►
From YouTube: Kubernetes WG IoT Edge 20190118
Description
January 18 2019 meeting of the Kubernetes IoT Edge Working Group - demo of ioFog https://github.com/Eclipse-ioFog
A
B
A
B
So
for
those
unfamiliar
or
new
to
kubernetes
the
their
the
way
they
run
the
community
is
that
each
working
group
and
special
each
working
group
and
special
interest
group
is
allowed
to
propose
an
intro
and
the
deep
dive
session.
There's
no
guarantee
they
take
them.
But
I
don't
know
my
experience
is
as
as
long
as
you
follow
the
guidelines
for
what
you
propose.
B
I
think
they
do
take
them.
Kubecon
barcelona
is
coming
up
in
may,
and
the
deadline
to
put
the
proposal
in
for
these
sessions
is
today
midnight
pacific
time.
The
sessions
are,
if
you
do
an
intro
and
a
deep
dive,
they
reach
35
minutes.
This
is
a
new
policy,
but
as
of
this
meeting,
they
want
original
new
content.
That's
never
been
presented
anywhere
else
there.
So
formerly
when
they
did
these
coupons
across
regions,
they
actually
were
happy
to
take
one.
B
You
know
you
could
present
in
europe
and
then
do
the
exact
same
session
again
in
shanghai,
but
they've
now
announced
that
since
people
can
watch
the
recording
and
they
record
all
of
them-
they
don't
want
that
to
happen
anymore.
So,
as
a
group,
if
we've
got
something
worthwhile
to
present
talk
about
or
even
have
a
birds
of
a
feather,
you
know
we
can
get
get
that
proposed.
B
I
don't
know
about
dion's
attitude,
but
some
of
us
I
see
preston
on
the
call
too,
and
I
didn't
look
if
cindy
is
here,
but
you
know
we,
we
sort
of
organized
some
panels
at
the
seattle
event,
but
we'd
like
to
mix
it
up
and
give
an
opportunity
for
other
participants
rather
than
leadership,
to
present
we're
not
looking
for
sales
pitches
for
anybody
new
here.
So
the
kind
kind
of
content
we're
looking
for
is
kind
of
you
know
open
source
things
that
are
use
reusable
across
the
whole
community.
B
It's
okay
to
even
talk
about
kubernetes
in
the
context
of
futures
like
gee.
It
would
be.
You
know
there
are
these
issues
with
applying
it
for
iot
use
cases
or
at
edge
and
it'd
be
great
if
items
one
two
or
three
could
be
addressed,
so
that's
that's
a
perfectly
valid
talk
too.
So
with
that
said,
I
guess
we
can
throw
it
out
for
open
discussion
as
to
whether
anybody's
got
specifics
to
propose,
and
if
you
look
at
that
meeting
document
that
I
posted
a
link
to
in
the
chat.
B
We
did
cover
a
little
bit
of
subject
matter
last
week
or
at
the
last
meeting
for
the
group
going
forward
in
the
year
and
that's
kind
of
a
potential
topics
to
bring
up
as
sessions
and
just
if
you
can't
make
barcelona
the
deadline
for
kubecon
shanghai,
which
it
will
be
in
june
a
month
after
the
europe
event.
I
think
we've
only
got
about
three
weeks
to
get
a
proposal
in
for
that
too.
So
it's
coming
up
very
soon.
B
And
if
anybody
on
this
call
wants
to
participate,
you'd
be
expected
to
be
physically
present
there,
so
that
might
I
know
I'm
gonna,
be
my
employer
has
told
me
that
I'm
to
book
flight,
so
I'm
very
likely
to
be
physically
present,
but
that
doesn't
yeah.
I
do
I'd
be
happy
to
turn
over
the
keys
to
that
to
somebody
else
who's
who
who
wants
to
take
over
one
of
those
slots.
B
You
know
which
entails
getting
an
email
and
a
bio
that
doesn't
automatically
get
copied
over
to
the
session
file
that
that's
about
it.
It's
it's
fairly
informal.
In
addition
to
the
abstract,
you
can
put
what's
called
a
detailed
description,
that's
only
used
by
the
reviewers,
and
that
has
a
long
the
limit
on
words.
There
is
probably
double
the
size.
I
don't
remember
the
exact
limit,
but
you
can
put
more
words
there.
The
actual
abstract
will
be
what's
exposed
to
the
public
when
the
session
is
listed.
B
Just
other,
since
I've
proposed
a
number
of
times
too
you
you
don't
want
to
have
you
don't
want
to
like
rush
and
abstract
with
typos
in
it
and
spelling
errors,
because
many
of
the
reviewers
will
just
automatically
reject
it.
So
it
pays
to
you
know
not
rush
russia
to
compose
one
in
the
last
20
minutes
before
the
deadline,
because
yeah
you
know
in
order
to
get
in
you,
you
want
something
that
is
going
to
get
a
favorable
review
from
the
panel
that
reviews
these
things
there.
There
are
other
things
too.
B
D
A
B
And
we
could
continue
this
conversation
on
slack
or
the
mailing
list,
so
we,
I
don't
think,
let's
just
get
a
show
of
hands,
maybe
of
people
who
think
they
might
be
prepared.
First
of
all,
if
you're
not
even
going
to
be
there
obviously
you're,
not
a
candidate
for
this
right
and
if
you've
got
suggested
topics,
maybe
just
list
them
in
the
agenda
doc
for
the
meeting
and
those
who
raise
their
hands.
We
can
maybe
have
a
conversation
on
slack
later
today
to
carry
on
with
this.
A
B
B
F
B
B
A
Maybe
because,
especially
for
the
euro,
people
like
it's
it's
getting
close
to
the
end
of
the
day,
so
it's
you
know
team
chance
that
we
will
manage
to
do
it.
