►
From YouTube: Kubernetes WG IoT Edge 20190925
Description
September 24 meeting of the Kubernetes IoT Edge Working Group
A
Okay,
I
just
started
the
recording.
This
is
the
meeting
of
the
kubernetes
iot
and
edge
working
group.
These
meetings
are
public
and
recorded
and
will
be
uploaded
and
available
on
youtube.
We
already
started
this
and
missed
a
little
bit.
We
have
no
agenda
today,
but
the
three
of
us
are
showing
up
are
just
having
a
conversation
on
trends
and
thoughts
on
what's
going
on
in
the
field
of
iot
and
edge
yeah,
so
go
back
and
continue
with
your
thoughts.
Moretz.
B
Yes,
as
I
said,
I
just
had
a
really
interesting
talk
yesterday
with
a
colleague
that
he
was
actually
like
managing
edge
devices
with
operators
like
like
kubernetes
operators
and
crds,
and
the
the
interesting
idea
you
came
up
with
is
that
basically
kubernetes
is
allowing
you
to
kind
of
build
your
system
in
a
model
view,
and
then
you
just
hand
this
model
over
to
your
to
your
cluster
and
your
cluster
kind
of
connects
your
model
to
the
real
world.
B
So
it
kind
of
creates
this
digital
representation
of
edge
device
or
of
the
state
that
you
want
to
run
your
your
your
edge
in
and
that's
actually
something
like
a
lot
of
different.
Other
techniques
like,
for
example,
uml
and
other
modeling
techniques,
are
trying
to
solve
like
connecting
this.
B
This
model
view
aspect
with
your
real
world,
but
it
often
diverges
because
I
I
don't
know
who
actually
uses
oml,
for
example,
but
it's
really
hard
to
like
keep
your
real
world
application
in
sync,
with
your
model
and
kubernetes
kind
of
bridges
that
you
know
what
I
mean.
A
C
C
C
Yeah
yeah
figure
out
the
gotchas
in
the
architecture
before
you
build,
but
but
then
steve
I've
seen
what
you're
talking
about
too,
and
it's
absolutely
true-
that's
the
first
thing
to
get
out
of
date
with
the
real
world
software.
It's
a
big
danger.
So
therefore
the
the
more
accurate
you
try
to
make
it
the
more
likely
it
is
to
break
down
yeah
yeah.
A
But
that
said,
I
think
the
real
interesting
thing
about
kubernetes,
with
its
architectural
model,
of
having
a
an
api
where
you
state
a
desired
goal
or
condition
and
then
have
these
control
loops
in
background
relentlessly
attempting
to
achieve
it
is
really
something
that's
been
used
in
the
iot
field
for
decades.
You
know
for
what
is
often
called
supervisory
control,
where
you've
had
something
like
a
a
tiering
of
low-level
controllers
things
you
know
in
the
u.s
they're
called
loop
controllers
or
temperature
controllers
plcs,
and
then
a
secondary
tier.
A
That
often
you
know
download,
set
points
to
these
things,
but
it's
designed
so
that
you
can
sever
communications
to
this
lower
tier
and
they
still
go
on
attempting
to
achieve
their
last
known
settings
of
desired
goals,
and
it's
something
that's
known
and
proven
to
scale
and
also
be
real.
You
know
resilient
in
the
face
of
failures
of
isolated
parts
of
the
system
and
failures
and
communication,
and
you
know
that
kubernetes
model
can
be
used
to
do
that
for
any
number
of
problems.
A
A
You
have
to
take
your
time
to
architect
decent
custom
resources,
but
then
the
actual
code
and
the
operators
could
simply
call
existing
legacy
projects
to
do
the
implementation,
or
you
know,
that's
my
belief.
C
I
think
it's,
I
think
it's
right
on
the
money,
because
you
actually
have
a
lot
of
work.
That's
been
done
to
figure
out
what
should
happen
at
the
low
level
and
oftentimes.
You
have
a
break
in
the
the
status
of
reporting
and
the
frequencies
and
things
right.
Some
things
only
happen
in
edge
environments
very
slowly
and
some
happen
very
rapidly,
and
in
no
case
should
you
try
to
set
a
standard
tempo.
So
what
you
can
do
is
you
can
use
custom
resources.
C
You
can
use
all
of
the
centralized
stuff
to
drive
it,
but
then
you
just
make
an
interpretive
layer
and
that
handles
all
of
the
nuances
of
the
particular
situation.
Maybe
you
need
to
query
parameters
actively
and
you're
used
to
getting
status
reported.
So
something
does
that
and
it
looks
native
and
then
you
just
it's
really
all
about
making
interpreters
right
and
that's
a
very
scalable
model,
especially
when
you
can
reuse
code.
A
Yeah
yeah
there's
any
number
of
these
interesting
projects
in
the
edge
iot
space.
