►
From YouTube: COINRG Research Group Interim Meeting, 2020-02-05
Description
COINRG Research Group Interim Meeting, 2020-02-05
A
Okay,
we're
recording
the
meeting,
because
I
will
probably
take
just
a
few
notes
and
we
will
be
able
to
send
the
meeting
also
to
Jeffrey
who
actually
be
in
Shenzhen
with
something
that
sounds
like
cold
or
flu.
Symptoms
is
currently
at
the
doctor
being
checked,
and
my
heart
goes
to
him
because
it's
probably
extremely
stressful.
A
A
There's
been
a
lot
of
stuff
happening
in
our
little.
It's
very
funny,
because
coin
in
French
is
corner,
so
I
think
in
order
little
corner
of
the
world
is
lots
of
stuff
happening.
There's
a
lot
of
good
work
being
done
and
I.
Think
Jeffrey
even
I
wanted
to
have
this
interim
to
make
some
statements
about
what
is
happening.
We
are
still
very
much
I
would
say
still
in
a
startup
mode.
A
Although
the
hackathon
is
not
the
big
point
of
this
group,
it
still
I
think
a
fantastic
way
of
getting
together
with
the
other
groups.
It's
probably
the
only
time
during
I
eat,
EFI
or
TF,
where
both
the
research
groups
and
the
engineering
groups
can
meet
and
share.
So
I
think
the
participation
to
the
hackathon
is
as
much
for
a
face-to-face
social
interaction
as
much
as
doing
great
code.
So
you
can
go
through
the
next
slides,
which
is
all
the.
A
A
Oh
yeah,
the
gold
whoops-
oh,
that's
it,
okay
go
yeah,
we
can
just
revise.
This
is
actually
yeah
the
the
previous
ones.
You
know
cause
it's
still
looking
at
everything,
that's
happening
in
the
the
computing
and
the
network
and,
as
I
said
for
me
anyway,
for
the
in
the
past
three
months,
there's
been
so
much
activity,
there's
been
so
much
work
being
done.
A
That
is
related
and
while
I
think
that
when
we
started
this,
this
group
there
was
a
strong
link
on
you
know:
what
are
we
going
to
do
with
the
data
centers
and
the
and
the
edge?
Is
it
going
to
be
both?
Is
it
going
to
be
in
between?
Is
it
gonna
be
just
switches
for
filtering
in
the
p4
world,
but
I
think
more
and
more
than
we're?
Looking
into
this,
there
is
an
extreme
move
right
now
to
distribute
the
networking
with
computing
capability
all
around
the
network.
A
So
this
is
something
that
is
impacting
them,
but
in
the
positive
in
a
positive
way,
there's
an
awful
lot
of
activities
and
I
think
one
reason
we
call
it.
You
know
we
call
it
ourselves
where
people
are
you
on
the
list.
We
call
ourselves.
The
gems,
which
is
Jeffery
even
Murray,
shows
a
and
the
reason
the
gems
wanted
to
have.
This
interim
was
also
to
to
put
a
status
on
what
is
happening.
Okay,
next
exit,
so
I
want
to
have
very
briefly
or
we
want
to
have
very
briefly
a
review
of
the
milestones.
A
I
just
kept
the
ones
that
we
have
not
achieved
the
ones
for
this
year,
I
think
for
Vancouver,
something
that
we
could
do
is
to
start
thinking
of
new
milestones
again
because
of
the
incredible
amount
of
work
that
is
being
done
in
the
research
community
on
things
that
relate
to
what
we
do.
I
want
to
do
a
trough
update.
A
We
have
a
number
of
active
of
drafts,
we
have
a
few
directs
Pires
and
at
least
one
that,
although
it's
expired
because
of
somebody
getting
a
job
elsewhere,
I
don't
know,
what's
going
to
happen
to
it,
and
we
have
one
new
draft
that
actually
emerged
from
some
work
on
Io
T
at
the
European
Union.
We
want
to
do
some
a
review
of
the
potential
new
work
and
again
what's
happening
in
meetings
and
conferences.
A
The
vancouver
meeting
we
have
three
milestones,
which
is
identify,
coin
related
ecosystem
dependencies,
and
this
is
actually
looking
at
some
of
the
use
cases.
We
had
a
number
of
drafts
that
were
for
that
to
are
expired,
there's
a
new
one,
so
I
think
we're
progressing
on
this,
but
obviously
there's
more
work
needed
the
case.
Studies
again
is
that
it's
really
to
the
previous
one
and
the
one
that
I
think
that
we
need
to
really
focus
on,
maybe,
as
and
the
Eve's
you
can.
A
This
is
actually
also
part
of,
if
you're,
a
list
of
of
reduce
that
to
suggested
so
interrupt
me
anytime.
We
want
to
discuss
and
catalogue
some
of
the
coin
requirements
and
the
implications
for
the
network
elements
like
I
said
in
my
little
intro
we
had
at
the
beginning.
You
know
we're
talking
almost
a
year
and
a
half
ago,
which
seems
seems
to
be
like
the
time
of
the
dinosaurs.
We
were
mostly
looking
at.
A
Yeah
I
think
I,
don't
know
who
else
is
from
California
that
you
know
we
have
to
have
kudos
to
e4
for
supporting
this
at
6:30
in
the
morning
so
and
I
think
it's
not
for
today's
meeting
to
to
decide,
but
maybe
also
regarding
the
new
work
and
regarding
the
evolution.
Maybe
we
want
to
start
thinking
of
new
milestones,
because
the
other
one
that
we
have
is
is
a
very
generic
one.
A
That
I
think
we
had
put
last
year
when
we
started
the
group
and
I
think
we
may
want
to
have
now
that
we're
legit
to
have
milestones
further
down
in
to
a
2021
and
further
out
and
again.
This
is
not
for
today,
unless
someone
has
an
ID
for
a
milestone
that
we
should
add
today,
but
I
think
this
is
something
that
we
should
have
on
our
on
our
list
of
things
to
think
about,
and
things
to
do
and
again
look
at
the
evolution
of
the
work
from
currently
defined.