Something
like
that.
But
if
anybody
has
an
idea,
I
still
said
like:
let's
put
them
in
the
in
the
in
the
agenda
dock
and
maybe
if,
if
the
deadline
is
extended,
we
can
work
on
that
next
week,
steve
you're,
muted,
I
see
you're
trying
to
speak.
A
Okay,
let's
move
on
so
the
other
other
administrative
topic
I
have
for
today
is
is
the
the
new
meeting
times
so
I
I
sent
something
on
the
on
the
mailing
base,
how
it
should
work
based
on
the
based
on
the
what
we
have
the
votes
and
and
the
requests
we're
having
from
the
people
in
in
the
the
asia
time
time
zones.
So
it's
just
a
proposal,
so
it's
open
for
discussion.
A
It's
you
know
we
can
try
it
for
a
while
see
see.
You
know
how
much
respondents
do
we
have
in
in
each
of
the
the
time
slots
and-
and
then
you
know,
adapt
to
everything.
But
you
know,
maybe
if,
if
it's
okay
with
everybody,
maybe
from
the
next
in
two
weeks,
we
can,
we
can
start
doing
the
new
time.
Anybody
have
any
thoughts
about
it.
E
A
A
A
You
know
the
sixth
and
fourth,
I
I'm
not
sure
I'd
be
regular
on
that
one,
but
the
meeting
should
be
recorded
and
we
should
plan
for
the
for
the
you
know
interesting
topics
for
the
people
participating
in
in
in
participat.
You
know
in
actual
actual
groups,
but
you
know,
let's
maybe
give
it
a
try
and
see
how
it
goes.
C
B
A
Yeah
so
there's
two
things
as
far
as
I
remember
from
creating
the
the
working
group,
so
we
have
our
official
calendar
I'll,
try
to
find
it
out
and
share
the
link
so
that
one
will
be
updated
and
then
the
on
the
in
the
it
still
said
in
the
github
in
the
description
of
the
of
the
working
group,
we
we
should
change
it
as
well.
Yeah.
F
D
A
D
F
You
how
about
now
yeah
cool,
I'm
sorry.
I
was
needed
by
the
phone
david
berka
from
red
hat.
Thank
you
for
doing
that.
I
think
I
joined
twice
or
three
times
I'm
the
global
technical
lead
for
the
internet
of
things
initiative
here
in
red
hat
er.
I
work
with
ian
pretty
pretty
often
and
obviously
kubernetes
is
a
big
team
for
red
hat
so
eager
to
learn
and
enjoy
the
the
smart
people
here
and
at
some
point
contribute
as
well.
So
thank
you.
A
H
Okay,
cool,
oh
hi,
everybody,
so
normally,
when
I
do
these
demos,
I
can
go
for
a
long
time.
There's
a
lot
to
cover.
So
what
I'm
going
to
try
to
do
is
touch
on
the
main
points
and
those
main
points
really
comprise
like
a
set
of
patterns
that
that
we've
seen
at
the
at
the
edge
over
the
last
years
that
that
seem
to
hold
pretty
steady.
H
So
I
think
those
will
hopefully
be
useful
to
everybody
and
around
those
those
patterns,
I'll
showcase,
some
of
the
functionality
everything
out
there
is
is
something
that
I'd
like
to
you
know
discuss.
So
if
there's
you
know
questions
about
it,
we'll
have
time
for
maybe
a
little,
but
for
the
most
part
you
can
just
ping
me
and
we
can
talk.
You
know
deep
dive
about
about
any
of
this
stuff.
H
H
The
main
concepts
are
that,
like
everything
involving
you
know,
the
distributed
systems
edge
computing.
There
is
on
the
actual
edge
hardware.
There
is
an
agent
and
we're
gonna
spend
some
time
with
the
the
io
fog
agent
and
looking
what
it
does
on
the
hardware
and
stuff.
And
then
there
is
a
controller
that
we
have,
and
that
controller
is
where
the
merge
point
you
know
will
be
with
kubernetes,
possibly
even
replacing
the
the
eclipse
io
fog
controller,
with
pure
kubernetes
going
forward.
H
If,
if
there's
a
way
to
do
that,
that
we
feel
really
allows
us
to
still
do
great
stuff
at
the
edge.
So
I
have
here
a
couple
of
raspberry
pi's
that
are
in
my
edge
compute
network
and
I've
installed
the
agent
and,
of
course
the
installation
is
as
easy
as
apt-get
install.
If
I
agent,
I
won't
do
that
because
it'll
upgrade,
because
we
had
a
release-
and
I
haven't
tested
it
yet
for
this
demo.
But
so
I
have
some
some
commands.
H
In
fact,
let's
see
what
version
I've
got
here,
version
yeah,
where
I'm
on
I'm
on
1.0.6
and
so
of
course,
like
all
things
coming
out
of
eclipse,
you've
got
either
epl1
or
epl2.
This
is
eclipse,
io
fog
is
epl2,
and
so,
if
I
do
io
fog
agent
status,
I
can
get
some
information
about
what's
going
on
here
on
the
hardware,
and
these
are
just
your
your
basic
stats.
H
So
I've
got
this
on
both
of
these,
and
both
of
them
are
pretty
much
sitting
idle
with
no
running
micro
services.
And,
of
course,
we,
since
we,
when
we
started
this,
the
the
only
api
that
we
felt
was
was
stable
enough
to
program
against,
for
containerization
was
docker
and
so
still
got
that
in
play.
They've
got
nothing
running
via
you
know
via
docker.
H
So
what
are
we
going
to
do
with
these
things?
What
more
can
we
do
with
these
things?
