►
From YouTube: Kubernetes WG IoT Edge 20230809
Description
August 9, 2023 meeting. Presentation by Liam Randal on WebAssembly & the CNCF sandbox project wasmCloud.
wasmCloud does orchestration compatible with Kubernetes but is not dependent on it - presentation, discussion and Q&A on this can be used for edge use cases.Also discussed advancement of the Edge Native white paper.
A
Hi
welcome
to
the
August
9th
meeting
of
the
cncf
iot
edge
working
group
on
the
agenda
today
we're
going
to
open
with
a
presentation
on
wazen
cloud
as
it
relates
to
Edge.
Then
assuming
there's
time
remaining.
We
also
have
an
item
on
the
agenda
to
to
discuss
the
edge
native
white
paper
and
promotion
of
it
at
the
upcoming
kubecon
North
America.
So
that
said,
I
made
you
co-host,
Liam
I'll.
Let
you
introduce
yourself
and
take
over
great.
B
Hello,
everyone,
my
name,
is
Liam
Randall,
I,
think
I
know
most
of
the
people
on
this
call
at
some
level
and
today
I'm
going
to
be
introducing
cncf,
Blossom
cloud
and
also
talking
about
webassembly
and
some
of
the
neat
use
cases.
We
see
around
our
ecosystem.
Just
a
quick
little
about
me,
so
I've
been
in
the
cloud
native
ecosystem
since
2014,
when
we
kicked
off
with
kubernetes
I,
actually
launched
the
very
first
kubernetes
company.
B
It
was
a
company
called
critical
stack
which
I
sold
to
Capital
One,
where
I
helped
to
work
with
an
amazing
team
to
create
other
cncf
projects
like
Cloud
custodian,
and
you
know,
help
to
move
a
big
Bank
into
the
cloud.
I
love,
webassembly
and
I
genuinely
think
that
it's
the
next
epic
of
tech
and
I'll
try
to
make
that
argument
here
today,
but
I
would
ask
that
everyone
on
this
call
feel
free
to
raise
your
hand,
ask
questions
interrupt
I'm
here,
just
to
build
to
help
communicate
I'm,
not
just
to
speak
at
you.
B
If
you
think
about
the
last
epics
of
tech,
there's
this
pattern
that
we
see
of
continued
higher
abstractions,
where
everyone
looks
around
and
says,
what's
the
common
complexity,
let's
make
that
the
platform
and
build
up
from
there
and
from
where
we
are
with
containers.
Today,
there's
still
a
ton
of
common
complexity,
and
this
was
what
originally
drew
me
to
webassembly
was
simplifying
the
ecosystem
and
the
life
cycle
for
the
management
of
applications
and
early
on
in
the
web
assembly
Journey.
B
B
Sakai
I
had
this
crazy
idea
of
what,
if
we
could
take
modern
languages
like
C,
plus,
plus
and
transpile,
to
JavaScript
and
boom
we're
running
Doom
in
the
browser
and
since
then,
a
number
of
Standards
bodies
have
spun
up,
but
webassembly
is
actually
standardized
by
the
w3c
as
the
fourth
language
for
the
web,
and
it's
really
a
compilation
Target.
This
is
these
are
the
members
of
the
by
code
Alliance,
which
is
an
organization
that
builds
and
helps
to
run
web
assembly.
B
Now,
if
you're
new
to
webassembly,
it
starts
as
with
first
principles,
but
it's
neither
web
nor
assembly.
It
is
a
compilation,
Target
and
I
think
the
best
way
to
think
about.
It
is
as
a
little
tiny
virtual
machine
that
takes
numbers
in
and
numbers
out,
and
that's
it
it's
a
little
CPU
with
no
operating
system,
no
standards,
and
it
runs
everywhere,
you're
already
using
it
today,
as
we
get
into
the
into
presentation,
I'll
try
to
highlight
some
neat
use
cases,
but
it
runs
inside
of
browsers.
B
It
runs
on
side
of
runtimes
for
servers
and
it
can
be
embedded
into
other
applications.
So
it's
been
adopted
all
over
this
cncf
as
the
sort
of
Faz
runtime
of
choice.
It's
safe
and
secure.
It
gives
you
little
tiny,
limited
memory
space,
it's
pretty
performant
and
it
is
a
polyglot,
so
you
can
actually
bring
multiple
languages
down
to
it.
Now
really
underneath
the
hood
webassembly
has
some
challenges,
because
it's
just
three
integers
a
bunch
of
integers
in
a
trench
coat.
B
B
But
what
is
really
powerful
is
that
it
runs
everywhere.
So
organizations
like
Amazon
Prime
video
use
webassembly
to
deliver
their
primary
app.
It
allows
them
to
iterate
faster.
It
allows
them
to
increase
the
performance
of
their
applications
without
deploying
a
full
kit
of
firmware
to
over
8
000,
unique
devices
webassemblies
adopted
in
places
like
Shopify
as
a
configuration
of
runtime
for
Envoy
in
this
studio.