A
The
drafts,
as
we
as
we
said
before
in
this
group,
is
that
the
drafts
are
not
be
on
the
output
of
this
of
this
research
group.
We
also
want
to
encourage
presentation
from
the
research
community,
even
if
it
is
you
know,
conference
papers
and
stuff
like
that
to
make
sure
that
we
stay
really
up-to-date
with
what's
happening
outside
outside
our
group
and
be
able
to
better
link
to
the
other
conferences
and
other
meetings
that
are
related
to
us.
So
I
think
this
is
an
action
item
for
everybody.
A
C
C
That
I
would
what
I'm
hoping
is
that
don't
be
a
more
substantial
change
before
the
next
meeting.
You
know.
One
of
the
questions
that
Dave
had
in
his
very
thorough
review
of
the
document
early
on
was
why
focus
on
edge
data
discovery?
Is
there
something
more
comprehensive
from
a
lifecycle
standpoint
that
we
could
you
know,
maybe
that
that
should
be
the
focus
of
the
drafts
way?
That
is
that's
the
key
point
for
discussion
with
the
with
my
co-authors
is:
do
we
make
that?
Do
we
make
that
shift
or
not?
So
that's
under
review.
D
Yeah
actually
we're
thinking
about
extending
that
draught
or
maybe
submitting
a
new
one,
because
we're
now
thinking
about
adding
in
some
aspects
about
intrusion,
detection
and
so
we've
got
a
new
colleague
working
here
at
our
chair
and
yeah
we'll.
We
are
still
in
discussion
whether
we
want
to
simply
extend
a
security
paragraph
of
that
draft
or
whether
we
want
to
do
an
additional
draft
about
that.
A
We
do
not
have
of
anybody
from
the
author's
list,
but
I
understand
that
this
is,
you
know,
progressing
and
again
it
helped
a
lot
focusing
some
of
the
ideas
that
we
have
and
so
I
expect
this
to
be
continuing,
and
you
know
we
one
thing
that
I
think
in
in
Vancouver
that
we
think
that
we're
going
to
start
having
is
maybe,
except
at
least
one
of
these
drafts
as
research
group
drafts,
and
this
one
is
high
on
the
list.
So
we
will
have.
A
Simply
expired
of
there's
one
of
mine
and
working
with
Alex
Alexeev
and
Ischia
shying
to
make
a
new
version,
so
this
should
be
reinstated.
The
next
one
was
Jeffrey's,
I
I
think
he
will
not
continue
this,
but
since
he's
not
on
the
call,
I
cannot
talk
to
him,
but
I
had
understood
that
they
were
not
going
to
continue
that
and
the
last
one
is
a
little
bit
like
a
in.
A
Like
in
a
limbo
one
derp
trough
and
who
is
actually
one
of
the
main
author
on
this-
had
left
interdigital
and
is
not
working
for
who
I
weigh
so
I,
don't
know
if
and
we
will
communicate
with
started
I
think
last
week
or
something
like
that.
I
want
to
communicate
with
him
and
the
other
authors
to
see
what
they're
going
to
do,
what
they
want
to
do
with
this.
If
they
want
to
continue
it
or
discontinue
it
or
change
the
authors-
and
this
is
actually
something
that
we
need
to
review,
there's
a
new
one.
A
That's
going
to
be
presented,
actually
there's
two
new
ones
there's.
Actually
we
have
one
of
the
authors
of
the
first
one
Alessandra
bossy
on
a
reference
architecture
for
agricultural
4-0.
He
Alexandro
had
made
a
presentation
in
in
Singapore
and
now
we're
I'm
actually
a
co-author
also
on
this,
but
it's
mainly
his
work.
Where
he's
doing
actually
the
basic
agriculture
Alessandra.
Do
you
want
to
say
something
about
it.
A
A
A
This
horizontal
horizontal
ization
of
the
computing
in
the
network
in
two
layers
and
this
common
data
layer
is
interesting
because
it
allows
to
address
things
like
data,
formatting
data
structures,
advanced
distribution,
cognitive
networking
and,
of
course,
security,
which
was,
if
some
of
you
remember
one
of
the
major
comments
that
we
got
from
our
IRB
review.
So
it
I
put
it
as
to
come
because
I'm
actually
right
now
in
the
process
of
recruiting
authors
for
this,
because
there
was
a
lot
of
really
good
efforts
made
in
the
in
the
proposal
and
independently
of
this.
A
If
this
is
accepted
or
not
actually
distilling
the
common
data
layer.
Out
of
that
is
actually
I
think
a
very
good
idea
and
can
show
that
rely.
It
actually
makes
all
the
silos
that
are
currently
part
of
IOT
and
a
lot
of
communication
system
being
connected
together
and
connected
together,
very
very
close
to
where
the
use
case
or
the
application
is
being
activated
or
consumed.
So
this
is
to
come.
C
And
I
think
Alicia
say
that
this
is
exactly
along
the
lines
of
the
comment
that
came
back
around
the
edge
stated
discovery
proposal,
because,
if
you
think
about
it,
you
know
we're
focused
on
finding
the
data,
but
that's
one
only
one
part
of
you
know
finding
data
so
that
you
can
marshal
it
for
two
ready
it
for
a
computation
that
may
be
placed
somewhere.
And
how
do
you
do
that?
Coordination,
for
example?
C
That
was
the
thinking
behind
the
edge
data
discovery
draft,
but
then
there's
the
equally
important
okay,
you
have
this
computation
and
what
happens
to
it?
What
happens
to
the
output
of
that?
How
to
you
name
it?
How
do
you
maybe
cash
it
in
multiple
places?
How
do
you
you
know?
How
do
you
decide
what's
the
best
rent
transcoding
for
it
or
transformation
of
it
in
order
for
it
to
be
most
accessible,
etc?
C
A
And
actually
part
of
what
the
questions
were
like
yeah,
how
do
you
discover
the
coordinate
of
yours
rate?
