
center.
So
that's
all
computing
somehow,
but
to
be
honest
as
yourself.
So
what's
the
research
there
I
mean
this
is
mostly
server
virtualization.
Maybe
you
have
some
specialized
hardware
system
that
understand
real-time
a
little
bit,
but
I
think
this
is
the
ITF.
C
C
C
So,
let's
really
look
built
into
the
technical
property.
Is
a
criteria
that
interesting
to
discuss
these
things
in
a
meaningful
way?
So,
for
example,
I
think
it's
interesting
to
understand
what
are
the
interaction
models
so
stateless
function
or
more
like
a
server
X
are
more.
What
do
we
actually
mean
by
H
or
maybe
in
network
computing?
So
it
are.
We
talking
about.
You
know
intercepting
flows
looking
into
packets.
Is
that
more
sample
application
data
units
or
just
you
know,
coop,
request
response
where
we
trigger
trigger
and
computation?
C
Sometimes
you
you
have
the
notion
of
mobile
code.
So
what
what?
What
is
that?
So,
what
is
your
execution
environment
for
that?
Or
how
do
you
program
that
and
also
really
important?
What
is
the
security
and
it's
a
trust
model
in
these
things?
So
so
what
platforms?
How
can
you
trust
platforms?
How
can
you
trust
trust,
the
the
functions,
the
results
of
those
functions
and
what
I
think
we
have
really
learned
in
what,
meanwhile,
that
it's
not
acceptable
to
postpone
that
discussion?
C
So
these
are
questions
that
have
to
be
asked
from
the
beginning
and
so
in
the
same
way
as
IOT
is
a
very,
very
broad
topic.
Of
course,
edge
computing
itself
is
used
in
different
way
and
Brian's
today.
So
this
is
a
Etsy
mobile
computing
diagram
and
well,
it's
kind
of
interesting
like
to
some
parts
of
the
industry,
but
because
you
can
do
interesting
things
but
technically
from
a
say,
networking
perspective
or
from
awesome
management
perspective.
C
I
mean
this:
is
just
flour
computing
extended
to
external
pots,
on
base
stations
and
doing
something
radio
specific,
perhaps
so
not
irrelevant,
but
doesn't
require
that
much
say
networking
or
internet
research.
So
when
you,
when
you,
when
you
say
edge
computing,
what
about?
What
do
you
mean?
So
do
you
mean
which,
like
versus
gateway
platforms
between
this
whole
continuum,
from
from
cloud
to
edge,
or
do
you
mean
something
like
offloading
compute
in
constraint,
networks
perhaps
also
interesting,
possibly
relevant?
Or
do
you
mean
what
I
showed
earlier
like
disability
computing
stream
processing?
C
So
quite
often
when
people
say
actually
they
actually
mean
something
like
doing
something
on
the
gateway
with
cloud
concepts
and
cloud
technologies,
and
these
technologies
are,
of
course
very
well
understood,
very
successful,
widely
deployed,
so
mm
question
is:
is
edge
computing,
actually,
the
best
term
anyway,
so
edge
computing,
as
I
explained
earlier
often
has
this
connotation
of
extending
the
cloud
to
the
edge
and
basically
applying
the
same
technology,
same
concepts,
but
also
maybe
inheriting
the
same
trust
models
or
trust
anchors
in
those
systems
and
well
there's
a
bursty
potential
of
the
danger
of
well.
Basically,.
C
Dependencies
on
the
cloud,
so
things
like
centralization
or
maybe
promoted
motivated
by
business
models,
that
expect
some
continuous
you
know,
contract
renewal
and
so
on
so
unnecessary
dependencies
or
unwanted
dependencies.
So
that
may
not
be
the
best
notion
to
start
with
in
the
beginning,
and
so
I
I
thought
a
bit
about
okay.
C
C
That
may
be
a
bit
expensive
in
certain
constraint,
IOT
scenarios,
so
because,
essentially
that
this
is
a
circuit
like
connectivity
model,
where
it's
very
difficult
for
me
to
say:
where
should
I
actually
offload
these
computations
to?
There
are
aquifers
that
know
where
it's
available,
compute
capacity
and
so
on,
but
they
don't
really
know
how
the
network
is
feeling
at
a
certain
point,
because
it's
so
cheap
and
over-provisioned.