You
know
like
the.
What
is
that
project?
That
does
the
device
twins
and
things,
and
I
think
some
of
them
already
seem
to
have
work
going
to
kubernetes-
is
them,
but
others
don't.
But
I
think
the
crd
model
is
flexible
enough
that
you
could
apply
it
to
most
of
the
things.
C
Yeah,
in
fact,
like
what
what
I
often
see
is
that
people
are
taking
the
you
know
some
of
these,
especially
like
eclipse
iot
projects,
many
of
which
were
made
for
particular
scenarios
of
device
management
or
firmware
updates
and
this
type
of
stuff,
and
you
can
deploy
it
out
with
the
kubernetes
infrastructure
right.
So
how
does
that
even
get
stood
up
at
the
edge
and
that
stuff
that
you
know?
Of
course
you
know
we
look
at
with
like
io
fog,
but
yeah.
C
It's
it's
interesting,
because
you
have
a
lot
of
efforts
to
handle,
you
know
rtos
environments
and
then
to
handle
device
communication
where
you're
pulling
stuff
off
of
serial
ports,
and
then
you
want
to
standardize
it
and
then
there's
digital,
pointing
and
all
of
this
stuff.
You
put
it
together
and
you
really
probably
have
most
of
the
pieces
that
you
need
with
a
couple
of
glue
and
sandwich
layers
right.
You
don't
you
don't
have
to
write
so
much.
I
don't
think
in
order
to
achieve
most
results.
A
Yeah,
it
seems
like
the
three
of
us
are
in
agreement
and
I
think
it's
a
fascinating
opportunity.
We
just
need
the
framework
to
go,
build
this
stuff
and
and
convince
and
enlist
our
army
of
followers.
C
Well,
I
think
the
you
know
the
biggest
the
biggest
trick,
at
least
what
I
see
happening
at
the
edge
next
is
defining
standardizing.
The
openings
that
can
be
used
for,
for
common
relay
of
you,
know,
actual
data
and
device
in
for
attributes
and
information,
and
I
just
I
I
don't.
I
don't
know
exactly
what
form
it's
going
to
take,
but
you
know
I
personally
believe
that
something
like
a
well
I
mean
mort
you're,
describing
a
modeling
right.
So
what
is
the
language
that
was
used
to
do
the
modeling?
Was
it?
B
I
think
they
they're
kind
of
seeing
the
the
custom
resources
or
the
the
definition
of
the
the
replication
control,
but
not
not
a
replication
controller
like
the
the
yaml
file
you
hand
into
your
cluster
as
a
modeling
language.
So,
okay.
C
B
C
Here
you
know
yeah,
that
if
you
can
define
the
structure
of
what
a
device
is
supposed
to
look
like
in
terms
of
how
it
maps,
then
that
that
that's,
I
think
one
of
the
trickiest
parts
is
just
finding
a
universal
way
to
describe
anything
that
can
can
interface,
and
then
that
means
that
the
building
out
of
those
interpreters
is
not
not
so
difficult
anymore.
It's
just
mapping.
B
A
A
You
know
they're
going
to
have
things
like
firmware
versions
that
maybe
are
akin
to
versions
of
a
contain.
You
know
a
container
image
and
they're
going
to
have
things
like
configuration
that
are
akin
to
things
you
try
to
do
with
container
based
apps,
and
I
think
you
should
be
able
to
manipulate
those
with
code
in
controllers
to
handle
things
like
doing
firmware,
updates
on
devices
that
aren't
even
traditional
computers
and
aren't
really
container
images.
C
Yep
and
calling
calling
functions
on
them
right
to
do
things
at
the
edge
that
are
probably
related
just
to
that
device,
but
if
properly
labeled
should
be
callable
just
like
any
other
updating
any
other
attribute.
A
A
Infrastructure
and
communication
protocols
is
somewhat
akin
to
istio,
although
they're
they're,
probably
using
non-tcp
pro
you
know,
istio
was
optimized,
I
believe
for
layer,
7
http
based
tcp
and
you're,
going
to
want
to
do
something
similar
of
a
central
control
plane
with
observability,
to
manage
the
myriad
of
industrial
and
edge
appropriate
protocols,
but
that's
a
big
problem
that
needs
solving
too,
and
I
think
you
can
use
the
model
that
was
used
initially
for
kubernetes
cloud
data
centers
out
at
edge.
C
You
know
from
one
protocol
to
another
receive
data
in
which
is
just
byte
order
right
and
something
has
to
slice
it
up
and
and
translate
it,
and
then
you
have,
after
that,
I'm
usually
dating
a
state
that
you
can
consider
universal
right.
It's
it's
you've
got
keys
and
values,
and
it's
your
choice.