Well,
output
agent
info
will
tell
you
some
of
the
configuration
that
the
agent
has
here
on
this
on
this
particular
piece
of
edge
hardware
and
there's
some
stuff
that
we
pick
up
automatically
such
as,
if
you
set
it
to
dynamic,
we'll
determine
what
what
network
adapter
seems
suitable
for
reaching
the
controller.
You
can
change
what
controller
you
point
to,
and
we
have
this
this
mode
on
for
this
developer
mode.
H
With
that,
we
bypass
a
lot
of
the
security
checks
that
you
would
need
in
order
to
be
able
to
talk
to
a
controller
and
be
verified,
and
so
on,
verify
the
controller
from
the
agent
site
and
that's
a
great
way
for
people
to
get
started
and
build
stuff
and
allows
you
to
just
kind
of
skip
all
of
the
the
identity
certs,
and
things
like
that.
So
you
can
set
some
limits
here:
ram
usage,
disk
usage
where
you're
storing
logs
and
where
are
you
storing
data
messages?
H
If
you're
going
to
use
the
the
eclipse,
io
fog
distributed
data
bus
and
also
you
can
do
some
tuning
as
to
when
this,
this
edge
node,
actually
checks
in
it
works
with
the
controller
and
so
out
on
you
know
these
remote
oil
fields,
you've
got
2000
pumps
and
you're
trying
to
save
money
on
your
cellular
cost.
So
you
probably
want
your
edge
nodes
to
only
be
working
with
the
controller.
Maybe
you
know
once
per
hour
or
once
every
five
minutes.
Definitely
not.
H
You
know
once
every
10
seconds
or
once
every
second
or
or
whatever.
So
all
of
that
stuff
is
tunable
and
then
there's
like
there's
low
power
mode-
and
you
know
stuff
like
this
on
these
on
these
edge
notes,
so
the
real
magic.
Let's
go
and
have
a
look
in
the
in
the
controller
here,
and
I
have
a
little
cheat
sheet
that
we're
going
to
be
using
with
with
large
enough
text.
H
So
everyone
can
can
take
a
look
at
the
commands,
but
I
want
to
introduce
some
of
the
the
concepts
that
you
can
find
here
right
in
the
the
documentation
so
iofog.org
and
here
you
should
have
a
way
of
getting
a
an
environment
set
up
on
your
local
machine.
So
that
way
you
can
do
the
controller
and
the
agent
right
from
your
your
windows
or
or
mac
laptop
and
under
documentation.
H
H
H
So
you
know
in
a
shipping
center
in
mexico
and
then
this
one
is
down
in
brazil
and
you're
getting
data
off
of
oil
pumps
there
and
all
of
that
stuff
should
be
mixable
as
long
as
you
on
the
edge
compute
network.
So
I
have
here,
for
I
o
fault
controller.
Let's
take
a
look
at
the
nodes
that
I
have,
and
so
this
will
give
me
a
list
of
the
the
nodes
that
are
in
my
ecn,
my
edge
compute
network.
At
the
moment
I
have
a
couple
that
were
set
up
here
for
this.
H
This
kubernetes
edge
working
group
demo
and
that's
the
two
that
are
over
here.
This
is
so
we'll
call
this
kubernetes
working
group
demo,
one
and
demo
two
and
you
can
see
here.
We've
got
some
information
that
the
the
agent
damon
is
running
and
then
we've
got
some
information
about.
You
know
last
time
we
heard
from
it
and
you
know
I
can
take
a
look
at
the
the
cpu
usage
right
now.
H
H
H
So
you
have
to
have
some
storage
location
for
for
containers,
and
we
all
know
that
you
know
private
registry
is
typically
used
for
you
know,
deploy
and
if
you're
going
to
share
stuff
around,
you
might
want
to
use
docker
hub
or
some
publicly
hosted
registry.
You
can
mix
and
match
them
all
with
with
io
fog,
because
one
of
the
the
things
that
we've
seen
is
that
once
people
develop
microservices
that
are
industry,
specific
really
good
example
is
like
a
scada
to
mqtt
translator.
H
H
It
means
that
you
have
some
way
to
actually
draw
it
down,
and
so
you
can
mix
and
match
match
registries,
and
so,
if
we
take
a
look
here
at
the
the
documentation
too,
we
have
for
controller,
you
have
registries
and
you
can
add,
remove
you
know,
and
they
can
be
private
and
public
and
if
they're
private,
you
need
to
provide
the
credentials
and
we'll
allow
you
to
draw
from
from
all
across
those.
And
so
what
ends
up
happening?
H
Is
you
list
the
microservices
across
various
registries
that
you
would
like
to
be
able
to
access
and
control
from
this
this
control
plane?
We
call
that
the
catalog,
and
so
the
microservice
catalog
list
will
give
you
the
the
listing
of
everything
that
you've
set
up,
that
you
would
like
to
be
able
to
take
advantage
of,
and
so
we
have
some
that
we'll
be
deploying
here
and
some
simulators
and
some
useful
ones
like
you
know,
instant,
rest,
api,
container
and
stuff,
and
you
put
these
in
as
the
administrator.
H
H
H
I'm
going
to
create
a
what
is
called
a
flow,
and
let
me
talk
about
what
a
flow
is
and
then
I'll
use
that
to
deploy
microservice
there's
another
pattern
that
has
emerged
outside
of
just
the
concept
of
the
the
edge
compute
network
and
that
pattern
is
one
of
overarching
orchestration
across
multiple
edge
nodes,
regardless
of
their
location,
location.
Being
another
pattern
that
we
can
talk
deeply
about.
But
so
we
have
this.
H
H
H
We
go
great
so
now,
if
I
do
that
that
flow
list,
of
course
we're
in
and
our
identifier
is
five
and
we'll
use
that
to
deploy
a
service
so
rather
than
have
me
mistype
anything,
I
have
it
here
in
my
cheat
sheet.