C
Another
little
issue
with
this
approach
is
that
often
we
find
ourselves
and
having
to
translate
between
different
namespaces,
so
when
using
a
DNS
and
so
on.
That
means,
if
we
want
to
you,
know,
offload
something
efficiently
in
a
lightweight
fashion.
What
we
may
be
required
to
do
lookups
first
and
have
the
infrastructure
the
secure
interceptor
for
that
installed.
C
So
thinking
about
interesting,
say
functionality
or
say
use
cases,
perhaps
in
say
computing
with
considering
things
scenarios
so
function
of
loading.
I
guess
is
very
interesting
for
power
saving
of
a
load
concentration
management,
things
like
trigger
execution
or
reactive
programming.
If
this,
then
that
so
when
new
data
is
generated,
but
also
perhaps
managing
all
that
data,
so
then
you
can
remove
all
scenarios,
perhaps
offloading
data
to
a
custodian
transfer
or
storage
service,
and
then
the
data
presenting
pipelines
that
I
mentioned
earlier
so
be
able
to
construct
different
pipelines
in
a
lightweight
fashion.
C
So
just
a
bit
full
of
thought.
So
when
I
said
in
the
beginning,
I
hope
th
computing
will
never
happen.
That
was,
of
course,
outrageous,
but
what
I
actually
meant
is
I
want
to
discuss
with
you
how
we
can
make
it
happen
in
the
ITF
and
and
the
thing
to
think
research
group,
because
I
mean
to
be
honest.
We
know
that
there's
so
many
forums-
and
you
know
industry
alliances
that
do
something
was
edge
computing
right
so
and
we
have
to
ask
ourselves
so
where
can
we
actually
make
at
the
end?
C
What
are
the
research
questions
that
are
interesting
and
not
yet
solved,
and
what
so
I
normally
also
prefer
doing
for
this,
like
the
unchartered
territories,
these
days
really
pursue
an
application
driven
development
approach.
Where
you,
you
know,
identify
few
really
interesting
things
that
are
maybe
hard
to
do.
C
These
days
was
legacy
stuff
and
then
really
do
a
deep
dive
and
thing
outwards
to
what
you
need
to
make
this
happen,
and
but
it's
really
important
to
really
leave
that
you
know
cozy
business
case
level
and
and
and
really
try
to
figure
out
what
the
issues
are,
and
so
the
interesting
questions
to
look
at
is
really
what
what
what
needs
to
happen.
What
are
the
interaction
models?
The
compute
computation
models
that
we
would
like
to
support.
C
The
the
important
pillars,
in
my
point
of
view,
are
decentralized
lightweight
approach.
This
joint,
optimization
of
computing
I
think
it's
very
promising
hasn't
been
done
very
much
and
I
didn't
mention
it
earlier,
but
object.
Security
is
probably
as
a
key
to
a
lightweight
non
connection,
based
approach,
of
course,
since
we
are,
we
are
doing
something
with
computing
and
network
in
the
proposed
coin.
Rg
so
I
think
this.
So
these
ideas
here
could
tough
hire
quite
nicely.
Was
the
coin
agenda.
C
D
Think
it's
one
additional
challenge
for
edge
computing,
especially
in
decentralized
fashion
and
that's:
what's
a
secure
execution
environment
entail?
Oh,
yes,
it's
not
only
the
object,
security
security,
the
data
being
transmitted
around
it's
like
how
do
I
get
a
blob
of
code
and
run
untrusted
code,
an
untrusted
platform
right.
You
know
with
other
pieces
of
untrusted
code
in
the
same
interested
platform.
Wait.
How
does
that
actually
happen
and
I
think
there's
technologies
to
do
that.
They'll
still
be
standards
for
how
we
deploy
these
things
and
manage
routes
of
trusts
and
and
so
forth.
Right.
D
D
F
That
custom,
Omen
reputations
are
pretty
important,
word
I.
Think
in
this
context.
So
the
problem
that
I'm
really
interested
in
in
this
context
is
one
word
that
has
been
said
about
50
times
in
this
room
already,
but
that
nobody
seems
to
know
exactly
how
to
handle
in
the
IOT
space,
which
is
the
word
event.
So
there
is
an
event
happening,
and
what
do
we
do
now?