C
Whether
you
want
to
send
that
up
toward
the
you
know
via
tcp
and
send
it
up
to
an
http
receiving
system,
maybe
a
rest
api
or
whether
you
want
to
keep
it
down
there
at
the
edge
and
put
it
back
out
over
mqtt
or
whether
you
want
to
put
it
over
the
serial
line.
So
it's
it
really
isn't
a
matter
of
of
a
fixed
network
translation.
I've
always
thought
it.
C
C
There
were
some
attempts
to
do
that
with
bluetooth,
low
energy,
basically
report.
You
know
things
in
a
certain
structure
up
to
rest
apis
or
over
over
tcp.
So
you
can
convert
bluetooth
data.
You
know
signals
to
wi-fi
data
signals
and
then
pick
them
up
on
a
server
on
the
on
the
lan
or
something
and
to
the
best
of
my
knowledge.
You
know
only
the
demo
devices
that
came
with
those
gateways
really
functioned
and
nothing
else
really
worked
because
you
have
to
establish
a
session
right.
It's
just
a
totally
different
model
of
networking.
A
Yeah-
and
I
think
some
of
the
reasons
those
proprietary
very
specialized
hardware-based
transports
exist
is
they
have
very
unique
requirements
that
don't
map
you
know
to
l2
ethernet,
and
they
never
will,
and
you
know,
trying
to
determine
that
you
always
have
to
convert
to
be
able
to
manipulate
them.
Just
seems
wrong
that
you
need
a
management
layer
to
manage
those
specialized
tren
networks
based
on
specialized
transport
and
leave
them
the
way
they
are
simply
take
on
the
management
and
observability.
B
B
Sorry
go
ahead
mark,
I
think,
one
of
the
the
biggest
problems
with
all
these
different
protocols.
B
I
know
what
you're
talking
about
because,
especially
in
manufacturing
like
we
have
a
lot
of
these
proprietary
protocols,
they
try
to
kind
of
do
the
management
on
your
own
instead
of
like
they
want
to
kind
of
keep
them
in
your
system
instead
of
saying,
okay,
we
we
want
you
to
use
different
protocols.
B
You
know,
like
profibus,
for
example,
tries
to
to
keep
everything
together
and
they
implement
the
management
plan
themselves,
and
I
think
that
that's
one
of
the
biggest
problems
right
now
that
everybody
is
doing
his
own
thing
instead
of
just
like
coming
together
and
like
creating
one
control
plane
that
can
handle
that.
C
Model,
but
until
open
source
ate
the
world.
A
C
Yep
there's
over
1500
protocols.
Well
I
mean
steve.
You
know
this
from
your
past
right
there's
over
1500
protocols
in
the
in
the
industrial
and
and
what
is
now
called
iot
space
and
there's
there's
a
you
know,
a
top
20
or
so
that
are
everywhere
and
the
rest
of
them
trail
off,
depending
on
which
vertical
you're
in
and
which
vendors.
A
Yeah,
I
think,
for
industrial
control,
I'd
even
go
off,
you
know.
If
you
want
to
be
a
big
player,
you
might
have
to
go
well
above
the
top
10
into
40
to
60,
but
then,
if
you
get
into
other
specialized
niches
like
automotive,
I
mean
there's
potential
here,
just
look
at
the
number
of
buses.
If
you
I
don't
know,
I'm
a
you
know
hobbyist
mechanic
and
work
on
my
own
car.
A
But
if
you
even
look
at
the
different
buses
connected
up
to
your
obd2
port,
there's
potentially,
I
believe
10
of
those
not
every
one
of
them
is
in
every
car,
but
there
are
a
lot
of
them.
C
Yep,
that's
it's
actually,
there's
some
really
great
obd2
port
bluetooth,
low
energy
modules
that
you
can
get
it's
hard
to
find
one
that
that's
got
the
right
configuration.
C
A
Yeah
yeah
I've
I've
got
one
of
those
from
mine
and
it
does
most
of
those
protocols,
but
I
mean
I
found
looking
into
it
that
the
anti-lock
brakes
might
be
on
one.
You've
got
low-end
i2c
stuff,
routed,
routed
there
for
other
things,
like
entertainment,
buses,
door
and
window
access
control
security.
A
C
Yeah,
it's
a
little
scary
too,
especially
the
like
there's,
you
can
clear
the
flag
for
the
the
check
engine,
and
so
just
I
mean
that's
a
convenient
way
to
get
that
to
turn
off,
and
you
know
you
don't
want
it
off.
A
Yeah
well,
I
was
surprised
because
those
consumer
obd2
readers
just
show
you
the
surface
part
of
the
iceberg
and
when
you
look
at
these
specialized
ones
that
are
required
to
do
firmware,
updates
on
the
modules
and
things.
That's
when
you
really
discover
how
many
of
these
protocols
are
there?