How
do
you
make
sure
that
it's
secure,
you
know,
there's
a
bunch
actually,
if
I
have
the
the
proposal
we
did
and
the
discussions
we've
had
with
with
Ericsson
and
with
Dirk
is
like,
we
have
a
whole
bunch
of
questions
and
the
discovery.
Obviously
so
yes,
and
so
I
think
this
is
interesting,
because
also-
and
you
know
we
have
on
the
call
and
again
Alessandro
is
also
there.
A
These
are
basically
discussions
that
we're
raised
by
this
idea
of
the
use
cases
in
manufacturing
and
agriculture
and
transportation
where
there
is
a
functional
distribution
between
and
cloud
cloud
infrastructure
and
edge
infrastructure
and
a
bit
of
in-between,
and
it
goes
into
distribution
of
cognition,
also
in
all
kinds
of
stuff.
So
I
think
this
could
be
maybe
at
one
point
even
more
than
one
document,
but
let's
see
what
we
have,
and
it
has
raised
a
lot
of
interest
anyway.
A
So
I
think
it's
there
there's
a
bunch
of
people
on
the
call
that
could
have
other
ideas
and
again
these
are
the
ones
that
exist
or
some
of
the
ideas
that
are
there,
but
there's
other
there's
other
work
being
done
and
related
for.
Actually,
the
common
data
layer
is
also
very
much
related
to
things
that
are
happening
and
things
to
things,
for
example,
and
even
in
a
name
function
networking.
A
So
this
you
know
there's
a
lot
of
common
interests
there
and
if
anyone
on
the
call
or
in
your
research
groups
have
other
related
ideas
that
you
think
could
be
interesting,
please
submit
them
again.
I'm
going
to
talk
about.
You
know,
what's
going
to
happen
in
Vancouver,
it
doesn't
have
to
be
a
draft
if
you
just
want
to
have
a
presentation.
One
thing
that
we
have
started
looking
at
and
I
see
Frank
is
on.
A
The
call
is
this
idea
of
critical
information
inside
computing
in
the
network,
so
by
computing
we
may
want
to
identify
critical
information
and
that
information
may
need
to
be
transported
with
some
kind
of
FEC
or
some
kind
of
coding.
This
is
something
that
we
started
talking
to
Jeff
with
my
other
group,
actually,
the
network
coding
group
with
the
people
from
the
University
of
Louvain,
and
that
could
also
lead
to
more
things
in
the
common
data
layer.
So
the
criticality
of
information
is
I,
think
also
something
that
is
very
important
for
us.
A
A
It's
going
to
be
a
short
meeting:
okay
next
slide,
okay,
so
this
is
related
to
which,
as
I
said,
so
potential
new
work.
We
have
obviously
a
use
case
for
dye.
Ut
I
talked
about
common
dead,
the
common
data
layer,
there's
some
ideas
and
distributed
cognition.
This
is
actually
from
erickson
research
and
both
in
Erickson
Canada
supposed
to
talk
to
them
feed
Canadian
port
tomorrow,
but
it
is
something
that
is
strongly
related
to
the
some
of
the
XR
issues.
A
Also,
so
X
are
being
extended
reality,
so
this
is
something
that
could
lead
to
more
work.
Roberto
Morabito,
who
is
part
of
that
Ericsson
research
team
is
currently
in
Princeton,
so
he's
talking
to
all
the
right
computing
and
the
network
people,
and
that
could
be
also
something
that's
going
to
happen.
Tactile
internet
I
am
not
seeing
fat
and
shining
from
equal.
The
technology
super
here
in
Montreal
I'm,
actually
in
Montreal
right
now,
waiting
for
a
snowstorm
and
all
flights
are
canceled
tomorrow
and
so
tactile,
Internet
and
I
would
say.
A
This
is
my
list
and
actually
one
thing
that
I
would
like
to
submit
to
the
to
the
to
the
list
and
to
this
group
currently
today.
So
raise
your
hand
if
you
agree,
I
think
it
would
be
good
to
have
some
kind
of
a
you
know,
part
of
the
wiki,
but
maybe
part
of
the
github,
a
list
of
conferences
that
we
know
are
related
to
this.
A
A
A
So
I
mentioned
that
proc
that
practicing
about
now
three
weeks,
it's
in
Paris,
it's
part
of
something
that's
called
I,
see
I
M,
which
is
a
you
know,
conference
on
computing,
where
there's
going
to
be
also
a
panel
at
the
end
on
things
related
to
computing,
a
network
that
proc
right
now
is
very
small.
We
have
a
number
of
papers,
but
we
also
have
AP
for
hackathons
last
tutorial.
It
was
the
first
year
was
organized
and
I.
A
Think
it's
going
to
be
going
to
be
nice
I,
like
small
conferences,
there's
a
doctoral
seminar
in
March
and
Dave
and
Eve
and
I,
and
probably
a
few
other
people
on
this
list
and
almost
Collins
going.
But
your
walk.
You
know
I
think
it
would
be
nice
and
a
bunch
of
people
who
are
part
of
what
I
call
the
current.
You
know
team
of
drafts,
authors,
they're
crossing
your
guts
and
everybody
are
going
to
be
there.
It's
actually
on
compute,
first
networking,
which
is
one
way
of
looking
at
computing
in
a
network.
A
These
seminars
are
cool
because
it
allows
people,
ok,
so
Colin,
you
won't
be
there,
but
we
will
take
good
notes
and
those
seminars
are
really
cool
because
they
allow
people
to
really
get
together
and
be
immersed
for
a
few
days
into
a
specific
topic
and
they
publish
them.
You
know
nice
notes
and
stuff
and
I.
Think
Eve
is
one
of
the
organizers
actually
with
Dirk
and
I.
Think
we
will
make
sure
that
salient
outputs
of
this
are
going
to
be
published
to
the
list.