F
Yeah
now,
if
only
one
event
is
happening,
it's
not
a
big
problem,
but
if
several
of
them
are
happening
and
local
memories
of
things
are
overflowing
and
so
on,
we
need
something
like
the
custodial
transfer
you
talked
about,
and
how
do
we
actually
run
this
in
an
environment?
We
are
really.
We
aren't
that
were
not
only
do
these
things
have
nefarious
jobs
that
we
are
talking
to,
but
even
even
if
they
are
all
on
our
side
and
wonderful,
are
they
going
to
work
I
mean?
F
Is
it
a
custodial
transfer
to
something
that's
going
to
burn
together
with
me
in
five
minutes?
Instead,
a
useful
thing
to
do,
and
some
reputation
is,
is
not
just
past
behavior
here
it's
it's
also
knowledge
about
how
the
thing
in
all
its
fingers
will
actually
act
in
a
particular
environment,
right
and
I.
Think
that
there's
so
much
that
we
don't
understand
about
how
to
structure
systems.
Of
course,
you
can
always
say
that's
the
job
of
engineering.
They
can
plan
this
and
and
it's
all
wonderful,
but
we
know
that
most
systems
are
really
engineered.
F
This
way,
if
I
build
a
single
bridge,
I
can
make
sure
that
this
bridge
is
kind
of
connected
properly
and
so
on.
But
if
I
have
a
real
IOT
system
like
a
smart
city
or
a
smart
nation
or
even
home,
this
will
evolve
over
time
and
it
would
be
nice
if
we
really
had
ways
to
get
these
storage
problems.
These
problems,
where
do
we
put
the
data
if
we
had
a
way
to
resolve.
M
Hi
this
is
Schuler
and
so
I
loved.
Your
talk,
of
course,
I
like
that.
You
we're
talking
about
pillars,
or
at
least
adjectives
that
sort
of
describe
the
space
that
we
should
target.
One
of
the
questions
that
I
think
will
be
echoed
in
the
next
talk
and
the
draft
that's
associated
with
it
is.
There
are
many
problems
in
this
space
or
building
blocks
or
I
love
the
building
block
idea.
M
There
are
many
building
blocks
that
we
could
build,
but
which
are
the
ones
that
are
most
appropriate
for
this
culture,
for
the
idea
that
our
they
don't
necessarily
have
to
be
our
expertise
now
but
become
our
expertise
but
sort
of
sit
squarely
in
our
domain
versus
some
of
the
adjacent
standards
organizations,
so
that
I
think
you
should
think
about
that
as
you
as
we
go
along.
So
that
was
really
why
I
wanted
to
get
up
here.
But
then
the
other
thing
that
you
raised
and
when
I
think
about
building
blocks
is
edge.
M
Computing
is
completely
a
misnomer,
because
this
really
isn't
just
about
computing.
It
is
about
reimagining
the
data
center
and
all
the
things
that
the
data
center
does.
In
a
place,
that's
more
proximate,
because
all
these
IOT
use
cases
have
requirements
that
that
really
far
away
cloud
can't
meet
and
so
edge
computing
is
a
combination
of
computing
and
storage
and
networking
and
control
points
and
data
management,
and
probably
some
other
things
so
which
of
those
are
our
domain
to
work
on
and
then,
when
I
think
of
the
edge.
M
C
So
yeah,
thank
you
so
I
understand
what
you're
saying,
and
so
they
say
notion
of
reimagining.
Center
I
have
some
sympathy
for
that.
On
the
other
hand,
I
was
wondering
whether
we
should
maybe
just
start
a
little
bit
smaller.
You
know
and
not
do
all
of
this
only
maybe
it
could
be
very
constrained
in
its
center.
Oh.
M
Sure
and
and
I,
so
let's
go
back
to
you're
constrained.
You
know
this
grows
out
of
some
of
the
co-op
activities,
so
constrained
devices
I
think
the
edge
challenges
that
the
IOT
equals
constrained
devices
because
you're
going
to
do
this
offload,
they
can't
all
be
in
constraining
devices.
So
it
challenges
that
a
bit.
So
you
could
talk
about
the
very
edge.
Sometimes
the
Bleak
called
the
bleeding
edge,
so
yeah
focus
there
on
the
kind
of
reimagined
data
center
in
quotes.
That
would
live
there
and
maybe
doesn't
provide
all
of
the
functionality
of
a
data
center.