B
So
it's
really
got
a
ton
of
options
because
it
lets
you
give
developers
choice
of
languages
that
they
start
with
and
it's
performant
and
it
runs
seamlessly
across
different
architectures
and
operating
systems.
You
know
it
was
designed
for
the
web
if
it
sounds
familiar,
think
of
what
flash
aspired
to
do
or
what
Java
aspired
to
do
in
the
browser.
B
It
gives
us
an
opportunity
to
start
thinking
about
how
we
might
take
all
of
these
disparate
languages
and
suddenly
build
applications
out
of
composable
Lego
blocks,
and
this
is
where
runtime
and
application
runtime,
like
cncf
wasmcloud,
comes
into
play.
So
what
wasmcloud
aims
to
do
is
to
give
you
a
great
developer
experience
around
adopting
and
running
webassembly
in
your
organization
and
there's
a
lot
of
great
open
source
and
other
runtimes
that
are
around.
B
But
this
one
is
in
the
cncf
and
it's
got
around
150
contributing
companies,
including
big
Banks,
defense
contractors,
Automotive
companies
and
lots
of
other
organizations,
and
it
actually
leverages
a
ton
of
common
standards
like
open,
Telemetry,
cncf,
Nats,
Cloud
events
and
open
application
manifest
in
order
to
sort
of
deliver
a
unified
experience.
That's
familiar
to
users
in
the
cncf,
but
I'm
certainly
certainly
extends
the
ecosystem.
There's
a
separate.
If
you
go
to
waslamcloud.com,
there's
a
slot.
B
We
do
community
meetings
on
Wednesday,
There's,
YouTube
channels
with
a
couple
years
of
meetings
on
it,
I
mean
it's
generally,
a
pretty
friendly
community.
In
fact,
our
our
meeting
is
at
one
o'clock
today
right
after
this
meeting.
If
anybody
wants
to
hop
on
and
join
now,
webassemblies
will
be
mentioned
earlier
was
just
a
CPU,
so
there's
a
standard,
that's
called
Wazi
webassembly
system
interface
that
lets
us
Define
building
blocks
and
these
building
blocks
are
organized
into
things
that
are
called
worlds.
C
B
B
That
we
call
the
web
assembly
component
model,
and
the
neat
thing
about
components
is:
is
that
these
contracts
or
apis
are
defined
and
standardized
and
the
components
that
you
use
to
deliver
them
can
be
implemented
in
multiple
languages,
more
even
with
different
implementation
types,
because
webassembly
is
a
Sandbox
you're
oftentimes
implementing
a
contract
like
Wazi
SQL
might
have
a
plug-in
for
MySQL
for
postgres
or
for
Azure
cosmodb.
Those
are
example,
components
that
come
in
wasmcloud
and
you
can
then
hot
swap
between
those
without
updating
your
application.
B
So
what
we're
trying
to
do
is
to
address
this
compile
time
problem
that
containers
suffer
from
when
we
build
containers,
we
actually
take
all
of
their
dependencies,
their
non-functional
requirements,
and
we
embed
them
at
compile
time.
What
webassembly
enables
to
do
is
to
load
these
various
dependencies
at
runtime,
and
what
that
means
is
is
that
platforms
can
load
the
most
recent
version
at
runtime,
giving
you
the
latest
the
latest
most
up-to-date
version,
so
you're
not
embedding
of
vulnerabilities
a
thank
you
Kate
for
putting
the
wassi
clear
link
into
chat
there.
B
That's
called
wit,
a
web
assembly
interface
types.
So
you
start
with
your
API
declaration
and
then
they're
pluggable
from
there
now
I'm
going
to
take
a
quick
diversion
here
into
the
webassembly
roadmap,
which
was
which
was
just
published
by
Bailey.
Hayes
Bailey
is
a
director
at
cosmonic
and
she
is
also
the
user
developer
or
the
developer
representative
to
the
board
of
directors,
the
bytecode
alliance.
So
we
just
published
this
three-tier
road
map.
That
I
can
point
you
at
there's
a
great
blog
here
on
the
bycode
alliance
website.
B
If
we
think
of
webassembly
components
as
the
as
the
sort
of
new
as
the
new
container,
the
new
abstraction
components
can
be
created
by
Common,
Language,
tooling
and
then
be
used
to
create
other
components.
But
what's
neat
about
these
components
is:
is
that
just
as
as
containers
were
portable
across
different
runtimes
and
implementations,
components
are
also
portable
across
different
runtimes
and
implementations.
So,
if
you're
building
components
for
cncf
wasm
cloud,
those
components
could
work.
For
example,
at
Kate
golden
rings
company,
which
is
a
fermion.
B
They
have
an
open
source,
runtime
called
spin,
or
they
could
work,
I
mean
theoretically,
even
on
other
webassembly
users,
such
as
fastly
or
cloudflare,
who
are
using
webassembly
to
deliver
their
experiences
without
speaking
to
their
product
roadmaps.
That's
the
intention
of
where
we're
going
with
these
components.