Oh
yeah
certain
v
on
certain
vehicles.
A
A
But
anyway,
the
bottom
line
is,
I
think,
that
mapping
out
all
of
these
protocols
that
are
edge
and
iot
you're,
going
to
find
that
you
really
need
a
solution
that
takes
plugins.
If
you
wanted
to
attempt
to
do
the
you
know,
control
plane
for
these,
whether
it's
kubernetes
or
something
like
an
istio
that
is
non-kubernetes
but
works
with
kubernetes
you'd
best
develop
it
so
that
it
was
extensible.
C
Oh
yeah
yeah,
that's
the
only
approach,
that's
going
to
work,
that's
things!
Change
too
fast
and
you'll
never
incorporate
all
of
that.
That's
out
there
there's
an
interesting
project
at
the
eclipse
foundation.
That's
that's
getting
more
and
more
attention,
which
is
a
spark
plug
and
there's
a
working
group
forming
around
it.
C
The
one
of
the
original
creators
of
the
mqtt
protocol
is
heading
it
up
his
name's
arlen
and
really
really
amazing
guy,
and
so
the
project
is
intending
to
put
standard
data
definitions
around
mqtt
messages
such
that,
no
matter
what
you
get
off
of
a
device
if
you're
going
to
use
mqtt
as
your
backbone,
you
you
translate
it
into
this
final
agreed
upon
format
across
you
know
across
verticals
or
companies
in
the
same
vertical,
and
you
have
some
agreement
such
that
doesn't
matter
how
you
implement
it
but
you're.
C
We
finally
have
kind
of
a
point
of
of
common
data
structure,
and
I
I
think
it's
great
it's
about
time,
especially
because
it
rides
on
top
of
the
the
open.
You
know
the
open
bytes
of
mqtt,
you
know
packets.
It
doesn't
doesn't
matter
what
you
had
in
the
past.
If
you
write
a
little
translator,
you're
still
good
to
go.
A
Yeah
that'd
be
interesting,
and
I
could
see
making
devices
talk
in
this
new
one,
but
another
way
of
doing
it
that
may
be
brings
legacy
to
the
party
would
be
to
have
sort
of
a
brokerage
receive.
Everything
and
attempt
to
you
know
manipulate
them
like
a
kubernetes
admission
controller,
where
you
know
something
gets
accepted
and
potentially
modified
and
put
into
some
kind
of
nosql
database.
That
would
put
these
in
a
standardized
form
for
people
to
consume.
C
Cool,
well,
here's
all
I'll
add
something
to
that.
Then
there's
a
lot
of
information
context,
information
relative
to
the
you
know
or
related
to
the
the
solution
that
you've
got.
That
needs
to
be
added
to
the
simple
low-level
data.
If
you've
got
some
really
dumb
temperature
sensors,
that
are,
you,
know
or
moisture
sensors
that
are
out
for,
like
smart
agriculture,
you're
not
going
to
time
synchronize
all
of
them,
this
right,
2
000
of
them,
you
what
you
really
want.
Is
you
want
them
to
report
their
raw
data
and
the
time
stamp?
C
And
the
identifier
you
want
to
assign
those
on
an
upper
layer?
Because
you
also
don't
want
to
take
the
you
know.
Let's
say
it
has
something
like
a
mac
address
right.
Maybe
it's
on
laura
wan
or
you
know
something
that
has
similar
similar
features
to
your
your
mac.
Address
that
we're
familiar
with
in
the
tcpip
network
on
ethernet
and
so
something
similar
to
that
a
unique
identifier.
You
can
read:
you're,
probably
not
going
to
know
which
one's
which
so
somewhere
you've
got
them.
You've
got
a
map
and
you
could
assign
a
timestamp.
C
You
could
use
that
map
to
assign
contextual
information
like
a
label
of
which
sensor
it
was.
Maybe
you
have
some
geo
coordinates
for
when
you're,
where
it
was
placed
and
all
of
this
stuff
is
you
think
about
it?
A
time
or
a
temperature
reading
might
be
four
bytes
coming
from
the
device
and
you're
sending
this
over.
She
said,
especially
over
like
bluetooth,
energy
or
lorawan,
or
whatever
everybody
really
counts.
If
you've
got
you
know,
5
000
of
them.
A
And
I
don't
know
I-
I
don't
know
that
it's
you.
This
is
my
idea,
but
I've
I've
thought
about
it.
Is
this
idea
that
you're
going
to
have
all
this
lte
5
delivered
data,
some
of
it
with
properly
labeled
some
not,
but
if
you
can
take
it
to
a
higher
tier
that
maybe
has
good
clock,
synchronization
nose.
Loop
knows
the
location
and
appends
to
it.
At
some
level,
some
of
the
stuff
is
sensitive
enough
that
you
even
want
to
authenticate
it
so
that
it
can't
be
tampered
with
later.