A
Well,
essentially,
it's
the
users
of
the
application
or
of
the
system
and
the
interest
data
messaging
in
ICN
and
all
the
namespaces
that
were
defined
or
really
really
good.
For
that.
So
we
want
to
extend
a
little
bit
the
focus
of
of
ICM
2020
towards
these
applications
again.
The
other
is
for
for
us
to
fill,
and
we
will
create
this
this
list
and
make
sure
that
the
old
community
is
is
aware.
A
If
you're
going
to
talk
very
soon
so
I'm
my
end
of
things,
we
will
not
have
real
presentations
today,
because
we
first
we
didn't
get
some
and
also
I,
think
it
was
more.
The
goal
of
this
meeting
was
more
to
to
put
a
status
for
everybody.
He
didn't
and
to
have
specific
presentation.
So
next
slide
now.
Is
she
only
great.
C
A
The
goal
is
obviously
looking
at
the
architecture
revolution
and
some
tools,
development,
but
the
forked
sector
evolution
essentially
also
it's
interesting
to
think
about
it
when
we're
inside
the
hackathon,
because,
like
I
said
at
the
beginning,
it
is
one
of
the
few
occasions
inside
the
IETF,
where
all
the
groups
can
meet-
and
you
know,
there's
cross
table
virtualization
here,
which
is
kind
of
interesting.
So
I
will
let
Eve
talk
about
her
multi
stream,
processing
project,
okay,.
B
A
C
A
C
A
And
logistics
well,
for
now
we
know,
please
send
us
agenda
items
and
your
talks
that
you
would
like
to
do
and
the
column
actually
sent
a
message
on
off
everywhere,
but
there
were
day
passes
available
and
if
there's
still
some,
but
if
you
feel
that
you
know
somebody
could
would
not
come
if
they
have
to
pay
the
whole
thing
or
people
from
the
West
Coast
Vancouver
so
close
by.
So
you
know,
students
or
things
like
that
or
young
researchers
that
sometimes
don't
have
budgets
that
could
be
a
way
to
get
more
people
I.
F
C
A
B
A
B
C
It
needs
it's
very
likely
that
you're
going
to
need
to
scrutinize
the
streams
that
converge
at
various
collection
points
to
assess
whether
it's
going
to
fit
all
of
that
data
is
going
to
fit
on
the
next
hop
going
forward,
and
not
only
is
it
going
to
fit
on
the
link,
but
it
is
benefit
on
the
computer,
either
in
memory
or
in
storage
or
on
it.
You
know
it's
internal
bus,
whatever
whatever
the
case
so
computing.
The
network
has
this.
C
You
know
this
role
for
these
upstream
flows
to
you
know,
assess
the
resource
availability
and
it
turns
out
that
if
you
look
at
a
lot
of
the
immersive
media
work
going
on
or
even
the
dense
IOT
work
happening,
you
know,
we've
got
so
many
dense
deployments
of
devices,
whether
those
devices
or
cameras
or
sensors,
that
you
know
in
the
aggregate
when
they
collect
at
points
they're.
Just
there
isn't
enough
room
there.
C
So
what
are
the
kinds
of
functions
that
need
to
live
at
those
aggregation
points
to
do
interesting
things
now,
while
I
sort
of
introduced
this
case?
If
these
upward
flows-
and
you
know
that's
made
for
compute
at
those
at
those
nodes
in
the
in
the
very
first
Boff
session
at
a
later
session-
I
actually
came
in
and
talked
about.
You
know
I
guess.
C
F
C
Is
the
idea
that,
in
these
dense
deployment
there
are
so
many
witnesses
to
any
event
so
I
as
a
sensor
may
sense
something
anomalous
happening
and
I
and
I
believe
I've
used?
The
use
case
of
you
know
in
a
smart
city,
there's
a
camera
and
something
important
happens
like
a
crash
in
the
middle
of
an
intersection
which
is
actually
where
the
bulk
of
most
accidents
happen
or
very
large
percentage
of
them,
and
you
know
so.
C
It
senses
that,
but
by
the
way,
so
do
all
the
other
sensors
nearby,
whether
they're
cameras
or
vibration,
sensors,
sensors
or
audio
sensors
and
they're
all
sending
information
to
their
anomaly
detectors.
And
so
the
idea
was
for
this
ubiquitous
use
case.
What
are
the
kinds
of
functions
that
might
that
you
might
perform
at
that
collection
point?
So
what
I
have
here
is
somebody's
rebranded,
Tofino
switch
at
the
bottom
and
it
presumably
has
match
action.
I,
don't
know.
If
you
can
see
my
cursor,
you
see
my
cursor
moving
around
when
I
move.
It.
C
I,
do
okay
great,
so
you
know
it.
Of
course
the
programmability
is
all
about
these
match
action
tables,
at
least
from
a
programmable
switch
standpoint
or
a
p4
standpoint
and
I
I
started
a
thread
I
think
right
before
the
last
or
maybe
right
after
the
left
meeting
the
left
IETF
around
one
of
the
kinds
of
things
that
your
switch
do.
What
could
it
do
to
not
only
assess
whether
a
single
flow
matches
some
rule,
but
could
we
have
groups
of
streams?
C
You
have
functions
that
are
very
deliberately
looking
at
in
this
case,
I've
just
for
examples
sake,
a
B
and
C
and
I
should
probably
show
this
more
visually
on
this
picture,
but
a
B
and
C
are
streams,
their
flows,
their
data
flows
coming
through
the
switch
and
I
really
want
my
my
match
function
to
be
about.
Can
you
tell
me
if
these
streams
are
too
textured
related?
And
that
begs
the
question?
What?
C
C
From
the
perspective
of
the
switch,
because
this
is
of
course
going
on
in
this
switch-
and
this
is
a
function,
this
is
I
mean
and
that's
that's
a
great
question.
It's
these
are
mass
action
functions
that
need
to
be
installed,
and
so
I
guess.
The
answer
to
the
me
question
is
who
have
installed
the
master
function
so,
but
if
you
know
it
may
be
an
operator
because
what
I'd,
what
I'd
like
to
have
happen,
is
I'd
like
to
provide
a
service.