M
L
You
mentioned
that
there,
obviously
all
these
activities
ongoing
even
the
standards
getting
an
idea
of
what
the
different
standards
groups
do
in
that
space
is
already
a
tremendous
task,
a
different
way
of
approaching
this
topic.
Here's,
however,
by
looking
at
what
the
folks
in
the
room,
for
example,
are
already
doing
in
terms
of
deployments
and
and
learn
from
that
and
maybe
take
it
more
from
a
sort
of
a
blight,
could
be
the
science
type
of
approach
and
see
like
what
worked
for
them
and
what
didn't
work.
You
know.
C
L
If
you
collect
that
sort
of
input,
I
think
there
may
be
also
some
interesting
sort
of
areas
of
the
investigations
that
arise.
That
actually
can
then
feed
back
to
the
idea
of
community
potentially
yeah.
Do
you
have
something
in
mind
already?
Well,
if
it
was
just
talking
I,
they
have
edge
computing
solution.
We
have
a
computing
solution.
There,
probably
five
other
companies
in
this
room
who
have
a
similar
solution,
Microsoft,
where
they
sure
Microsoft,
has
something
there's
folks
here
from
Amazon.
They
have
a
solution
and
so
on
and
yeah,
okay.
Okay,
thank.
I
So
much
like
what
we
do
in
the
data
center,
where
we
take
elements
in
Iraq
and
we
compose
them
into
a
platform
on
the
fly
right
and
we
have
protocols
and
security
and
things
to
do
that
in
a
trusted
way.
That's
what
we
really
want
to
do
at
the
edge.
So
I
want
to
move
compute
over
to
this
element,
but
it
doesn't
have
a
GPU
to
do
my
AI
algorithm.
I
So
I
combine
that
with
something
else
close
by
that
has
that
and
and
put
storage
over
here,
because
it's
it's
convenient,
that's
what
we
really
want
to
do
with
the
edge
and
and
I
think
that
that's
kind
of
what
you
were
saying:
Eve
move
the
data
center
down
towards
the
edge.
That's
a
big
vision
but
I
think
we
can
start
off
small
and
take
take
pieces.
What
are
those
protocols
to
do?
You
know
those
those
five
questions
you
had
at
the
beginning?
We're
really
good
right.
Let's
look
at
those
take
a
piece
of
that
yeah.
C
C
Yeah,
thank
you.
Yeah
I'm,
of
course,
from
familiar
with
these
models
and
solutions.
I
think
it
could
be
one
task,
perhaps
or
one
interesting
activity
to
find
out
whether
this
constrain
things
and
then
maybe
offloading
to
more
powerful
gateways,
but
that's
the
only
model
or
whether
they're
also
you
know
other
distributed
computation
models
that
are
interesting,
and
so
my
se
perspective
right
now
is
that
maybe
it
could
be
more
fruitful
to
you
know
not
ossify
on
the
edge
gateway
model.
D
So
I
like
your
idea
of
a
joint,
optimization,
a
dynamic
system.
However
I
want
to
occur
to
me
recently,
but
the
building
blocks
approach
that
is
taken
by
like
and
that's
the
problem,
emergent
properties
so
as
an
example
of
a
merchant
property
going
to
worry
about
that's
privacy,
so
a
data
format
by
itself
may
not
really
be
analyzable
terms
of
privacy.
A
data
access
mechanism-
you
know,
may
not
mean
anything.
D
Just
you
don't
know
what
data
you're
accessing,
but
together
there
could
be
a
privacy
risk
because
the
data
could
be
accessed
in
a
certain
way,
our
associate
with
a
person
or
whatever
right
so
I
think
one
one
thing
to
worry
about
here
is
that
we
can't
necessarily
decompose
things
nicely.
You'd
also
think
how
they
how
they
stack
together.
We
use
cases
yeah
and
I.
Think
when
you
look
at
these
joint
optimizations
you'll
get
build
additional
constraints
into
that,
like,
for
example,
competition
cannot
migrate
across
the
national
boundaries.
It
has
certain
data.
B
H
Okay,
good
afternoon,
ladies
gentlemen,
this
is
the
problem
state
of
IOT
integrated
with
the
edge
computing,
so
my
name
is
dawn
when
home
I'm
working
for
July,
so
first
I
thanks
to
chair
to
provide
another
time
to
show
our
demo.
Thank
you.
So
it
is
the
content.