So
these
webassembly
components
are
portable,
they're
pluggable
and
they
run
everywhere
and
can
be
made
from
different
languages
as
well
as
linked
together.
So
building
applications
now
becomes
this
effort
in
just
giving
developers
the
smallest
piece
of
code
to
work
on
as
possible.
B
B
All
X
teams
need
to
stop
and
Patch
those
vulnerabilities,
and
there
are
many
techniques
to
try
to
automate
that,
but
imagine
a
world
where
we
can,
instead
of
having
5
000
teams
patch
5000
apps,
where
we
can
have
one
team
patch,
one
component.
That
then
is
adopted
and
pulled
in
at
runtime.
This
is
a
huge
opportunity
for
cloud
native
and
was
the
primary
reason
that
I
saw
this
as
the
biggest
opportunity
in
Tech
the
most
common
complexity.
B
That
we're
looking
to
abstract
away
from
now
I,
don't
think
that
it's
fair
to
compare
and
say
that
this
conversation
is
excuse
me.
You
know
webassembly
components
versus
containers
in
the
same
way
that
it's
not
fair,
to
compare
containers
to
Virtual
machines
or
virtual
machines
to
real
Hardware.
These
things
all
work
together
in
and
around
Cloud
native,
and
they
essentially
offer
you
different
value
propositions
for
where
they
can
go.
What
capabilities
they
have?
How
reusable
are
they
so
just
to
do
a
quick
comparison
between
containers
and
webassembly
components?
B
Components
are
compatible
with
everything
that
came
before.
There
are
multiple
ways
to
run
components
and
webassembly.
On
top
of
kubernetes,
you
can
put
it
inside
of
a
container
there's.
Even
a
native
implementation.
That's
looking
to
try
to
orchestrate
a
web
assembly
directly
in
kubernetes
or
things
like
wasmcloud,
allow
you
to
run
on
kubernetes
or
directly
on
Tiny
Edge
devices,
which
is
where
we
brought
that
up
in
this
particular
conversation.
B
Now
one
of
the
advantages
to
Containers
is
that
we
were
able
to
lift
and
shift
most
of
our
application
right
from
Linux
and
just
put
them
in
a
container
and
run
them
on
someone
else's
application.
While
that
can
sometimes
happen
with
webassembling
components,
I
view
webassembly
components
as
more
of
a
green
field
where
you're
building
new
applications
and
able
to
bring
in
libraries
and
Frameworks,
but
oftentimes
you're
thinking
about
how
to
re-architect
to
take
best
advantage
of
the
unique
properties
of
webassembly
that
we
have
here.
B
So
what
is
Blossom
Cloud
really
bring
to
webassembly
what
wasmcloud
gives
you
is
not
only
a
developer
experience
around
starting
running
managing
and
stopping
these
things,
but
it
also
introduces
the
sort
of
common
components
that
we
think
Enterprises
want.
So
awesome,
Cloud
signs
all
the
web
assembly
artifacts,
it
signs
the
invocation
requests
and
it
enables
you
to
invoke
and
operate
webassembly
modules
across
disparate
infrastructure,
underneath
the
hood
it
simply
embeds
Nas.
But
from
a
user
point
of
view,
it
means
that
you
can
use
capabilities
regardless
of
whether
running
in
a
adjacent,
webassembly
module
or
not.
B
And
this
is
very
powerful
because
it
suddenly
means
that
you
can
build
applications
that
are
self-forming
that
are
self-healing,
that
run
across
clouds
or
edges
all
without
modifying
your
applications.
So
it's
we
view
awesome
Cloud
as
distributed
webassembly,
and
the
reason
why
we
think
that's
important
is
any
application
that
matters
is
going
to
have
properties
like
highly
available
reliable
and
in
Watson
Cloud.
These
become
knobs
that
you
can
turn
which
we'll
see
in
the
demo
now
underneath
the
hood,
a
web
assembly,
Watson
Cloud
uses
a
manifest
that
uses
oam
to
Define
an
application.
B
So
let
me
pull
up
an
example
here.
So
if
we
look
at
an
example,
OEM
Manifest,
this
will
look
very
familiar
to
kubernetes
people.
You
can
define
an
application
with
a
version
and
then
describe
the
webassembly
components
that
you
want
to
use
to
pull
these
together
and
a
webassembly
component
can
have
requirements.
B
So
here
we
have
components
that
have
requirements
on
on
SQL,
for
example,
and
I
can
say
that
I
want
to
also
load
the
progress
module
and
then
here
I'm,
giving
a
manual
definition
for
what
is
the
link
definition
for
that
a
particular
module.
Where
does
that
component
connect
to,
but
these
could
be
loaded
into
stores
as
well?
Just
like
you
can
with
kubernetes
is
supposed
to
be
a
simple
example,
and
then
common
components
like
web
servers
and
database
components
can
then
be
loaded
underneath
the
hood.
B
These
components
live
in
oci,
just
like
we
do
with
containers,
I
mean
we
work
very
closely
with
Docker
and
with
Microsoft
and
other
Cloud
providers
who
are
aligned
and
participants
in
the
greater
webassembly
ecosystem
to
ensure
this
Better
Together
experience.