A
If
you
could
apply
blockchain
like
journaling,
I'm
not
saying
it's
literally
blockchain,
because
there's
some
issues
with
that,
but
you
know
put
put
enough
journaling
on
them
that
you're
going
to
be
able
to
atta
you're,
going
to
be
able
to
detect
somebody
attempting
to
revise
history
by
tampering
with
the
data
and
then
feed
it
up
with
this
intact,
where
you
can
authenticate
that
this
is
the
actual
reading
from
this
application.
A
C
Imagine
you
have
a
supply
chain
integration
and
at
your
warehouse,
you're
receiving
you
know
two
million
dollars
worth
of
components
and
on
several
pallets
and
they
come
in
and
there's
some
readings
taken,
not
only
for
the
contents
of
the
box,
but
the
weight-
and
you
know,
there's
all
types
of
stuff
that
you
use
to
take
measurement
before
you
release
the
the
invoice
payment
and
you
automatically
decide
that
the
shipment
doesn't
look
complete.
So
you
pass
it
instead
of
settling
that
day,
you
pass
it
off
for
manual
review
and
it
takes.
C
You
know
10
days
for
someone
to
get
around
to
looking
at
it,
because
you
no
longer
have
people
on
the
dock
all
the
time
and
the
company
that
sent
you
the
stuff
needs
that
working
capital
and
now
they're
mad
at
you
and
they
say,
prove
it
well
who
owns
the
data
and
who
is
able
to
determine
that
it's
entered
into
an
immutable
state
and
that
that
that
information
that
was
used
to
make
a
business
transaction
happen
automatically
or
not
happen
automatically
is
still
pure
and
was
generated
correctly
right.
C
This
is
really
important
stuff
for
for
automating
the
world
which
we're
far
from
doing,
but
we
are
aiming
that
direction
to
to
have
fewer
manual
checks
and
balances.
A
Right
yeah,
I
I
think,
there's
a
lot
of
potential
for
things
that
are
say:
gateway-like,
where
you
take
really
low
capability
devices
and
when
they
first
hit
a
more
capable
gateway.
You
could
append
a
lot
of
information,
maybe
attempt
to
make
them
tamper
resistant,
maybe
even
engage
in
feeding
them
up
through
redundant
channels
of
communication.
You
know,
for
example,
some
of
these
forecasts
of
lte5
are
projecting.
A
You
know
in
a
competitive
world
anyway,
cheaper
transports
so
that
maybe
you'd
go
with
multiple
carriers
for
resiliency.
You
know
if
you're,
in
an
area
where
you
could
get
two
cell
carriers
or
three
or
four
that
all
could
provide
service.
If
it's
cheap
enough,
maybe
you
go
with
two
and
reaching
some
gateway
devices.
C
Yeah
well
also,
you
know
when
the
when
the
the
the
reaches
such
a
point,
that
it's
spread
across
almost
all
of
the
physical
infrastructure
available,
it
becomes
equivalent
to
the
physical
infrastructure
again
right,
so
there's
only
so
much
only
so
much
redundancy
you
can
build
into
something
that
has
failure
points
you
know
inherent
in
its
underlying
hardware,
but
to
you
to
your
point
about
the
edge
and
the
edge
communication
in
5g
yeah.
C
It's
like
if
I
really
could
have
reliable
communication
that
where
it
wasn't
priced
at
the
absolute
limit
of
my
willingness
to
pay,
I
would
definitely
throw
redundancy
in
there
and
I
would
start
using
it
for
more
and
more
data
pathways
too
right
now.
I
I
try
to
conserve
what
I
send
over
cellular
because
there's
an
expense
associated
with
it.
Yeah.
A
Yeah,
in
a
way,
I'm
not
sure
that
in
reality,
building
into
your
hardware,
the
ability
to
use
two
carriers
might
be
a
great
insurance
policy.
Right
I
mean,
if
you
look
in
the
u.s,
there's
a
lot
of
cell
phones
that
only
have
a
radio
that
is
in
the
device,
that's
unique
to
one
of
the
carriers,
but
that's
a
lock-in,
and
once
you
get
tens
of
thousands
of
them
out
there,
they're
gonna
know
you
can't
go
anywhere
else.
A
C
Yeah,
there's
also
a
big
difference
between
these,
these
iot
devices
and
iot
gateways
and,
like
you
know,
smartphone
you,
you
don't
go
back
out
to
you,
know
the
oil
well
at
the
pump
head
to
change
that
gateway
when
you
decide
to
change
carriers
and
therefore
you
really
are
locked
in
if
you're,
on
a
single
single
connectivity
provider,
because
the
the
the
sheer
you
know
cost
of
going
out
and
changing
the
hardware
out
or
changing
the
radio
module
is
just
prohibitive.