So
let's
see
oh,
you
know
what
actually,
let's
take
a
moment
to
look
at
this
real,
quick
because
well
here
I
want
to
get
started
then
we'll
do
that
all
right!
Let's
go
ahead!
This
one
will
be
kind
of
interesting.
A
H
And
then
we
know
it's
flow,
five
all
right,
that'll
be
our
identifier
for
which
edge
node,
and
then
this
is
my
flow
identifier
that
I
reserved
for
this
so
for
which
edge
node.
Let's
put
it
on
this
one
here,
so
I've
got
a
unique
identifier
for
the
edge
node,
and
let's
do
that
should
give
me
a
thumbs
up,
nope
duplicate
name,
because
I
did
it
just
earlier
today
to
test.
H
Let's
just
call
it
bluetooth,
api
2
and
there
we
go
okay,
it's
on
its
way,
so
I
also
wanted
to
show
this,
so
somebody
has
to
abstract
away
some
of
the
the
hardware
concerns
because,
as
we've
you
know
discussed
in
prior
working
group
meetings.
You
know
there
is
a
diversity
of
hardware
and
that
makes
some
challenges
well,
so
you've
got
you
can
specify
for
any
of
the
edge
nodes
that
you
create.
H
You
can
specify
either
that's
arm
or
x86
or
I
don't
know,
but
that's
just
the
very
beginning,
so
controller
right
now.
The
eclipse,
io
fog
controller
is
aware
of
the
different
hardware
types,
the
different
architectures.
Actually
because
it
goes
even
you
know
well
beyond
this,
and
it
won't
be
long
before
the
listing
will
include
you
know
ios
and
android
devices,
which
you
know
you're
not
going
to
jailbreak
these
devices
at
commercial
scale.
So
they
need
to
be
able
to
do
something
other
than
containers.
H
So
it's
most
likely
going
to
be
the
the
equivalent
of
serverless
functions
and
then
there's
the
ability
to
run
unit
kernels
on
with
with
eclipse
io
fog.
But
that
would
be
that
on
devices
that
are
are
listed
as
as
possibly
being
a
different
architecture,
and
so
but
the
gist
of
it
is
you
want
some
way
of
being
able
to
list
for
each
edge
microservice
list
where
it's
suitable
to
run
and
for
things
like
lambdas.
H
It
they'll
be
suitable
to
run
on
a
larger
number
of
devices
and
for
things
like
vms,
obviously
then
you're
more
constrained
and
that's
okay,
because
the
edge
is
a
very
weird
nasty
place
where
hardware
looks
very
different
across
all
different
deploys
and
you
want
to
be
able
to
still
control
it
orchestrate
it.
But
you
need
to
know
more
about
it
and
and
how
to
set
that
all
up.
H
So
let's
have
a
look
here.
It
should
be.
I
think
this
unit
will
be
the
one
that
has
yes,
so
the
microservice
that
I
deployed
with
that
that
flow
was
pushed
to
this
node,
and
so
we
have
one
running,
and
here
it
is
so
what
I've
deployed
is
I've
deployed
a
bit
of
hardware
abstraction
so
that
I
can
do
some
cool
things
on
this
on
this
edge
node.
H
You
can
then
instantiate
the
hardware
abstraction
that
you
need
just
the
same
way
that
you
instantiate
any
other
edge
application.
If
you
don't
need
bluetooth,
don't
load
that
container
and
if
you
do
there,
you
go
and
then
it's
accessible
to
all
the
other
containers
same
concept
of
having
a
database
up
before
you
allow
the
the
other
microservices
that
tap
into
it.
And
so
since
this
thing
is
here,
I'm
going
to
show
a
little
bit
about
hardware
abstractions.
This
is
kind
of
a
cool
trick.
H
So
now
that
I
have
this
up,
I
can
address
it
and
you
can
either
lock
this
down
or
you
can
keep
it
open
like.
I
did
it's
always
at
10
500
for
this
container.
Unless
you
change
it,
I
want
to
take
a
look
at
what
bluetooth
low
energy
devices
are
available
on
this
on
this
node
and
there
we
go-
and
this
is
the
list
I
had
yesterday
in
the
office
was
pages
long
and
clearly
in
my
house.
I
don't
have
as
much
emf
interference
in
my
in
my
living
room.
H
So
these
are
the
devices
that
I
have
sitting
there.
As
you
can
see,
it's
a
json
output-
and
I
don't
know
if
anybody's
ever
done
in
the
group-
has
ever
done.
Bluetooth
low
energy
programming
directly,
but
it
can
be
kind
of
a
pain
in
the
butt.
It's
much
nicer
from
a
edge
microservice.
It's
much
nicer
to
hit
a
rest
api
and
be
able
to
use
that
to
communicate
with
with
bluetooth
stuff.
D
H
So
we
do.
We
do
the
interfacing
with
the
damon
and
there's
there's
a
lot
of
room
here
for
understanding.
You
know
what
people
want
to
do
at
the
edge.
We
occasionally
get
a
request
to
open
something
where
someone
said.
I
can't
do
what
I
could
do
with
with
the
raw
container
controls.
Could
you
you
know
any
plan
to
open
that,
and
of
course,
once
we
hear
that,
there's
there's
a
legit
use
case
for
it.
H
You
know
we
put
it
on
the
road
map
and,
and
ultimately
the
best
is
to
just
represent
the
pure
capabilities
of
the
full
containering
system.
So
that
way,
you
don't
even
have
anything.
That's
missing
question
is
a
security
come
up
of
course,
so
who
should
be
authorized
to
do
that,
because
the
trusted
edge
node
should
stay
trusted
and
not
become
a
hacking
playground,
but
yeah
cool.