The
first
we
describe
what
the
change
on
out
rep
and
ii
will
show
a
short
time
show
for
our
implementation
of
edge
computing.
The
goal
is
to
support
not
left
yeah
TC
stored
history
of
our
drift
in
I
trade,
ITA
103
meeting.
H
We
pretend
out
Rev
first
into
to
Giles
Terry
meeting
on
the
time
the
file
is
the
Jerome
version.
So
at
the
time
we
show
two
demos
scenario
as
a
use
case
of
IH
computing.
The
first
one
is
not
construction,
providing
or
moisturing
services
over
construction
site,
and
second,
is
real-time
control.
Monitoring
system
by
a
lottery
inverted
pendulum
system.
In
the
last
meeting
we
present
out
ref
in
on
free
ITF
work.
Meeting
at
the
time
to
fail
aim
is
to
zero
to
version
okay,
o
for
this
meeting.
H
C
H
File
aim
to
threaten
fidelity,
blah
blah
and
second,
we
integrated
the
sub
a
and
kept
analysis.
I.
Remember
that
Sabir
presented
and
discussed
at
IDF
100-miles
meeting
so
as
a
result
of
the
integration
duo.
Sir,
for
example,
Sabir
meteos,
if
torque
were
added
in
this
threat,
so
I
show
you
what
is
the
change
in
with
the
EOC
on
poor
example,
in
Section
3
background
the
material
suggested
we
should
clearly
separate
here.
What
is
the
color
computing
and
what
is
the
edge
computing?
H
So
we
try
to
add
the
separation
cloud
and
edge
computing
and
we
slightly
modify
on
second
Pole.
You
change
of
aut,
for
example,
strict
latency
and
jitter
Oh
bring
cost
on
interrupted
services
and
privacy
and
security
and
section
5.
There
are
some
to
edit
or
change
in
section
6.
Also
in
this
yeah
there
are
so
many
changes
section
7
use
cases
of
edge
computing
is
to
rename
state-of-the-art
of
ith
computing,
the
sub
a
and
the
cab
analysis.
H
By
tab
years,
I
thought
many
parts
are
moved
to
section
7.1,
common
aspect
of
at
ith
computing
service
platform,
and
then
the
many
detailed
texts
are
included
in
appendix
a
mobile
of
ith
computing.
So
I
think
that
you
can
find
variable
information.
What
is
H
computing
and
water
currently
use
open
source
product
over
standard
and
any
other
related
project.
So
you
can
refer
some
developer,
up-to-date
information
about
H
computing
yeah.
H
This
is
one
of
the
content
of
the
the
it's
computing
in
this
threat.
I
would
like
or
discussed
the
Gateway
based
architecture
or
ith
computing
I
emphasize
that
this
approach,
this
architecture
to
not
cover
or
the
case
of
edge
computing.
This
is
one
of
a
particular
way
when
we
implement
edge
computing,
for
example,
or
it
drew
tricht
or
similar
to
IOT
gateway.
H
To
connect
the
sensor
Y
device,
we
must
provide
a
genetic
function
in
the
southbound
area
and
to
connect
to
cloud
network
who
are
sober.
Then
we
need
to
educate,
wave
function
to
dos
bound
area,
and
one
important
thing
is
to
if
the
bidder
hk2
a
function
so
in
educated
function,
we
can
imagine,
for
example,
storagee
processing,
analyze,
pre-processing
intelligence
whatever,
so
we
suggested
this
kind
of
dog
it
way.
Architecture
of
ith
computing
yeah,
but
are
we
noted
there
are
lots
of
different
edge
computing
approach?
H
For
example,
each
cloud
educate
way,
distribute
edge,
North
divides
married,
a
node,
etc.
So
in
the
next
revision
we
will
add
some
different
edge
computing
approach
in
now,
trapped
and
I
would
like
to
ask
is
already
42
G
urge
adoption
what
we
need
a
word
up
to
the
progress
so
before
decision
this
adoption
I
want
to
show
our
demo.
What
is
age
computing.
H
Yeah,
so
in
current
draft
we
only
suggest
that
80
based
on
age
computing,
but
I,
also
discussed
about.
We
must
provide
many
cases
of
each
computer
in
a
bony
gateway,
pH
computing.
So
as
the
doc
says
that
we
must
consider
also
the
current
environment
as
computing,
also,
so
we
hoped
in
the
next
revision.