If
I
could
describe
the
relationship
between
you
know,
Cloud
native
and
webassembly
as
a
whole,
it's
very
collaborative,
and
we
really
aim
to
be
this.
B
B
And
the
second
thing
is:
is
that
in
wasm
Cloud
we
run
what
are
called
lattices
and
a
lattice
is
essentially
a
flattened
topology
for
an
application,
and
you
could
have
multiple
lattices
that
each
are
scoped
to
their
own
unique
permissions
domain.
So
think
of
them.
You
know
very
similar
to
what
you
might
have
with
pods
on
kubernetes.
B
Yes,
you
know
wasm
itself
is
you
know
similar
to
plain
text,
so
it's
so.
There
are
multiple
tools:
there's
a
great
tool
called
exodism
and
some
other
things
in
the
ecosystem
to
allow
you
to
scan
webassembly
modules
and
even
see
a
was.
Does
this
consist
of
other
components
because
it
can
be
Turtles
all
the
way
down?
You
can
have
components
and
components.
B
Victor
asked
a
question:
does
the
cosmotic
control
plane
run
on
kubernetes
a
cosmotic
itself,
as
a
company
does
not
run
on
top
of
kubernetes?
For
our
back
end,
we
actually
use
a
Hashi
Nomad
in
order
to
orchestrate
and
run
our
webassembly.
We
just
released
a
very
popular
plugin
that
brings
psyllium
to
that
ecosystem
and
that
world
called
net
reap,
which
is
a
command
that
lets
you
use,
psyllium
and
all
the
amazing
stuff
that
you
can
do
with
edpf
outside
of
kubernetes.
B
Now
we
are
compatible
with
kubernetes
you
can
of
on
wasm
cloud.
We
actually
ship
a
Helm
chart
that
you
can
spin
up
that
will
deploy
wasm
Cloud
across
a
kubernetes
cluster
and
within
cosmonic
itself,
there's
an
even
easier
way
to
do
that.
What
cosmotic
aims
to
provide
is
orchestrated,
webassembly
and
that's
the
company
I
work
for
a
question.
B
This
registry
is
called
Ward.
There
is
an
early
version
of
it.
What
Ward
provides
is
component
metadata,
so
you
can
query
it
and
say
give
me
all
of
the
components
that
match
Wazi
SQL
and
it
has
a
labeling
system.
So
you
could.
Even
you
know,
you
know,
query
for
other
additional
properties
similar
to
how
you
can
query
for
containers
and
the
underlying
storage
for
those
is
on
oci.
B
Shan
I
would
actually
push
back
a
little
bit
on
the
dependency
nightmare
here
and
let
me
share
my
experience
when
I
was
working
at
Capital
One.
This
was
what
I
did
decided
was
our
biggest
problem.
I'm
in
Tech
yeah
I
was
I,
got
promoted
as
VP
of
innovation,
I
own
the
open
source
office,
and
my
boss
said.
B
Congratulations,
you
own,
you
have
a
million
vulnerabilities,
it
wasn't
that
number,
but
it
was
a
really
large
number
and
when
I
dug
into
that
problem,
I
couldn't
believe
it
and
what
the
the
sort
of
Source
the
problem
was
was
that
we
tried
to
bootstrap
our
developer
experience
by
giving
them
a
goal.
Golden
template,
and
this
golden
template
was
a
great
way
to
get
started.
I
mean-
and
it
essentially
was
an
application
template
that
had
all
of
the
the
Enterprise
goo
that
we
needed
in
order
to
build,
run
and
scalar
applications.
B
Logging,
metrics,
open,
Telemetry
all
of
those
things,
but
as
we
distributed
that
across
thousands
of
applications,
we
were
really
suffering
from
multiple
problems.
We
had
some
ability
to
Auto
patch
and
auto
upgrade
people,
but
essentially
the
developers
would
typically
need
to
get
in
the
loop
to
patch
things
like
log4j,
so
by
Distributing,
unique
copies
of
each
app.
We
essentially
created
a
bigger
nightmare
than
the
ability
to
centrally
manage
individual
components
and
I
see
this
as
the
sort
of
goal
of
organizations
that
are
adopting
things
like
service
mesh
or
platform
engineering.
B
B
You
know
that
I
think
is
the
real
goal
for
what
we
want
and
when
we
look
at
how
cloudflare
and
fastly
or
platform
engineering
really
Embrace
is
they
try
to
let
the
platform
provide
the
latest
copies
of
everything
at
runtime
versus
compile
time
and
webassembly
has
some
unique
properties
that
I
think
make
this
even
better.
Webassembly
is
so
small
that
it
has
a
cold
a
cold
start
time.
That's
oftentimes,
measured
in
milliseconds
Kate.
B
You
gave
that
awesome
talk
at
kubecon,
EU,
wasm
Day,
where
you
scaled,
10,
000
web
assembly
modules
in
just
a
few
seconds,
is
that
right,
yeah.