H
H
H
And
flow
a
number
five
which
we
just
had
and
let's
set
it
down
there,
we
go
so
that'll,
take
care
of
everything
that
was
that
was
there
and
then
I
have
under
this
flow.
I
have
some
additional
concepts,
so
I
have
some
some
micro
services,
as
well
as
some
some
routes
between
them,
and
so
let's
go
ahead
and
update
that
one
flow
three
and
make
that
one
active
it'll
take
a
moment
for
all
of
that
stuff
to
wind
through,
and
so
we
can
take
a
moment
to
talk
about
data
flows
and
routes.
H
So
there
is
a
deeper
dive
into
security
that
I
think
a
lot
of
folks
in
this
working
group
have
been
wanting
to
do,
and
I
think
it
was
even
that
nicholas
from
nokia
mentioned
it
in
seattle.
At
the
the
face-to-face
meeting,
saying
I'd
really
like
to
expose
some
of
the
security
issues
at
the
edge
and
how
we
intend
to
resolve
them
and
then,
of
course,
preston
you
mentioned
just
last
last
meeting
two
weeks
ago.
H
Is
you
know,
here's
some
things
that
we
have
yet
to
really
kind
of
get
some
some
good,
solid
answers
on,
but
here's
some
starting
ideas,
so
we
have
and
I'm
happy
to
dive
into
it.
As
like
a
separate
working
group,
you
know
meeting
discussion
if
anyone
wants
on
the
agenda,
but
we
have
that's
what
we
call
the
the
eight
requirements
of
secure
commercial
grade
edge
computing
and
some
of
the
concepts
here
are
self-evident,
which
is
like
you
know.
H
You
have
to
be
able
to
know
that
the
software
hasn't
been
tampered
with
and
this
and
that,
but
being
able
to
trust
the
hardware
and
trust
what
software
is
running
on.
What
hardware
is
is
very
common
pattern.
It's
a
I
put
something
out
at
the
edge
and
it
can
be
physically
compromised
in
a
way
that
the
cloud
and
data
center
hardware
typically
can't
a
lot
of
the
infrastructure
that
we
have
for
our
cloud
and
data
center.
H
H
But
how
can
you
be
assured
that
the
data
that
you'd
like
to
move
between
edge
locations
or
from
edge
to
cloud
or
from
cloud
down
to
edge
is
truly
going
where
it
should
go
and
was
authorized
properly
and
so
on?
And
so
to
that
extent
we
have,
if
you'd
like
to
use
the
api,
that's
provided
by
eclipse,
io
fog,
you
can
pass
your
data
via
the
the
distributed
data
bus,
which
uses
the
ecn
to
to
route
everything,
and
that's
that
concept
of
routing
that
I
mentioned.
So,
let's
take
a
look.
H
H
So
here
we
have
a
microservice
that
is
gathering
some
weather
data
and
that's
the
one.
That's
actually
reporting
it
into
the
the
distributed
data
bus
and
if
I
go
and
I
grab
the
status
again,
we're
going
to
see
that
it
continues
to
report
some
data.
And
so
then
we
have
this
other
thing.
It's
called
core
networking
going
along
the
lines
of
deploying
capabilities
dynamically
because
we
essentially
have
the
ability
to
move
software
to
the
edge
on
an
ad
hoc
basis.
Why
don't?
H
We
also
use
that
to
deploy
our
networking
capabilities
so
when
you
think
about
it,
what
you
want
is
an
edge
agent
that
is
as
stable
as
possible
and
allows
you
to
deploy
additional
capabilities
as
if
that
were
the
edge
applications.
Well,
one
great
edge
application
is
the
improvements
to
the
the
the
agent
itself
and
the
agent's
capabilities.
H
It
reminds
me
of
the
old,
the
actual
first
8-bit
nintendo
entertainment
system.
I
don't
know
if
you,
you
guys,
know
this,
but
they
ever
increase
the
capabilities
by
having
the
the
main
board
in
the
thing
do
actually
very
little
other
than
the
interfacing,
with
your
television
screen
and
the
controllers,
and
so
on.
Most
of
the
the
compute
capabilities
came
on
the
cartridges,
and
so,
as
that
system
was
used
over
the
course
of
a
very
long
life
for
a
video
game
system,
it
was
amazing
how
new
new
games
had
new
capabilities.
H
Well,
why
not
do
the
same
thing
with
the
edge
if
it's
really
expensive,
to
deploy
hardware,
you
know
because
think
about
not
only
the
management
but
the
the
actual
rollout
costs
right
if
you've
got
200
vehicles
that
you're
outfitting
with
edge
compute,
if
you
have
to
bring
them
back
in
to
make
improvements
as
it
costs
you
a
lot
of
money.
So
how
about
keeping
the
agent
as
simple
as
possible
and
then
using
the
agent's
capabilities
to
dynamically,
deploy
things
like
internet
working
or
hardware,
abstraction.
H
D
And
the
other
thing
that's
coming,
I
think,
in
that
same
space
is
fpga,
where
colleagues
work
on
that,
where
you
can
basically
say
hey,
which,
which
hardware
interfaces
do
you
have?
You
know
how
many
spy,
how
many
this?
Well,
it
doesn't
matter
what
the
hardware
platform
shipped
with.
If
you
have
fpga
you
just,
can
software
define
new
hardware
interfaces,
basically.
H
H
Is
the
jetson
tx2
for
doing
like
edge
image
recognition,
and
you
know
nlp
stuff,
and
when
you
do
it,
you
just
have
to
have
some
exposure
to
the
onboard
resources
and
it
doesn't
even
really
matter
what
it
ends
up
being
exactly
so
now
we
get
to
that
kind
of
that
universal
concept.
And
then
you
can
allow
people
to
say
well,
look
you've
got
the
infrastructure
out
there,
just
get
that
vendor's
microservice
for
interfacing
or
flash
that
fpga.