C
So
I
think
there's
it's
it's
an
insurance
policy
for
having
available
connectivity
and
yeah
financial
leverage
for
sure,
which
is
that
you've
built
in
the
rights
to
to
flip
a
switch
and
change
over
instead
of
needing
to
bear
the
physical.
You
know,
change
out
cost.
A
They
shared
that
lt5
is
going
to
have
as
big
an
impact
on
the
world
as
the
transition
from
dial-up
modems
over
to
broadband.
Did
you
know
if
you're
old
enough
to
remember
those
days
of
using
those
dial-up
modems
on
landline
phones?
You
really
didn't
know
what
you
were
in
for
when
those
things
became
obsolete
and
permanent
connectivity
to
the
internet
became
the
norm
rather
than
you
know,
this
intermittent
contact
and
the
lt5
they
were
contending
is
going
to
be
as
big
a
technology
shift.
A
In
terms
of
the
use
cases
it
enables
where
whole
businesses
were
essentially,
I
would
contend
created
that
couldn't
have
been
done
without
broadband
internet,
and
you
know
people
didn't
even
dream
of
what
the
possibilities
were
anyway.
In
their
talk,
they
went
on
to
quote
another
forecast
from
gartner.
That
was
saying.
A
If
I'm
remembering
their
data
correctly,
they
were
saying
that
there's
been
this
tremendous
trend
of
not
necessarily
applications
moving,
but
new
application
workloads
are
being
hosted
in
public
clouds
and
it's
been
one
of
the
biggest
growth
sectors
in
technology,
but
that
the
the
edge
enablement
brought
about
with
lte5
among
other
technology
factors
is
going
to
start
to
create
a
scenario
where
the
bulk
of
new
workloads
is
going
to
be
hosted
out
at
edge,
and
I
think
their
figure
was
that
by
the
year
2022
70
of
new
workloads
were
going
to
be
edge
hosted
rather
than
you
know
that
could
be
edge
or
on-prem.
C
It
doesn't
even
have
to
be
100,
true
yeah.
I
I
certainly
like
that.
I
mean,
given
you
know
from
where
I
sit,
but
I
mean
at
least
it's
at
least
partially.
True
and
I'd
say
you
know,
50
true
would
be
was
wonderful,
but
when
you
think
about
it,
there's
a
there's
a
lot
of
advantages
to
what
we
get
from
the
the
the
hosted
cloud
infrastructure,
but
at
the
edge
you're
paying
for
your
own
electricity
you're,
paying
for
or
have
set
up
your
connectivity
and
you
likely
own
the
hardware.
C
Those
are
a
couple
of
things
that
you
were
paying
for
in
in
the
the
the
cloud-hosted
infrastructure
and
if
you
have
a
workload
that
really
doesn't
need
to
be
run
centrally.
All
of
these
things
already
taken
care
of
at
the
edge
means.
You
have
very
good
scale
economics
for
the
amount
of
processing
that
you
do.
C
If
you
can
keep
up
with
the
workload,
it's
incredibly
cost
effective,
but
all
of
that
stuff
became
really
hard
to
manage
historically,
and
so
we
decided
you
know
centralizing
was
better
and
and
at
the
time
I
think
that
was
right
and
I
think
the
you
know
all
of
what
was
in
the
cloud
is
good.
A
I
think
their
definition
of
edge
and
most
people,
though,
would
entertain
scenarios
where
it
isn't
necessarily
you
own
it
they're
they're,
classifying
things
run
on,
let's
say
operator,
hardware
operator,
internet
carriers
and
telcos
that
would
host
in
cell
phone
towels
towers,
switching
offices
as
also
being
edge,
hosted.
C
Wow,
well,
that's
amazing.
I
mean,
I
hope,
that's
the
case.
That'll
be
great,
but
there
yeah,
I
guess,
they're
the
business
drivers
are
ones
of
of
latency.
I
don't
know
what
the
cost
will
look
like
right.
What
will
the
spot
pricing
be
for
running
a
microservice
like
you
know,
right
at
a
cell
tower
versus
running
it
in
the
in
the
public
cloud.
I'm
not
sure.
A
C
Oh
I'd
say:
definitely
there's
a
business
there.
Definitely
in
fact
it
would
for
a
lot
of
carriers.
It
would
change
them
over
from
being
dump
pipes
right.
I'm
just
going
to
move
your
data
for
you
to
being
an
active,
active
service
provider
of
compute
infrastructure.
That's
a
that'd,
be
a
great
transition
for
them.
Normally
they're
just
pathways.
A
Yeah,
I
mean
even
look
at
the
potential
market
there.
If
you
could
host
that
edge
and
you
owned
certain
sensors.