It
was
a
great
talk
if
you'd
like
to
see
the
implications
of
that.
What
that
means
is
that
you
can
think
about
starting
your
web
assembly
modules
per
in
vacation
versus
having
a
bunch
of
warm
containers
or
a
bunch
of
warm
lambdas
here,
waiting
to
service
requests,
Shan
did
I
answer
your
question
or
maybe
speak
to
your
topic
of
dependencies.
B
Absolutely
thanks
a
lot
great
question.
By
the
way.
Thank
you
so
I'm
going
to
do
some
demos
here
that
I
hope
were
a
little
interactive
and
I'm
actually
going
to
use
cosmotic
to
run
them,
which
is
our
commercial.
A
platform
for
the
company
that
I
work
for
everything
I'm
doing
is
compatible
with
wasm
Cloud,
underneath
the
hood
and
what
cosmotic
does
is
just
gives
you,
the
batteries
included
experience
so
I'll
try
to
draw
out
the
differences
here.
So
this
UI
is
what
comes
with
our
product.
B
The
wasm
cloud
is,
you
know,
command
line
and
Via
yaml,
but
I
want
to
be
able
to
to
help
people
visualize
what's
happening,
underneath
the
hood
here
with
a
few
simple
apps.
So
here
I've
got
an
app.
That
is
very
simple:
it's
an
XKCD
generator
and
if
people
on
the
call
will
maybe
grab
your
phones
and
spin,
this
up
we're
going
to
use,
wasn't
Cloud
to
run
this
application
across
disparate
infrastructure,
including
kubernetes
I'm,
all
without
modifying
it
very
quickly.
B
Okay,
I
think
somebody
has
it
so
what
I'm
going
to
do
is
I'm
simply
just
going
to
delete
this
now
underneath
the
hood.
This
is
not
being
managed,
kubernetes
or
just
like
kubernetes
was
in
Cloud,
does
support
reconciliation
controllers,
so
you
can
sort
of
manage
State,
and
if
you
refresh
your
screen,
your
app
should
now
be
broken.
You
should
get
a
cloud
flare
app,
so
Alex.
If
you
just
pull
down
that,
should
fail
for
you
now.
B
Let
me
show
you
the
infrastructure
that
I've
spun
up
here
before
a
meeting.
What
I
did
is
I,
went
ahead
and
spun
up
Blossom
Cloud
running
on
different
operating
systems
and
in
different
clouds
running
all
across
the
country.
So
here
I've
got
this
host
called
bouncing
rain.
That's
running
on
my
local
MacBook
I've
got
a
host
in
Microsoft
Azure
running
on
x86
Linux
down
in
Texas
I've
got
a
host
on
Oracle
Cloud
that
is
running
on
arm
60
virtual
arm.
64.
B
I've
got
a
couple
kubernetes
nodes
here
in
the
middle
that
are
running
on
a
kubernetes
cluster
I've
got
a
host
running
on
cosmotic
cloud
in
AWS,
East
and
I've
got
a
host
running
another
host
here
running
on
my
local
MacBook,
and
this
is
also
on
my
MacBook
here
now
to
spin
these
up
in
wasmcloud.
You
would
have
to
run
Nats
yourself
centrally
and
then
spin
up
Blossom
cloud
and
point
them
at
Nats,
and
then
you
would
have
all
these
things.
B
Working
in
the
open
source
cosmotic
makes
that
a
little
bit
easier,
but
we
can
easily
spin
up
our
applications
I'm
using
wasmcloud
on
any
clouder
Edge.
So
here
somebody
wants
to
just
chime
in.
We
can
use
labels
in
order
to
schedule.
This
I'll
maybe
run
this
on
Oracle
Cloud.
Somebody
wants
me
to
run
it
anywhere.
Different
I
can
go
ahead
and
do
that
I
mean
now.
We've
spun
this
up
on
Oracle
and
if
we
refresh
our
app
thing,
should
work
now,
that's
great,
but
what?
B
But
normally
I
would
do
this
with
a
yaml,
a
spec,
and
that
was
what
was
that
was
that
bouncing
rain
was
that
my
Local
Host
I,
don't
even
remember
it
matters
and
it
and
it
doesn't,
and
it
doesn't
matter,
but
now
what
will
happen
underneath
the
hood
is
Knox
will
route
these
requests
to
the
closest
matching
provider
and
if
I
were
to
come
into
my
infrastructure
and
delete
that
bouncing
rain
host,
even
without
making
any
changes,
this
application
just
self-forms
and
self-heals.
So
underneath
the
hood.
B
This
is
an
abstraction
that
was
on
cloud
provides
that
lifts
us.
A
bit
higher
than
what
we
find
in
kubernetes
and
that's
the
goal
is
to
take
the
common
complexity
of
application
libraries
and
enable
that
to
be
the
platform
that
we're
building
on
itself
and
our
ecosystem
has
has
really
blossomed,
not
just
web
assembly,
but
laws
and
Cloud
in
particular.