That's
perfect.
D
F
D
The
other
is
I'm
guessing
a
lot.
The
answer
may
be,
because
you
just
kind
of
went
in
at
the
beginning
of
this
project
to
kind
of
solve
the
problem,
but
I
looked
at
the
core
networking
stuff
and
it
seems
like
you've
kind
of
reinvented
a
few
things
where
you,
you
know
you're
basically
doing
raw
sockets,
and
you
invented
your
own
binary,
serialization
protocol.
H
It's
like
there
are
things
out
there
that
that
we
may
not
even
be
aware
of
that
are
a
direct
hit
fit
for
what
it
is
that
our
users
are
doing
with
this
and
we're
happy
to
list.
You
know
several
of
the
core
networking
microservices
for
different
implementations,
also
allowing
people
to
take
better
advantage
of
software
to
find
networking
and
all
the
things
that
might
be
important
to
their
deploy.
H
So
I
think
the
the
phrase
that
I
used
when
I,
when
I
first
jumped
into
to
meet
you
guys
and
and
start
talking
about
stuff,
was
that
it's
time
for
eclipse
iofo
to
kind
of
like
grow
up
into
a
bigger
community
and
part
of
that
is,
is
asking
like.
Is
there
something
here
that
we
could
be
leveraging
that
other
people
already
have
in
place
on
that
note
of
service
discovery,
and
I
think
we'll
leave.
You
know
the
last
10
minutes
here
for
the
group
on
that
note
of
service
discovery.
H
We
actually
only
do
hardware
capability
discovery
at
this
time
and
we
don't
expose
running
services
to
the
control
plane
for
you
to
leverage
in
orchestration
and
that's
kind
of
a
first
order
of
business
to
do
a
kubernetes
integration
is
we
know
how
important
that
is,
and
so,
if
I'm
running
a
microservice
on
an
edge
node,
that's
that's
valuable.
For
others.
We
need
to
be
able
to
surface
that
so
that
that
becomes
exposed
as
a
as
an
available
service
for
hardware.
We
do
a
fairly
good
job.
We'll
tell
you
this
hey.
We
have.
H
We
have
some
usb
to
serial
adapters
available
and
stuff,
and
then
you
can,
you
know
this
one's
got
bluetooth
and
that
type
of
stuff,
but
yep.
So
always
a
lot
of
work
to
do,
and
we
are
at
the
the
eclipse
foundation.
We
are
actually
now
expanding
the
you
know
the
pool
of
folks
and
welcoming
in
people
to
drive
this
implementation
forward,
so
that
type
of
input
is
invaluable
to
us,
and
I
really
appreciate
it
cool,
so
I
think
to
wrap
this
up.
H
I
deployed
that
flow,
or
I
activated
that
flow,
and
we've
got
two
over
here
and
three
over
here
and
one
of
the
things
that
that
you
get
with
the
the
core
networking
stuff
that
we've
got
in
place
is
you
can
also
expose
the
the
data
pathways
into
any
of
the
edge
nodes
allowing
you
to
retain
data
at
the
edge
but
still
access
it
remotely.
So
it's
one
of
the
problems
we've
talked
about
at
at
the
edge.
Is
you
want
to
curtail
your
usage
of
bandwidth?
H
You
don't
want
to
transmit
raw
data.
If
really,
what
you
want
is
the
insights
and
the
outputs.
Well,
what
if
you
want
that
raw
data
to
be
part
of
model
training
or
something,
and
maybe
you
want
to
access
it,
and
so
I
have
a
setting
here
on
this
one,
and
you
know
this
is
the
demo,
so
there's
a
50
50
chance
here
that
I
did
something
I'm
not
going
to
fire
up.
H
H
No,
it's
all
good
thanks
for
the
presentation
sure-
and
this
is
about-
like
five
percent
of
what
I'd
like
to
show
you
guys
I
can
talk
for
for
hours
about
this
stuff,
but
so
just
ping
me.
If
anybody
wants
to
like
dive
in
and
don't
feel
bad
about,
pointing
out
something
to
really
like
the
way
you
did.
That
is
actually
really
not
compliant
with
what
we
all
want.
It's
like,
please
bring
it
on,
because
we
need
to
plan
our
road
map
and
you
guys
are
actually
helping
us
make
the
scene.
H
And
I'm
gonna
have
a
look
here:
okay,
cool,
let's
see
I,
and
so
that
was
that
was
that
was
matt.
That
was
you
that
was
just
speaking.
Okay,
great,
I
see
you're.
I
think
the
first
question:
let's
see,
do
you
contemplate,
I
fog
running
only
on
the
the
device
edge,
the
device
side
of
the
last
mile,
or
do
you
see
use
cases
where
it
might
also
run
the
infrastructure
edge
data
center
base
of
the
cell
tower
awesome
question?
H
We
actually
want
to
be
agnostic
as
to
what
location,
but
obviously
what
you
would
choose
to
run
at
the
base
of
cell
tower,
which
we
have
a
customer
by
the
way
that
is
in
just
a
couple
months.
You'll
you'll
see
an
announcement
from
them
about
the
embedding
of
io
fog
in
their
in
their
lte
base
station
product.
Is
you
know
so
it's
you
sh.
We
shouldn't
care
where
we
run
and
you
can
even
run
io
fog
like
inside
the
cloud
and
data
center
environments.
H
H
The
network
configurations
different
latencies
are
different,
but
we
want
you
to
be
able
to
see
the
the
edge
in
the
cloud
as
a
pure
continuum
and
being-
and
you
know
our
goal-
is
how
about
having
kubernetes
able
to
orchestrate
cloud
all
the
way
down
to
the
device
edge
and
every
point
in
between
and
obviously
you
know,
matt.