We
include
this
kind
of
the
cases
also
in
this.
F
F
H
Okay,
thank
you.
Thank
you.
Thank
you
yeah.
This
is
our
implementation
demo,
so
I
know
that
this
is
not
user,
but
as
Todd
says
that
there
are
a
lot
of
the
edge
computing,
for
example,
Amazon,
Microsoft
and
Google.
We
noted
the
companies
are
commercial
of
the
traditional
cloud
computing
company,
but
nowadays
the
company
also
moved
to
the
attraction
to
in
the
area
of
edge
computing
and
Samsung,
while
they're
or
arm
also
have
interest
in
edge
computing,
and
there
are
lots
of
the
project
regarding
of
the
edge
computing.
H
So
we
only
Soto
objects
of
this
demonstration
is
the
first
show
on
the
attempt
or
implementation
of
edge
computing
based
on
open
source.
Hx
HX
is
the
open
source
on
project.
There
is
the
umbrella
of
the
Linux
Foundation
on
air
f8,
so
you
can
use
it
edge.
X
free,
not
only
ITF.
There
are
lots
of
the
open
source
regarding
to
on
edge
computing,
so
you
can
choose
what
is
the
best
for
you
and
second
object.
H
The
core
is
to
provide
your
mapping
between
our
implementation
and
architecture
in
the
track,
and
the
final
is
of
all
important
thing
is
to
to
George
adoptions
of
sport
yeah.
This
is
the
service
scenario
yeah.
We
prepare
our
third
year
test
bed
in
the
front
desk.
So
after
my
presentation,
I
show
you
two
videos
the
working
scenario,
so
we
prepared
a
three-part.
One
is
to
send
thing
and
second
is
to
executing
system
part
and
find
the
third
is
to
attract
device
part.
H
So
what
is
our
eighth
complete
function
in
our
implementation
is
dead.
We
are
targeting
on
provided
intelligence.
When
you
talk
about
intelligence,
we
are
provide
two
pre-processing
data
and
prediction,
the
next
barrier
and
analyze
and
control.
So
to
do
this,
we
train
the
people
on
Vera
in
other
trouble
or
other
cloud
network,
and
then
we
copy
the
result
of
the
training.
The
ratted
is
the
tap
to
be
fired,
so
this
file
is
to
copy
our
AG
system.
So
if
we
retrieve
third
era
so
only
inference,
while
prediction
works,
don't
turn
in
HD
system.
H
So
the
main
Encore
of
our
service
Nereo
is
to
detect
normal
situation
and
on
normal
situation,
separation.
What
is
the
normal
and
what
is
on
normal?
If
we
think
about
our
previous
cases,
the
best
one
is
to
compare
the
threshold
failure.
So
if
some
period
exceeds
the
threshold,
then
you
can
imagine.
Oh
this
is
some
extent.
While
there
are
something
happen,
but
in,
for
example,
in
construction
area,
there
are
many
vibration
era.
H
There
are
many
big
noise,
so
if
we
compare
only
one
time
or
second
time,
the
threshold
failure
tent
or
desert
is
not
good,
so
our
intention
is
to
something
more
better,
so
it
is
related
to
the
machine
running
team
running.
So
in
this
case,
if
we
make
some
noise
there
are
some
patent.
So
in
our
IT
system
it
detects
it
is
the
abnormal
situation,
and
then
it
was
something,
for
example,
make
alarm
worst
make
some
action
to
the
actuator.
So
you
can
see
there
are
some
three
things.
H
H
So
first,
the
sensor
device
by
the
Arduino
port
or
it
is
connected
to
device
to
add
system,
and
then
it
cracked
air
up
from
sensor
and
then
it
through
process
Tara
and
prediction
make
some
inference,
and
then
he
analyzed
control
device
and
actual
device.
So
top
law
is
like
this
or
it
is
a
screenshot
of
each
process
so
for
it
click
let
received
sensor
from
the
earth
sensor
and
then
it
delivered
to
edge
system
using
MQTT
and
then
in
AI
platform.
It
analyzed
and
some
decision
and
this
decision
will
be
done
in
other
actuator.
H
So
we
implemented
a
model
in
the
padang.
This
is
the
pre
post
area
and
prediction.
I
know
that
a
lot
of
the
air
platform,
so
you
can
use
any
platform
as
you
want.