I'm
going
to
do
maybe
one
more
demo.
B
B
It's
got
a
neat
little
or
basic
little
UI,
that
I
can
upload
pictures
to,
and
then
this
application
can
route
to
different
machine
learning
models
depending
on
if
I
care
about
privacy,
performance
availability,
regulatory
domain
or
whatever
the
case
would
be
now.
If
you're
not
familiar
with
how
machine
learning
models
work.
Essentially,
what
happens
is
images
need
to
come
in
and
get
processed
and
that
basically
will
take
a
raw
image?
B
And
then
you
know
size
it
to
to
the
vector,
create
first,
take
the
image
and
standardize
it
so
that
it
matches
what
the
machine
learning
model
has
been
trained
on.
So
that
could
be
size,
color
balance
number
of
bit
depth
all
of
those
sorts
of
properties,
and
then
it
essentially
turns
it
into
an
output
Vector.
So
this
output
Vector
has
a
neat
property
to
it.
If
we
think
of
this
raw
image
as
being
a
MAG,
these
vectors
are
typically
much
smaller
than
that.
So
there
is
a
measure
of
compression
that's
happening
here.
B
So
if
we're
running
these
preprocessors
on
the
edge,
we
have
the
ability
to
minimize
our
Ingress
egress
charges
and
increase
performance.
As
we
move
these
things
around,
we
can
also
run
our
machine
learning
models
in
places
to
protect
privacy
or
in
places
to
respect
regulatory
domain,
for
example
around
what
data
can
move
in
and
out
of
countries
all
right.
So
enough
talking.
B
You
know,
picture
of
hot
dog
and
it's
pretty
sure
it's
a
hot
dog
I
could
submit
maybe
a
cute
little
kitten
and
it
thinks
it's
a
44
tiger.
Now,
if
I
wanted
more
accuracy,
I
could
submit
this
to
a
resnet,
which
is
a
larger
model.
That's
around
250
megabytes
and
now
the
model
will
route
those
same
requests
to
the
higher
Fidelity
model.
So
here
espresso
is
labeled
as
99.5
espresso
and
what's
powerful.
B
You
know
Vector
to
the
machine,
learning,
inference,
API
and
then
return
results.
So
the
the
distributed
properties
of
wasmcloud
mean
that
your
basic
application,
topology
can
have
radically
different
implementations.
I
could
develop
as
everything
running
on
my
Macbook
without
changing
my
code.
I
can
simply
deploy
this
across
multiple
clouds.
B
Without
changing
any
of
my
code,
I
could
run
this
across
multiple
edges,
different
operating
systems,
different
CPUs
and
the
sort
of
end
result
that
we're
really
aiming
for
is
this
experience
where
developers
can
write
the
smallest
piece
of
code
possible
and
then
operate
it
across
any
any
infrastructure.
Shanna
a
great
question:
what's
the
typical
size
of
a
webassembly
module
in
production?
Similarly,
what
is
the
size
of
application
that
have
all
things
was
on
components?
B
A
B
A
B
Yep,
so
different
languages
will
result
in
different
size,
webassembly
modules
due
to
how
those
languages
work
so
static.
Languages
like
C,
C,
plus,
plus
and
rust,
tend
to
produce
webassembly
modules
that
are
very
tiny.
Languages
like
dot
net,
go
and
python
produce
modules
for
a
little
bit
larger
because
they
bring
The
Interpreter
or
the
garbage
collector
with
them.
B
We
published
a
case
study
with
Adobe,
where
we
went
through
an
example
where
Adobe
was
running
some
containerized
Java
that
they
re-implemented
in
Rust
and
what's
powerful
about
this
particular
demo,
is
the
applications
were
good
in
wasmcloud
they
were
the
same
algorithm.
They
were
really
tiny.
They
went
from
consuming
two
gig
of
memory,
for
instance,
down
to
something
like
150
kilobytes
per
instance,
and
they
were
so
small
that
they
could
even
be
operated,
sometimes
within
the
in
the
user's
browser.
B
So
they
were
not
only
good,
but
they
were
fast
and
cheap
here
and
what
that
meant
was
in
most
cases
instead
of
uploading.
These
images
that
then
have
the
background
removed
for
the
cloud
English
egress
and
waiting
the
450
milliseconds.
For
that
to
happen,
all
of
this
work
could
take
place
right
in
the
customer's
browser,
and
that
meant
that
it
took
place
in
40,
milliseconds
and
Adobe
was
just
paying
for
the
loading
the
module
in
the
customer's
browser
versus
pulling
all
the
you
know,
megabytes
of
images
up
that
needed
to
have
their
background
removed.
B
It's
a
powerful
use
case,
so
I
think
to
make
sure
that
I,
Circle
and
answer
your
question
is
the
size
of
the
webassembly
module
depends
on
the
source
language
and
the
Frameworks
that
are
compiled
in
in
there,
but
generally
it
is
orders
of
magnitude
smaller
than
what
we
find
in
a
containerized
world
and
you
and
that's
typically,
because
even
with
containers,
it's
very
rare
that
people
are
doing
from
scratch.