C
Information
I
have
this
opportunity
to
say
only
forward
the
information
from
the
stream
that
has-
and
this
is
what
I
proposed
on
the
list
you
know
if
I
could
ascertain
that
these
these
streams
were
related,
meaning
they're.
Sending
me
related
information,
you
can.
You
could
define
contextual
related
in
any
number
number
of
ways,
but
in
this
use
cases,
how
proximate
are
they
to
each
other?
Are
those
cameras
pointing
in
the
same
place
and
have
they
captured?
Have
they
actually
captured?
C
You
know
the
jaywalker
who
was
walking
across
the
street
that
led
to
this
accident
in
the
middle
of
the
intersection.
You
know
something
to
that
effect
and
so
I'm
thinking
about
all
the
different
things
that
once
I've
ascertained
it
that
these
streams
related
might
be
useful
functions
to
match
on.
So
these
were
just
you
know,
top
of
mind.
C
C
These
are
on
streams
and
the
the
switch
is
needing
to
look
a
very
likely,
although
not
necessarily
very
likely,
I
want
it.
Certainly
in
the
second
example
to
look
into
the
packet
beyond
the
packet
header.
So
right
now,
it's
often
the
case
that
what's
happening
inside
of
the
switch,
is
that
it's
got
acceleration
hardware
for
doing
all
the
things
that
are
typically
done
to
packet
headers
as
they
come
in,
but
really
a
lot
of
the
most
interesting.
Well
sorry,
my
perspective
is
there's
additional
interesting
work
to
do
and
it
could
be
defined
as
computing.
C
The
network
that
happens
beyond
the
packet
header
and
so
the
not
only
are
we
interested
in
groups
of
contextually
related
streams,
but
we're
looking
at
wanting
to
process
stuff
beyond
the
packet
header.
You
know,
for
you
know,
a
very
human
in
the
frame
is
an
anomaly
detector.
You
know
all
of
these
things
could
be
about
information.
C
That's
in
the
deep
packet
inspection,
so
you
know
I,
don't
know
how
easy
or
hard
that
is
to
do
because
I
don't
really
have
my
p4
chops
yet,
but
but
that's
the
thinking
and
then
the
actions
might
be
something
as
simple
as
you
know.
Oh
these
things
you
know,
I
want
to
process
them
together
and
I'm
going
to
do
more,
sophisticated
things
on
them.
Let's
you
know
spin
up
a
VM
and
do
something
more
sophisticated
with
them
and
we're
using
the
switch
to
kind
of
identify
collections
of
things.
C
If
you
squint
at
that,
that's
a
that
may
look
like
I'm
forming
this
virtual
edge
cloud.
Those
those
strains
qualify
as
some
Federation
in
its
own
right
and,
let's
you
know,
do
something
edge
like
that.
Manages,
for
example,
I'm
going
to
have
established
some
policies
that
are
related
to
those
streets,
or
you
know
I'm
going
to
create
I'm
going
to
identify
where
there's
a
trust
anchor
for
them
or
other
forms
of
security
and
privacy,
for
example.
It
may
be
the
case
that
oh
I've
got
these
contextual
related
stream.
C
Let's
turn
on
that
anomaly,
detector,
and
so
that
I
can
come
back
and
do
some
anomaly
detection,
oh
and
then
this
anomaly
detection.
Let's
turn
on
the
you
know
human
in
the
frame,
finer-grained,
anomaly,
detector,
or
something
like
that.
So
you
can
see
there's
this
whoops.
That
was
not
what
I
want
to
do,
and
so
there's
a
whole
collection
of
these
functions
that
are
circular
or
iterative.
You
know
one
leads
to
the
other,
which
comes
back
to
a
different
kind
of
match.
C
Match
match,
leads
to
an
action
which
leads
to
a
match
to
which
leads
to
action
and
so
forth
so
forth,
but
I'm
also
quite
interested
in
you
know.
If
something
happens
that
it's
anomalous,
you
can
imagine
in
an
industrial
setting
that
you
might
want
to
vault
that
data,
or
even
in
a
setting
where
there
is
a
business
model
around
oh
I,
did
an
insurance.
C
Somebody
needs
to
you
know
or
there's
a
contract
for
a
guaranteed
quality
of
service
that
isn't
met.
I
want
to
vault
the
data
so
that
I've
got
evidence.
My
ubiquitous
witness
evidence
in
the
future
of
some
collection
of
data
around
a
point
in
time
and
space,
and
then
you
know
back
to
the
immersive
use
case.
C
The
immersive
media
use
case
and
I
have
in
the
past
also
commented
on
the
very
interesting
work
going
on
in
the
MPEG
standards
community
around
stitching
together
these
disparate
streams
of
flows
to
form
something
new,
which
is,
you
know,
360
degrees,
they're,
they're,
whole
collections
of
formats
and
points
cloud
formats
that
are
emerging
from
the
MPEG
guide
group.
But
the
idea
is
I.
Have
all
these
streams
they're
related
if
they're
in
the
visual
realm
and
I
could
stitch
them
together
or
even
to
view
partial
360
degree
views?
C
You
know,
maybe
it's
180
degree
view
that
would
still
allow
me
to
walk
around
my
data
after
the
fact
and
examine
what
actually
happened,
and
so
you
can
see
that
there's
this
there's
there's
no
shortage
of
thing,
so
we
could
do
for
this
kind
of
use
case.
The
question
is:
is
the
switch
up
to
it
and
what
would
happen?
Oh,
do
we
or
is
that
if
it's
not
the
switch
itself
is
there?
G
I
have
two
quick
observations.
One
is
some
things
are
convenient
to
do
with
the
data
flow
model,
which
is
what
you're
basically
forced
to
do
in
a
switch,
and
some
things
are
not
and
I.
Think
like
a
first
order.