So
in
this
case
we
use
air
platform,
which
is
also
emitted
by
a
three.
So
the
flow
is
like
this
and
then
in
analyzed
and
the
control
device
cases
and
finally,
so
I
think
that
the
most
important
thing
is
to
result
or
the
next
step
of
the.
H
H
Object:
detection
model.
Yes,
sorry!
So
in
the
normal
state
the
object
detection
model
is
disabled,
but
is
something
happen
and
at
least
an
critic
make
impressed
then
in
it
enable
the
object,
detection
motor
so
I
know
that
this
is
a
very
simple
kitchen.
So
please
understand
that
this
is
not
a
commercial
product.
It
is
kind
of
that
pure
she
kisses
yeah.
H
So
in
the
in
our
trapped
we
propose
gateway
based
architecture
about
edge,
comparing
like
this
and
how
our
implementation
is
related
to
the
threat.
I
met
a
trifle
mapping
so
to
Sony
device
who
are
actuate
located
this
area
and
the
connected
device
who
are
correct
era
and
the
actual
device
are
located
in
the
HEA
network
function
and
then
pre-processed
era.
While
prediction
analyze
control
device,
locate
is
a
tenet,
each
company
function.
So
in
our
demonstration
we
don't
have
any
HC
k
to
a
function
because
we
don't
have
any
outbound
connection.
H
C
H
E
H
H
The
screenshot
of
the
laptop
yeah
so
as
I
mentioned,
that
this
is
the
terminal
of
the
Itchy
process.
That
is
the
same
thing
there
are,
and
so
Dara
is
to
trigger
to
AG
system
and
air
prep
on
to
something
port
or
ml
operation.
And
then
the
result
is
to
deliver
to
racket
is
ok,
so
the
explanation
is
a
little
long,
but
demo
is
very
simple.
So
if
we
make
a
noise,
you
can
find
that
failure
of
the
net
period.
There
is
some
change,
so
a
DoD
George.
H
H
H
A
H
A
I
was
wondering
cause,
I
mean
we're
also
building
some
standardized
data
models
here
at
IETF
Centinela,
for
example,
would
be
interesting
to
see
like
how
good
fit
that
would
be
for
this
kind
of
cases
and
then
bring
the
learnings
back.
Okay,
these
are
the
shortcomings.
Here.
It
works.
Well,
user
could
be
improved.
That
could
be
maybe
something
an
angle
have
here.
Yeah.
H
F
Observation
whenever
looking
at
his
demo
I
think
someone
should
write
a
paper
about
what
good
sensors
alpha
GTM
us.
So
we
have
had
temperature
sensors,
and
these
are
we
bad
because
they
are
so
slower
and
having
a
noise
set.
So
also,
it's
may
be
kind
of
noisy
yes,
and
so
that's
one
one
interesting
problem
that
we
all
share.
We
want
to
demonstrate
something.
Of
course
it
has.
F
Up
very
freakin
so
yeah
that
actually
work
much
better
than
I
thought,
but
another
nice
thing
that
came
up
on
last
Friday
was
that
we
had
some
component
of
a
standard.
They
had
a
minimum
time
constant
of
60
seconds
for
something
and
we
decided
we
will
never
ever
do
that
again
because
in
testing
and
it
doubles
demonstrating
things.
That
means
you
have
to
wait
for
something
for
60
seconds,
which
is
completely
ok
for
the
actual
intended
usage.
F
F
D
So,
to
get
back
to
what
we
could
do
for
standardisation
opportunities
here,
so
we've
been
talking
about
this
quite
a
bit
and
you
know
AI
is
a
service
interesting
thing
to
have,
but
what
I
want
to
do
is
be
able
to
plug
and
play
in
AI
service
with
a
bunch
of
other
things,
because
they
I
can
plug
and
play
a
sensor.
You
know
into
a
system
so
I,
don't
to
use
like
an
orchestration
statistic.
D
Node-Red
to
plug
in
to
compute
surface
so
I
like
to
see,
is
first
of
all
standard
way
to
describe
edge
surfaces
that
fits
in
everything's.
We're
doing
you
know
your
open,
API
or
whatever
I'd
like
to
see
some
sterilized
edge
services,
regardless
of
who
deploys
them.
You
know
there
should
be
standardized
ways
to
do.