You
know
like
with
a
go
or
something
sham
did
I
answer
your
question.
F
Yeah,
in
a
way,
that's
I
think
it
makes
sense.
I
don't
know
one
related
question
to
that
was
so
there's
no
concept
of
having
like
the
OTA
part
right.
I
was
curious
about.
If
there
are
things
out
there,
where
people
can,
you
know
you
could
just
ship
out
an
OTA
was
a
module
Somewhere
Over
The
Air
to
a
very
Remote
device.
F
That's
supposed
to
do
this,
or
maybe
the
same
point
where
you
mentioned
like
I'm,
patching
the
the
logging
logging
component
bottom
component,
or
something
and
I
don't
have
the
let's
say
the
luxury
of
being
connected
to
a
centralized
registry,
or
something
so
assuming.
If
the
modules
are
small
enough,
are
they
are
there
some
specs
out
there
where
it
can
be
like
a
Well,
Suited
solution
for
over-the-air,
stuff,
yeah.
B
That's
called
webassembly
vert,
which
allows
you
to
virtualize
resources,
but
at
the
web
assembly
layer
we're
primarily
dealing
with
the
you
know
the
standards
and
then
the
developer,
experience
kind
of
move
out
to
the
different
open
source
projects
like
Blossom
cloud
or
wasm,
Edge
or
spin
and
and
those
ecosystems
to
manage
it.
Frank.
You
asked
a
great
question
around
can
on
the
AIML
example,
can
you
make
use
of
the
GPU
capabilities?
Absolutely
one
of
the
I
think
neat
properties
of
Blossom
cloud.
B
Is
we
enable
you
to
load
components
that
are
external
to
webassembly?
So
if
we
think
about
webassembly
as
this
tiny
virtual
machine,
it
doesn't
have
support
yeah
CPU
view
extensions.
So
one
of
the
properties
that
we
do
support,
though,
is
you
can
load
an
external
provider.
So
on
our
example
that
we
did
here,
let
me
switch
back.
B
B
Yes,
we
have
many
customers
in
the
Watson
Cloud
ecosystem
that
specifically
do
use
us
in
air
gapped
situations.
You
obviously
run
into
the
challenges
of
how
do
I
update
that
air
gap
system.
The
powerful
thing
about
how
awesome
Cloud
works,
though,
is
because
the
components
are
distributed
here.
You
have
the
ability
to
download
components
across
this
lattice
that
is
running
underneath
the
hood
here.
B
A
B
E
C
E
B
C
A
C
E
E
There's
some
places
either
in
intros
or
outros,
where
there
could
be
mention
of
the
project
or
of
the
particulars
of
of
the
interaction
we've
had,
and
so
those
things
seem
amenable
and
if
there's
any
other
places
where
we
could
do
some
kind
of
short
intro.
That
would
be
fantastic,
too.
I,
don't
know
that
anybody
wants
to
sit
necessarily
and
hear
us
drone
on
about
what
are
the
exact
principles
but
I
think
giving
some
awareness
as
to
hey.
This
is
the
next
tranche
of
guidelines
coming
out
of
the
iot
edge.
A
So
I
think
there's
two
elements
involved
here
right
now.
This
thing
is
still
kind
of
in
discussion
graph
stage,
not
that
that's
a
problem,
but
we
really
need
to
Aspire
to
have
this
be
wrapped
up
and
published
by
the
time
of
kubecon,
because
I
think
it's
a
lot
better
to
promote
it
as
kind
of
a
published
already
reviewed
document,
then
still
saying
it's
a
work
in
progress.
So
let's
try
to
push
this
forward.
A
A
I
think
the
closing
address
always
has
had
where
to
go,
to
learn
more
where
we
talk
about
this
group
and
it
would
be
a
natural
fit
there,
but
until
the
actual
speaker
is
appointed
to
deliver
the
closing
address,
you
know,
I,
don't
feel
comfortable,
In
Absentia,
cramming
it
down
some
yet
to
be
determined
speaker's
throat.
It's
up
to
them.
A
E
C
E
Here,
collectively,
for
several
weeks
now,
so
you
know,
I
can't
I
can't
push
any
more
than
what
we've
already
done.
So
the
fact
that
we
missed
some
deadlines,
nothing
really
I-
can
do
with
that
I
I've
attempted
to
do
it
right.
So
within
the
confines
of
this
meeting.
The
second
thing
is
yes,
I,
agree,
I,
don't
want
to
recapitulate
all
the
data
and
information.
E
You
know
more
interaction
post,
the
conference
right,
so
I,
don't
I
and,
lastly,
the
restrictions
on
how
many
presentations
one
can
give
at
coupon
or
nothing
that
we
can
control
so
of
of
which
I
know
that
at
least
Frank
and
I
are
kind
of
fully
booked
at
this
point
at
the
two
presentation
limit,
so
it
would
have
to
be.
You
know
some
other
method
of
delivery,
probably
not
by
either
of
us
and
I.