We.
We
have
a
lot
to
talk
about
with
this.
Maybe
even
some.
F
H
Coming
up
that
we
could,
we
could
check
out
with
that
that
type
of
configuration
yeah
thanks
for
that
answer-
yeah
very
cool,
let's
see,
and
so
then
oh
okay
and
then
and
then
we
had
from
tom
singing
at
drop.
That
would
appreciate
the
the
link-
and
I
don't
see
any
other
questions
in
here.
So
I
guess
the
floor
is
open.
Cindy
did
you?
Did
you
have
one
or
you
just
want
to
make
sure
we
have.
E
H
That's
right,
it's
it's
all
targeted.
This
is
this
is
why
the
merge.
This
is
why
the
integration
of
kubernetes-
okay,
we're
not.
We
have
no
interest
in
replicating
all
the
great
work
and
when
we,
when
we
started
doing
this
kubernetes
was
was
you
know,
existing
was
very,
very
nice
and
stable,
so
yeah,
so
our
whole
goal
is
all
of
that
capability
is,
should
be
actually
driving
the
engine
at
io
fog
instead
of
io
fog
having
its
own
layers.
E
H
E
H
What's
the
best
way
to
actually
manage
a
fleet
of
10
000
vehicles,
for
example,
or
400
oil
pumps
and
commercially
edgeworks
is
producing
great
sas
software
that
you
run
in
the
cloud
and
use
to
drive
your
edge,
and
you
know
it'll
allow
you
to
emerge
in
your
cloud.
So
that's
the
stuff
that
you
know
is
enterprise.
Use
cases
are
go
ahead
and
write
an
engine
on
you
know
for
with
the
ui
and
management
system.
H
That's
one
of
the
products
that
we're
launching,
that
is
a
commercial
product,
but
the
whole
goal
of
the
eclipse,
io
fog
engine,
especially
merging
with
kubernetes,
is
you
can
programmatically
do
stuff
at
scale,
but
you
are
going
to
have
to
be
the
one
scripting
and
we
all
know
that
anyway
about
the
cloud
infrastructures.
You
know
everyone
has
their
own
way
of
controlling
it
and
that's
fits
for
the
use
case.
There's
no
universal
ui
for
all
of
it.
E
H
Yep
yep,
so
we're
gonna,
be
we're
gonna,
be
releasing
first
kubernetes
integration
of
io
fog
in
just
a
couple
of
months,
and
you
know
I
we're
going
to
be
talking
to
you
know
you
and
annie
at
huawei
about
this
anyway
about
having
you
know
some
deeper
work
with
you
guys.
So
the
answer
to
that
is
yes,
and
essentially
you
have
to
for
for
kubernetes
to
get
a
good
grip
on
the
edge
you
have
to.
H
D
When
you
have
your
fleet
right,
you
have
the
you:
have
the
registry
of
nodes.
How
do
you?
How
do
you
approach
integrations?
I
think
that's
one
of
the
things
in
in
iot
in
general
is
that
it's
unavoidable
to
end
up
with
redundant
registries,
because
different
facets
of
the
features
and
capabilities
at
a
minimum
just
need
a
list
of
what
are
my
devices
right?
Do
you
have
you
done
like
one-to-one
mapping
integrations
where,
for
every
node
there's
a
kind
of
a
mirroring
into
another
asset
or
device
inventory
system.
D
H
These
things
on
the
warehouse
shelves
are
this
one
is
in
a7,
and
this
one
is
in
b42,
and
how
do
we
know
which
one
goes
where
so
exposing
those
unique
identifiers
so
that
they
can
be
routed?
You
know
right
and
and
and
you
can
go
grab
the
match,
as
is
the
first
order
of
business,
but
all
of
that
was
done
with
a
microservice
on
an
io
fog
node.
That
was
specially
written
to
do
that
that
that
translation-
and
this
is
that's
the
problem-
that's
going
to
grow
and
get
worse.
H
How
do
you,
and
also,
how
do
you
ship
hardware,
that
has
unique
identifiers
in
it
sim
card,
is
a
great
example
and
be
able
to
know?
Oh,
it's
gone
live,
and
now
I
need
to
allow
it
in
the
system
and
shipping
out
hardware
at
scale.
That's
ready
for
edge
computing,
where
you
basically
just
power
it
on
and
go.
H
Is
we're
just
about
to
send
out
some
some
cameras
that
have
gpus
in
them
where
the
the
identifier
is
is
flashed
in
the
same
way
that
you
flash
in
a
serial
number
into
a
you
know
window
is
when
dell
rolls
out
some
some
computers
and
so
on,
and
this
process
obviously
needs
to
get
faster
and
more
automated
and
possibly
even
allowing
you
to
take
like
kind
of
unidentified
hardware
and
give
it
some
sort
of
identity.
You
know
once
you
expose
it's
it's
underlying
sim
card
and
so
on
all
good
stuff.
H
Some
of
it
right
now
has
been
handled
with
microservices
written
for
the
application,
the
specific
use
case-
and
I
think
some
capabilities
in
the
you
know
the
engine
itself
are
probably
warranted.
Maybe
some
the
abstract
notion
of
of
unique
identifiers
that
are
one
to
one
mapped
and
then
what's
the
external
tie
right
and
how
do
you
reach
out?
How
do
you
bring
it
back?
Do
you
cache
it?
If
you
don't,
you
know
all
of
this
type
of
stuff,
yeah
yeah.
D
H
F
Yeah,
I
I
have
a
couple
questions,
but
I
mean
I
don't
know
you
have
time.
You
said
that
when
you,
when
you
deploy
io
fog,
basically
you
have
a
dependency
on
the
operating
system
right,
so
you
need
an
os
already
existing
in
whatever
an
xbox
that
you're
deploying
your
software.