Thing
to
try
to
do
is
to
categorize
those
things,
because
we
have
50
years
of
history
of
what
things
work
well
in
data
flow
models
in
which
things
don't
just
from
a
computational
standpoint.
So
maybe
going
back
and
looking
at
that
kind
of
stuff
would
be
useful
all
right.
G
A
second
thing,
which
is
happen,
which
happens
at
much
higher
rate
data
rates
than
what
we're
talking
about,
is
what
the
physicists
have
to
do
for
detector
data
in
particle
accelerators,
Google
it
they
have
to
combine
and
throw
away
more
than
99.99%
of
the
sub
crime
and
figuring
out
how
to
do
that.
Without
you
know,
spitting
special
purpose
hardware
is
actually
very,
very
tricky,
so
there's
active
learning
there
may
be
some
very.
There
may
be
some
learnings
there.
Yes,.
C
I
mean
that's
a
that's
a
really
great
point
because
make
no
mistake,
you
know,
I
I
show
a
picture
of
a
switch
here
because
we
have
at
least
in
the
past
few
hackathons
been
catalyzed
by
looking
at
p4,
and
so
that
visits
more
and
on
trip
to
the
discussion.
But
the
question
is
this:
is
not
just
a
software
flow
or
a
programmability
flow?
C
This
is
there
are
implications
for
the
hardware
which
are
super
interesting
for
the
for
the
design
of
any
kind
of
entity
that
lives
at
these
aggregation
points,
because
there
is
this
close
partnership
that
has
to
happen
between
the
router,
the
compute
engine
and
the
storage.
That
may
not
really
exist
as
yet
in
something
like
the
switch
that
I've
shown
here.
Yeah
I.
G
Agree
entirely
with
that,
what
portion
I
would
give
is
the
p4
is
way
more
powerful
than
any
switch
today
can
do
and
may
be
way
more
powerful
that
any
switch.
That
looks
anything
like
today's,
which
is
good
ever
here
right.
So
maybe
thank
you.
May
you
really
need
to
start
with
the
hardware
and
not
the
programming
land
I
can
do
almost
anything
before
I'm
one
two
right.
C
C
C
Well,
I'm
I'm
happy
to
again
because
this
sort
of
came
up
in
the
context
of
the
hackathon.
Could
we
come
up
with
a
project
that
was
more
long-lived
that
maybe
you
know
for
this
kind
of
hackathon?
The
first
thing
that
we
look
at
is:
how
easy
is
it
to
simply
do
the
base
function,
which
is,
could
you
could
the
switch
at
line?
Speed,
tell
me
which
things
are
contextually
related,
and
could
we
come
up
with
a
family
of
functions
that
do
progressively
more
sophisticated.
C
You
know
functions
around
what
does
context
mean?
You
know
it
could
just
be
an
identifier
inside
of
the
packet
header.
Well,
that's
simple.
The
answer
is
yes,
but
you
know
these
were
sophisticated
ones.
The
answer
is
probably
not
now.
The
I
took
another
to
drill
in
a
little
further
into
the
use
case
and
again,
I
can
kind
of
present
this
in
more
detail
in
a
you
know,
when
we,
when
we
meet
in
person
that
one
of
the
interesting
things
about
the
MPEG
I
standard
is
that
it
is,
it's
got
a
metadata
description
for
scenes.
C
C
You
come
up
with
one
data
model
for
like
how
do
you
represent
data?
How
do
you
represent
objects?
And
so
now
you've
got
MPEG
I.
It
understand.
It's
got
scene
intelligence.
It
can
recognize.
If
we
throw
some,
you
know
we
can
create
bounding
boxes
around
things
and
identify
them
as
objects.
Inside
of
these
scenes
and
each
one
of
those
objects
in
turn,
has
you
know
its
own
meta
information,
and
you
know
it
starts
to
look
a
bit
like
the
naming,
that's
required
for
ICN
or
some
other
complex
scene
where
data
is
related.
C
Now
not
everything
in
the
scene
is
going
to
be
going
to
have.
A
digital
is
not
going
to
have
computation
and
communication
capabilities,
but
some
of
them
are-
and
it
just
becomes
a
very
interesting
way-
to
link
the
visual
realm
to
the
smart
objects,
realm
and
kind
of
working
backwards
to
figure
out
how
do
these
digital
cyber
twins?
C
You
know
that
have
a
representation
on
the
computer,
then
map
back
to
the
physical
world.
So
there
are
lots
of
things
that
can
come
out
of
this.
You
know
it's
sort
of
endless,
but
in
that
case,
I
was
thinking
that
it
could
have
some
nice
implications
for
linkages
back
to
where
the
coin
discussion
even
started,
which
is
the
thing
to
thing
research
groups
so
so
kind
of
to
comfortable
circle.
Is
this
you
know
sort
of
how
do
people
think
about.
C
At
least
one
of
the
hackathon
projects
around
a
use
case
like
this,
whether
or
not
we
tie
it
to
a
physical
switch
or
whether
you
know
all
of
this
lives
and
software
of
some
sort,
and
we
map
it
back
to
these.
You
know
what
Dave
was
saying.
Is
you
know
what
are
these
the
kinds
of
functions
for
contextually
related
data
flows
that
are
easy
for
certain
kinds
of
hard
classes
of
hardware
to
do
versus
these
newer
functions
that
we're
going
to
need
something
new
for
so
we
could.
We
absolutely
could
go
through
that
exercise.
C
A
A
It's
like
big
data,
essentially
and
99%,
or
even
more
than
that,
it's
just
data
that
could
be
sent
somewhere
else
and
used
to
train
neural
network
algorithms.
But
they're
not.
You
know,
I
mentioned
the
word
critical
and
they
are
not
critical
at
this
point.
They're
part
of
yes,
the
training,
so
this
is
what
the
dice
plant
looks
like
and
oh,
my
god
now
there
is
something
that
needs
action.
It
still
needs
to
be
part
of
the
training,
but
it
also
needs
to
be
action
Don
right
now,
so
it's
the
anomaly
detector.