You
know
some
people
are
going
to
have
to
put
up
the
capital
for
these
sensors
in
a
lot
of
these
smart
cities,
projects
they're
owned
by
government
entities,
but
I
think
in
potentially
others
there
might
be
ways
for
privately
owned
collected
data
to
be
put
there.
That
is
of
value
to
other
people
other
than
the
people
who
collected
it
and
they'd
want
to
potentially
use
it
out
near
the
edge.
A
C
A
I'm
not
saying
it's
the
best
way
to
do
it,
but,
for
example,
if
I
had
the
vehicle
counts
and
speeds
of
every
freeway,
in
los
angeles,
I
would
contend.
I
could
give
you
a
weather
report.
I
mean
there
are
some
things
that
happen
when
it
rains.
You
know
vehicles
slow
down
in
predictable
patterns
that
from
vehicle
sensors
alone.
I
think
I
could
probably
give
you
a
good
idea
at
least
some
weather
conditions.
A
C
Yes,
actually,
there's
there's
business
around
that
already
taking
the
the
security
camera
footage
from
parking
lots
and
using
it
to
sell
back
to
retailers,
information
about
their
own
stores.
A
C
There
was
a
in
the
city
of
san
francisco.
There
was
a
a
project
that
came
through
this
program
called
startup
in
residence
or
stir
it's
really
amazing
program
and
the
startup.
I
don't
know
what
became
of
it.
There
was
a
company
called
birdie
and
I
think
it
was
and
up
until
this
project
came
through
in
partnership
with
the
city.
There
was,
I
think,
just
maybe
three.
C
Maybe
it
was
even
one
air
quality
sensor
that
was
that
was
owned
by
the
city
that
had
been
there
for
20
30
years,
and
it
was
its
position
was
you
know
up
on?
I
think
maybe
it
was
even
the
same:
the
same
hill,
there,
hill,
slash
mountain,
that's
that's
a
home
of
twin
peaks,
and
that
and
the
you
know
the
radio
tower
that's
up
there
and
stuff.
C
So
you
you
did
it
with
you
know
ten
thousand
dollar
sensor
units
and
you
put
out
a
few
of
them
and
you
got
what
you
needed,
but
then
along
comes
a
startup
and
it
asks
people
to
put
them
in
their
their
window
of
their
apartment
or
their
home
or
whatever
around
the
city
and
then
allow
it
to
talk
on
their
wi-fi
and
in
exchange
for
letting
it
ride
over
their
wi-fi
and
sending
the
data
back
to
the
bertie
cloud
or
whatever
it
was.
C
You
know
people
got
the
air
quality
report,
they
they
basically
gave
these
out
to
people
or
or
charged
a
low
amount
for
them,
and
so
now,
all
of
a
sudden
there's
thousands
of
data
points
giving
a
much
more
accurate
map
of
air
quality
in
the
city
of
san
francisco.
The
low
points,
the
high
points,
the
the
neighborhoods
that
are,
you
know
still
near
whatever
light
manufacturing,
I'm
not
even
sure.
A
C
A
Yeah
and
in
a
lot
of
cases,
I
don't
think
you
necessarily
have
to
pay
people
to
do
it.
If
it's
cheap
enough-
and
you
know,
there's
either
some
model
where
they
get
some
benefit
out
of
having
that
sensor
themselves
or
you
know
just
want
to
join
some
big
crowd-sourced
effort,
and
I
know
of
two
of
two
things
like
that:
there's
a
weather
company
called
weather
underground
that
allows
you
to
build
up.
A
A
They
have
readings
within
blocks
of
my
house
and
they're,
really
capable
of
showing
microclimates,
not
one
or
two
sensors
for
a
large
city,
like
los
angeles
and
because
of
that
density
of
data.
I
think
that
they've
actually
got
advantages
over
some
of
the
traditional
data
gathering
situations
already
there's
another
one
called
flight
radar
24
that
are
using
these
new
air
traffic
control
transponders.
I
don't
think
they
fully
officially
go
on
the
air
for
another.
A
Few
years,
but
the
old
model
was
that
control
towers
at
airports
used
radar
to
spot
incoming
aviation
and
the
bigger
airplanes
would
have
tags
broadcasting
some
kind
of
a
unique
identifier,
but
the
new
ones
involve
point
to
point
where
the
the
airplanes
themselves
are
required
to
have
a
device
that
has
gps.
So
they
know
their
own
position
and
just
broadcast
out
the
position
and
a
receiver
at
the
airport
then
doesn't
have
to
use
radar
to
know
where
these
things
are.
If
it
can
count
on
all
of
these
having
their
own
beacon,
saying.
A
Here's
who
I
am
and
here's
where
I'm
located
and
it
turns
out
that
you
can
buy
a
receiver,
an
antenna
off
amazon
now
for
about
forty
five
dollars
and
enough
people
have
put
these
on
the
air
in
urban
areas
that
there's
very
dense
coverage.