A
I
am
saying:
there's
some
interesting
standards
related
to
this,
like
onyx,
describing
neural
net
models
that
could
be
used
right.
D
I.
Think
AI
interesting
is
especially
interesting
because
it's
not
Turing
complete,
so
you
can
still
upload
a
computation
without
worried
about
someone
employees
virus
you,
because
just
awful
your
neural
net
model,
so
I
think
it's
a
very
interesting.
You
know
thing
to
deploy
a
service
with
I.
Think
related
to
the
idea
of
AI
is
a
service,
is
how
can
we
extend
discovery
mechanisms,
so
we
have
discovery
mechanisms
for
like
devices
the
coyote
thing.
Those
same
sure
mechanisms
also
work
for
edge
computing
services.
Okay,
thank
you.
Thank
you
for.
H
H
M
Would
echo
what
Michael
McCool
just
said
about
discovery,
but
one
of
the
aspects,
or
at
least
one
of
the
objects
that
I'm
really
interested
in,
is
all
the
data,
because
I
think
with
all
this
computation
everywhere
that
the
data
gets
scattered.
If
you
will-
and
the
question
is,
if
you're
going
to
do
computation,
how
do
you
marshal
the
data?
Where
do
you
put
it
after
you're
done
and
that
we
need
some
mechanisms
for
discovering
its
existence,
its
access,
control,
etc?
So
I
would
argue
that
there
there
may
be
some
utility
to
some
kind
of
discovery.
H
Thank
you.
So
actually
this
draft
is
the
title:
is
the
problem
statement
or
I
didn't
do
the
edge
computing,
so
we
are
not
taking
on
the
make
sound
solution
or,
as
some
propose
or
so
I'm,
considering
of
the
how
to
hinder
their
your
command.
So
I
know
that
you
can
everything
in
one
thread
so
I'm,
sorry
or
what
is
the
best
way.
D
Think
about
edge
computing
is
a
micro
services
model.
There's
a
set
of
services.
You
can,
you
know
plug
together
in
different
ways.
One
service
you
can
certainly
have
is
a
storage
service
right
that
allows
things
like
serial
transfers
and
so
forth,
and
you
know
access
control,
so
I
think
it's
a
very
important
component
service
want
to
consider.
I
can
also
do
historical
data
searches
in
this
little
thing.
M
You're
right
I
was
completing
your
talk
and
Kurt
and
Dirk's
talk,
but
I
think
it
is.
The
thing
to
put
into
this
draft
is
more
about
the
problem
statement.
Oh
there's.
We
need
to
find
our
data
if
we're
do
in
network
compute
and
then
for
dirts
outrageous
opinions
segments.
It's
the
you
know.
What
are
the
building
blocks
and
maybe
one
of
them
is
you
know?
How
do
you
find
objects
that
are
not
merely
you
know
the
DNS
end
host
address
or
name
okay,.
F
Like
the
question
about
Kafka
cover
take
offense
ago,
somebody
was
demonstrating
something
then
that
essentially
went
into
the
XMPP
groups
then
descended
that
has
recently
published
and
they
try
to
build
things
together
with
the
X,
would
be
infrastructure
and
didn't
quite
get
it
to
work
in
the
end
they
just
said
screw
it.
We
are
going
to
do
this
with
Kafka,
because
it's
going
to
work
and
what
we
are
demonstrating.
Yes,
not
XMPP,
but
the
actual
day
tomorrow
yeah.
F
So
we
have
this
this
interesting
little
disconnect,
which
also
came
to
me,
came
up
in
the
question.
Why
aren't
you
using
standards
you
and
we
certainly
have
to
work
more
on
aligning
the
realities
of
what
you
actually
can
like
together
quickly
today
on
based
on
some
some
open
source
projects
that
are
widely
used
and
the
standards
that
we
are
building
yet.
F
Piece
of
work
that
should
and
fit
when
you
are
talking
about
technology
like
like
edge
computing
or
computing
for
things
that
have
a
high
integration
aspect
to
them.
It
becomes
more
and
more
difficult
to
ignore
this
issue.
I
think
with
this,
we
are
out
of
time
only
for
a
while.
Thank
you
all
for
coming
and
I
hope.
We
can
put
something
interesting
ever
for
Singapore
and
see
you
all
in
Singapore.
You
think
you
like
it.
You.