A
It
needs
to
trigger
something
so
I
think
yes,
the
implication
in
MPEG
is
interesting,
but
I
think
the
more
you
know
the
even
more
important
one
is
this
in
network
compute
across
contextually,
related
data
flow
and
I
think
based
on
what
was
done
in
the
past.
Two
hackathons
is
something
that
is
a
step
further
Amil
or
alessandro.
You
were
there
in
Singapore.
Do
you
have
anything
to
add
to
this.
E
Good
no,
but
what
you
said
it's
a
very
true
I
mean
related
to
ddd.
Dates
have
been
coming
from
different
fields:
I
mean
agriculture
is
one
of
the
example,
but
industry
can
be,
can
be
a
similar
one,
transportation
or
or
whatever
I
mean
smart
city
and
so
on
and
so
forth
and
Indian
Amanda
would
be
very,
very
interesting
for
me.
G
So
this
is
there's
another
angle
or
that
I
think
we
want
to
be
aware
of.
Nick
McEwan
gave
this
fascinating
talk
and
hotness
info
in
November,
and
the
bottom
line
of
that
talk
is
that
servers
are
starting
to
have
the
same
problem
as
switches,
because
if
you
need
to
touch
the
data,
everything
falls
apart,
very
quickly
rates
above
40
gigabits,
so
he's
actually
proposing
new
server
architectures.
G
So
I
think
we're
going
to
see
a
lot
of
this
recapitulated
+
+,
+,
2,
2
things
we
really
ought
to
think
about
are
number
one.
The
switch
guys
really
don't
care
about
anything,
but
the
highest
rate
data
right
all
of
their
energy
goes
into
it
all
the
money
goes
into
it
all
of
the
ASIC
design
goes
into
it
and
we
had
tons
of
use
cases
that
are
two
orders
of
magnitude
below
what
those
guys
are
looking
at.
G
G
A
And
and
I
think
that's
exactly
what
I
said
at
the
beginning.
What
I
said
this
is
them
an
incredible
evolution
of
the
thinking
of
this
group
from
the
big
you
know
from
when
we
started
thinking
about
it,
and
you
know
I
would
say
well.
Poor
Jeff
is
not
here,
but
Jeff
had
more
of
a
data
center
and
hundreds
of
gigabits
per
second
whatever,
and
even
I
were
more
like
hold
on
that.
A
G
A
So,
but
what
I'm
saying
is
that
I
think
yeah
I
think
all
these
comments
are
good,
but
maybe,
if
you
know
I,
think
Eve,
you
should
post
this
slide
to
the
group
and
with
a
little
bit
of
your
ideas
and
cuz
how
we
could
we
could
do
that
and,
like
I
said,
there's
already
code
from
the
previous
packet
on
that.
Does
some
of
that
and
also
start
a
a
discussion
on?
A
Yes,
what
it
means
to
to
look
at
these
data
flows,
what
it
means
when
you
start
looking
again
just
add
the
headers
or
I
would
say
a
little
bit
beyond
the
header.
I
would
say
the
header
and
the
metadata,
and
not
just
you,
know
the
whole.
We
don't
maybe
need
the
whole
payload
to
make
a
decision
right.
C
And
actually,
what
I
didn't
get
to
was
that
it
turns
out
not
surprised
the
way
that
the
MPEG
I
standard
has
creates
voila
metadata
identifier,
so
object
identifier,
and
so
in
the
same
way
that
we
may
have
a
port
number
or
a
source
address
or
whatever
it's
got
metadata
identifier,
and
maybe
what
we
could
be
doing
is
simply
jumping
into
the
packet
and
grabbing
that
metadata
identifier
to
go.
Aha,
you're.
Looking
at
that
thing
to
you
know
and
and
okay,
great
we're,
contextual
e
related.
G
C
No,
no,
that's
that's
what
we're
doing,
but
but
reusing
the
identifiers.
Until
again,
this
has
wonderful
implications,
for
you
know
the
magic
of
main
spaces,
and
could
we
could
we
leverage
our
expertise
in
namespace
design
to
do
something
intelligent?
It
is
an
MPEG
world
or
propose
something
or
demonstrate
something.
Okay,.
C
C
But
so
this
is,
but
this
kind
of
goes
back
to
okay,
because
I'm
peg
I
is
in
flight,
just
as
the
one
data
models
work
is
in
flight
in
the
IOT
in
these
communities
which
seem
like
pretty
disparate
communities.
Can
we
get
some
people
like
ourselves
to
straddle
these
world
to
influence
one
the
metadata
model
to?
How
would
it
be
easy
to
create
to
extract
from
the
scenes
create
namespaces
that
are
reasonable
and
processable
and
you
know
etc,
and
that's
fit
into
these
models
of
networking
like
ICN
and
so
forth?
So.
G
You
know
another
sort
of
thing
right
out
of
my
rear
orifice.
Is
that
about
a
great
phrase
she
check
Helen?
Is
a
group
of
Wisconsin
came
up
with
a
scheme
for
fingerprinting
video
with
similarity
algorithms
that
actually
represented
similarity,
so
that,
if
you
computed
this
at
two
different
sources,
it
would
come
up
with
the
same
fingerprint
if
the
two
things
we
were
looking
at
were
sufficiently
similar.
That's.
G
Would
say:
I?
Don't
he
was?
He
was
the
P
I
I'm,
not
sure
he
actually
did
the
work.
We
should
go
through
some
forensics
on
video
fingerprinting,
because
that
sounds
more
like
what
you're
trying
to
do
in
terms
of
being
able
to
decide
what
to
do
with
some
video
based
on
how
similar
it
is
to
some
video
from
a
different
source.
Okay,.
G
But
I
think
that
I
think
there's
two
I
mean
it
would
be
extremes,
but
many
of
the
cases
in
IOT
are
actually
static.
Properties,
not
dynamic
properties
or
quasi
static
properties,
which
means
that
the
configuration
of
the
sensors
doesn't
change.
The
contextual
relationship
doesn't
change
and
not
true
for
video.