And
this
you
know
if
you're
inspired
to
go.
Look
at
this
go
load,
a
cell
phone
app
called
flight
radar
24
and
if
you're
in
an
urban
area.
A
Like
me,
you
can
sometimes
hear
a
helicopter
going
overhead
and
the
thing
will
actually
tell
you
what
that
is,
and
in
many
cases,
even
show
you
a
picture
of
the
you
know
the
vessel
and
it's
pretty
impressive
stuff
for
something
driven
entirely
by
crowd
source
data.
C
I'm
I'm
downloading
it
right
now,
thanks
for
the
the
new
hobby
same.
A
Yeah
you
can,
by
the
way,
if
you're
on
an
airplane
with
wi-fi,
you
can
run
that
app
over
the
wi-fi
and
see
your
own
plane
and
know
where
it
is
probably
with
better
accuracy
than
what
some
of
those
seat
back
things
do,
and
you
can
actually
see
the
other
airplanes
around
you
and
spot.
When
you're
coming
in
near
the
airport,
you
can
actually
spot
whether
the
airport
appears
to
be
running
circular,
holding
patterns
by
just
watching
it,
because
you
get
your
own
personal
radar
coming
up
on
your
device.
C
This
is
really
amazing.
I
want
to
say
that
I've
been
shown
something
like
this
before
in
a
web
app,
and
what
I'm
looking
at
right
now
is
just
a
whole
host
of
planes
around
you
know,
I'm
nearest
to
oakland
live
airport
berkeley
and
I
want
to
say
I've
seen
something
like
this
before
I
love
it.
A
Yeah,
it's
remarkable
that
they
have
people
go
out
in
once.
You
know
the
tag
number
of
an
airplane.
They
try
to
get
people
to
take
a
picture
of
it
so
that
I
don't
know
if
what
device
you're
on.
But
if
you're
on
a
tablet,
you
can
actually
see
the
picture
of
the
aircraft
and
know
whether
you
know
in
my
neighborhood.
I
want
to
know
whether,
if
I
hear
a
helicopter
hovering
if
it's
a
traffic
helicopter,
I
kind
of
know
there
might
be
an
accident
on
the
nearby
freeway.
A
A
That
site
by
the
way,
has
is
somewhat
controversial.
I
think
because
they've
they're,
because
it's
crowdsourced
and
I
think
it's
hosted
in
france-
they've-
refused
to
mask
planes.
So
some
of
the
police,
vehicles
and
air
force-
one
don't
actually
want
they
want
to
be
invisible,
but
that
database
leaves
them
leaves
them
fully
visible.
Yeah.
C
Well,
there's
you
know:
there's
a
information
wants
to
be
free,
which
is
often
misinterpreted.
Is
that
it
should
be.
You
know,
without
a
price,
but
that's
not
the
case.
It's
that
it
wants
to
be
known,
that
is
yeah
and
it's
how
very
how
but.
A
Anyway,
as
a
as
a
principal,
I
think
that
there's
potential
with
edge
and
iot
for
a
lot
of
this
private,
it
isn't
just
smart
cities.
I
think
there's
potentially
going
to
be
a
lot
of
privately
generated
information
that
is
of
value
to
other
people
and
if
you
ever
ended
up
with
this
inverted
cloud,
where
the
model
is
that
operators
are
hosting.
A
You
know
compute
capacity
near
where
this
data
comes
from.
They
might
be
in
a
position
to
also
host
access
to
this
data
kind
of-
maybe
you
know
like
right
now.
If
you
do
a
youtube,
video
google
will
place
ads
on
this,
and
let
you
monetize
it
well
what?
If
there
was
a
business
model
where
you
could
have
this
crowdsourced
data
have
relay
it
to
one
of
these
operators
and
if
they
could
give
you
some
vector
where
you'd
get
a
fraction
of
a
cent
from
every
anyone,
who'd
ever
use
it.
C
Yeah,
currently,
we
have
youtube
celebrities
and
we
assume
we'll
have
we'll
have
sensor
and
iot
data
moguls.
C
Yeah
well,
let's
you
know,
let's
hope,
that
people
out
there
who
have
the
time
to
listen
to
us
banter
or
at
least
picking
up
something
and
maybe,
if
nothing
else
that
will
entice
them
to
come
to
the
next
meeting
and
you
know,
show
us
a
thing
or
two.
A
Yeah
yeah,
if
anybody
does
pick
this
up,
you're
free
as
always
to
put
your
own
items
on
the
agenda,
so
this
working
group
is
always
open
to
whatever
okay,
I'm
gonna.
Stop
the
meeting
shut
down
the
recording,
but
maybe
I'll
talk
to
you
again
in
a
couple
weeks
in
the
the
us
series
of
the
meetings
good
night,
kilton.