That
has
access
at
least
to
install
the
agent
right.
So
you
you,
I
owe
it
doesn't
really
help
you
managing
the
lifecycle.
They
always
I
I
guess
it's
the
question
right.
Mm-Hmm.
H
H
What
what
should
we
target-
and
you
know
the
the
answer-
that
question-
was
major
distros
of
linux
using
containers
and
have
it
run
on
on
devices
that
that
have
the
capability
to
to
run
the
host
os
right
out
the
box
now
ever
increasing
we're
being
asked?
Well,
if
you're
there
and
you
do
deal
with
the
life
cycle
of
the
edge
agent,
because
eclipse
io
fog
will
let
you
roll
forward
remotely
and
roll
it
back?
H
What
about
helping
with
the
host
os
and
we
can
tie
into
like
you,
know,
tpm
and
the
secure
boot
and
all
of
this
stuff
to
help
with
you
know
these
are
edge
security
policies,
cool
stuff,
but
what
about
being
able
to
do
maintenance
of
these
remote
nodes?
And
you
don't
want
to
end
up
with
a
wind
river
and
eclipse
io
fog
and
then
saying
well,
the
two
of
them
together.
Give
me
what
I
want.
You
really
just
want
one.
F
Sure
yeah
yeah,
that
was
the
yeah
okay,
sounds
good
and
the
other
question
is
sometimes
we
talk
to
particularly
industrial,
big
factory,
secure
kind
of
bunkers,
they're,
pretty
strict
about
the
policy
of
some
external
service
reaching
out
to
things
that
are
deployed
in
the
factory,
so
that
is
absolutely
forbidden.
Nobody
can
open
anything.
F
Even
vpn
is
kind
of
tricky
and
you
said
that
you
have
the
ability
opens
the
stage
and
that's
a
requirement:
does
your
agent
have
the
possibility
to
open
the
connection
itself
from
the
inside
to
the
outside,
so
you're
establishing
secure
communication
and
then
at
that
point
it's
sort
of
trusted
based
on
search.
So
what
have
you
and
at
that
point
you
already
have
the
channel
to
start
following
whatever
changes
you.
H
Yes,
that's
exactly
the
idea
is
and
then
in
addition,
let's
also
note
that,
if
the
deploy
mandates
that
you
keep,
you
know
information
and
control
all
within
a
certain
building.
You
just
go
ahead
and
put
everything
on
the
land
and
you
can
drive
it
all
from
within
the
factory.
But
if
you
need
outside
remote
control,
then
yes,
you
want
the
agent
to
be
talking
to
the
control
plane
with
a
trusted
connection,
and
then
from
that,
then
you
can
begin
to
actually
move
data
or
to
administration
by
opening
an
ssh
tunnel
and
so
on.
H
H
You
know
commercial
edge
management,
you
know
ui
product,
but
the
the
programmatic
capability
is
already
there
in
the
eclipse,
io
fog
engine,
to
use
the
geographic
location
to
specify
the
boundaries
of
data,
and
that
of
course
means
that
the
microservices
need
to
be
using
the
eclipse,
io
fog.
You
know
data
controls
there,
but
we
can
also
watch
dog
whether
anything
outside
of
that
data
bus
is
being
used
to
transmit.
We
watch
dog,
the
network
and
so
on.
H
So
you
get
some
really
decent
guarantees
and
some
auditability
around
where
data
has
been
allowed
to
flow
geographically
and
has
flowed
outside
of
those
boundaries,
and
you
know,
or
you
can
even
set
policies
like
no
data-
is
allowed
to
enter
this
geo
fence,
and
that
would
be
ways
that
you
can
give
some
customer
assurance
that
the
infrastructure
is
indeed
acknowledging
the
location
awareness.
You
know
challenges
of
the
edge
okay.
G
H
So
yeah
there's.
Let
me
make
sure
that
I
I
understand
exactly
what
you
want
from
inter-process
communication,
but
you
can.
You
can
pass
information
as
kind
of
official
officially
sanctioned
data
between
the
microservices
right
there
and
we
use
a
shared
memory
space
for
for
for
moving
that
between
the
microservices.
H
Yes,
exactly
great,
that
it
is,
does
not
mean,
in
that
case
the
edge
compute
network,
instead
of
being
like
you
know,
across
nodes
which
is
like
you
could
think
of
that,
as
like
micro
vpns
in
this
case,
shared
memory
space
is
really
really
fast,
and
it's
right
there
between
edge
microservices
on
the
same
hardware,
we
determine
whether
those
that
hardware
is
separate
or
the
same,
and
you
don't
have
to
do
anything
with
the
the
io
fog
controller.
You
just
say
I
want
to.
H
What
are
the
latencies
on
that?
It's,
so
we
did
a
tracking.
We
were
doing
a
million
data
messages
per
second,
I
think
of
of
maybe
it
was.
You
know,
1k
per
message
and
latencies
or
sub
sub
20
microseconds
from
from
microservice
to
microservice
good,
but
that
hardware,
I
think,
also
was
like
an
intel
atom
and
maybe
one
gig
of
ram.
So
further
tests
needed
you're
welcome
to
try
to
crack
it
and
see
what
you
get
and
report
back
your
speed
test.
H
I
will
try
that.
Thank
you,
cool
yeah,
okay,
I
think
we're
10
after
and
I
think
for
folks
watching
the
recording.
I
don't
want
to
be
one
of
those
people
that
just
hangs
on
you
can
ping
me
with
any
questions.
Get
me
on
I'm
on
the
slack
for
this
working
group.
You
can
email
me
and
in
fact,
I'll
put
my
email
in
the
the
meeting
notes
next
to
my
name
on
the
attendee
list,
so
it
makes
it
easy
for
everybody
happy
to
talk
more.