C
You're
right
that
they're
going
to
there's
going
to
be
some
classes
of
meta
information
that
never
changes
and
then,
when
we've
got
mobile
things
moving
around
or
dynamic
scenarios.
Yes,
that
part
of
the
meta
data
space
is
going
to
be.
You
know
very
hard
to
attain
because
you're
it's
going
to
require
you
to
do
some
computation
on
the
stream
before
you
can
decide
what
the
metadata
is.
You
know
that
you're
going
to
plug
into
this
function,
that's
going
to
deduce
contextual
relationships.
C
So,
yes,
so
I
think
that's
one
important
point
that
you
to
tease
out
of
this
is:
could
we
find
interesting
elements
inside
of
the
metadata
that,
after
some
initial
compute
or
even
from
the
Gecko,
are
things
that
we
could
hang
our
functions
on?
Meaning
you
know
that's
what
we
focus
on
to
deduce
contextualism
and
that
would
certainly
speed
things
up.
C
So
you
know
that's
sort
of
a
meta
problem,
for
this
is
how
do
we
create
this
family
of
functions
that
don't
require
us
to
perform
compute
every
time,
because
after
all,
it's
certainly
produced
style
of
switches
or
network
elements?
They're
not
really
designed
with
hardware?
That's
going
to
be
accelerated,
it'll
accelerate
those
sorts
of
functions,
at
least
not
yet
so
yes,
that
was.
That
was
a
good
point
that
you
made
Dave
about.
You
know
what
things
are
we
going
to
look
into
the
package
stream
and
grab?
C
How
easy
will
that
be,
and
that
should
be
the
focus
of
I
think
the
hackathon
is
just
to
do
something.
This
is
this
is
the
big
picture,
but
the
small
picture
would
be
great.
Let's
work
on
functions,
some
small
collection
of
functions
that
will
figure
out
whether
these
streams
are
contextually
related
that
are
flowing
through
the
switch.
Now
we
don't
really
have
a
testbed,
that's
going
to
allow
us
to
do
that,
so
we're
going
to
have
to
simulate
that.
F
A
I
think
would
be
interesting
and
starting
a
discussion
on
the
list,
because
the
people
who
were
really
influential
and
the
previous
hackathons
are
not
on
well
Alessandro
and
ml
were
there,
but
yet
people
from
nobody
below
the
people
from
the
University
and
Singapore
that
did
a
lot
of
fantastic
work,
the
guy
from
the
Netherlands
who
I
suppose
was
part
of
one
of
the
telecom
operators,
their
date
without
their
on
the
list.
So
they
could
also
have
specific
ideas
for
this
and
I
have
their
names
anyway.
So
we
could
even
pick
them
personally.
A
G
A
B
A
So
I
think
I've
I've
taken
a
number
of
action
item.
One
of
them
is
to
make
sure
that
we
share
all
these
ideas
for
the
hackathon
for
March
21
22.
As
far
as
the
active
new
and
expired
drafts,
I'm
going
to
make
sure
that
I
contact
the
authors
of
at
least
the
their
cross
and
one
to
make
sure
to
see
how
this
is
going
to
go
forward
or
if
it's
going
to
be
abandoned,
I'm
going
to
work
with
alessandro.
A
On
the
use
case
for
agriculture,
I
will
contact
the
erickson
research
people
as
well
as
the
people
from
intel
germany
about
this
common
data
layer
that
was
defined
more
or
less
both
in
discussions
for
the
I
you
proposal,
but
also
a
lot
of
it
on-site
meetings
that
we
had,
and
also
that
involved
dirk
good
dinner.
Actually,
it's
very
funny
to
have
two
Dirk's
and
the
same
group.
So
it's
dirty
and
dirt
k.
So
this
one
at
dirt
k
and
will
obviously
use
all
of
this
to
define
a
gentle
items
and
talks.
A
A
Vancouver
is
not
that
bad.
Actually,
it's
it's
I
think
it's
nine
hours,
not
twelve
like
if
we
were
on
the
East
Coast.
So
you
know
it's
feasible.
It's
like
you
ieave,
when
you
communicate
with
with
Europe,
so
I
think
there's
potential
for
for
participation
and
I'm,
even
thinking
that,
if
it
gets
to
this,
we
could
even
have
them
record
their
presentations
and
just
play
it
back.
Obviously,
it
means
that
all
questions
would
be
them
taken
down
and
sent
remote.
A
You
know
sent
at
a
later
time,
but
I
think
we
have
to
make
sure
that
we
allow
people
to
you
know
to
participate
as
much
as
they
can.
I
also
know
that
out
of
the
net
proc
we're
going
to
also
contact
a
few
people,
there's
you
know
ways
of
also
having
more
presentations.
As
you
know,
the
Cambridge
group
and
the
UK
is
extremely
active
on
all
kinds
of
related
things.
I
also.
C
B
A
Call
for
agenda
item
and
talks,
I
think
you
Eve
and
just
when
you
feel
better
should
start
working
on
the
agenda,
but
for
the
moment,
I
kind
of
call
it
like
a
call
for
agenda
items,
there's
nothing
perfect
plans
except
for
for
obviously
the
people
who
have
the
active.
The
active
drafts
obviously
should
be
given
the
chance
of
presenting,
but
the
rest
is
is
wide
open
right
now:
okay,
yeah
and
I
think
yeah.
We're
think
we're
going
to
have
still
a
lot
of
stuff
and
but
I
think
with
two
sessions
we
should
be.
G
G
C
D
G
A
A
You
know
my
own
sentiments
here.
This
is
going
really
well
and
you
know
it's
blossoming
and
it's
really
cool
and
thank
you,
everyone
for
being
there.
We
had
I
think
a
very
good
cross-section
of
the
skills
in
this
meeting,
although
it
was
small
and
I
think
we
need
to
continue
the
work.
So
thank
you
so
very
much
and
take
you
to
Eve
for
being
a
the
early
bird.
What
you
kept.