Think
that
that
exit
discussion
or
the
closing
remarks
would
be
a
good
place
for
it.
A
D
D
So
just
a
was
a
call
of
action
for
maybe
Andy
or
Frank
or
I
would
be
willing
to
put
it
up
for
us
to
put
up
a
GitHub
issue,
and
that
basically
says
like
this
is
white
paper.
We're
working
on
here
is
the
Google
document
with
the
draft,
and
then
we
can
kind
of
have
subtasks,
maybe
like
Brandon.
You've
done
a
really
good
job
of
working
with
the
cncf
marketing
team
to
get
that
into
a
published
state.
So
that
could
be
a
call
to
action.
D
And,
yes,
it
is
disappointing
that
we
didn't
get
it
into
a
presentable
form
for
kubecon,
but
having
a
published
paper
might
be
something
that
we
can
then
transfer
our
energies
to
and
then
maybe
for
the
next
kubecon.
It
would
be
cool
to
have
a
talk
that
not
only
presents
the
content
but
also
gives
that
demonstration
of
an
application
using
these
principles
to
kind
of
really
land.
It.
E
Yeah
well
so
I
would
respond
to
that.
Just
so,
yes,
I
can
take
the
ownership
of
building
a
GitHub
issue,
I
think
much
the
same
I.
It
was
my
expectation
that
much
the
same
as
many
of
the
presentations
that
we
put
together
for
kubecon
or
any
other
conference
is
that
you
wait
for
the
opportunity
to
be
accepted
and
then
put
together
the
the
materials
to
back
that
up.
D
Not
at
all
no
I'm
just
trying
to
put
a
silver
lining
on
things
of
saying
we
maybe
can
put
effort
towards
getting
that
published
version
and
focusing
our
efforts
there.
Instead
of
splitting
it
into
presentation
and
getting
the
published
paper
out,
but
no
last
time
we
certainly
didn't
have
it
completed
before
the
conference.
So
I
would
not
say
that
is
the
standard.
Okay.
C
Maybe
a
question
to
Brendan
real
quick.
If
I
can
go
and
show
him
in
Brendan,
how
long
does
it
take
to
go
and
process
the
thing?
So
if
we
assume
that
we
would
rather
rub
it,
sorry
wrap
it
up
in
the
next
two
weeks
say:
how
long
would
it
take
to
go
and
do
the
processing
the
beautification
to
go
and
ultimately
get
it
published
as
a
PDF.
G
On
the
last
one,
we
did
have
it
in
near
final
draft,
State
I
think
ahead
of
the
last
event
where
we
presented
it
and
it
did
make
it
in
as
a
talk
around
the
topic
and
part
of
the
kind
of
the
call
to
action
afterwards
and
linking
to
the
paper
was
this.
Is
here
we'd
love
your
input,
come
read
it
and
give
us
that
feedback.
G
We
allowed
a
certain
amount
of
time
for
that
to
come
in,
and
then
we
felt
good
about
hitting
the
finalize
button
from
there
pros
and
cons
of
making
it
finals
sooner
rather
than
later
and
waiting.
That's,
maybe
something
for
us
to
to
figure
out,
but
definitely
there's
there's
still
an
opportunity
to
to
share
this,
this
goodness
with
the
community
and
build
on
it
from
here.
C
G
A
G
A
E
So
I
mean
I'm,
just
I'm,
sorry
I
know
I'm
I'm
coming
off
this
way,
but
that's
my
intent.
I'm
a
little
bit
frustrated
I
mean
we've
I'm,
not
gonna.
Do
the
spelled
milk
thing
I'll
just
make
mention
of
it.
We've
been
bringing
this
up
for
several
weeks
and
we've
hit
numerous
barriers,
and
now
this
information
comes
to
light
that
it
wouldn't
have
been
a
barrier.
E
You
know
where
I'm
confused,
you
know,
I
just
I,
don't
I,
don't
think
this
is
in
a
state
that
it
can't
be
released
to
the
public
and
people
can
have
a
chance
to
review
it.
So
that's
first
and
second,
you
know
if
we,
if
we
could
have
had
the
opportunity
to
do
it,
we
should
have
and
I'm
just
don't
know
why
we
waited
that's
all.
D
So
it
sounds
like
there's
clear
frustration
about
a
lot,
a
lack
of
facilitation
from
the
maintainers
of
this
working
group
and
getting
this
talk
as
our
maintainer
track
talk
for
this
working
group
and
that's
something
that
clearly
fell
through
the
cracks.
I've
been
out
a
lot
of
the
summer.
So
I
wasn't
at
the
past
several
meetings
to
initiate
that.
So,
if
you
want
acknowledgment
of
that,
then
I'll
acknowledge
that
I
didn't
help
move
that
forward.
The.
D
So
that's
something
that
we
can
kind
of
be
sure
to
evangelize
more
in
the
working
group
is
that
this
working
group,
when
it's
opportunities
for
speaking,
come
it's
not
just
for
a
maintainer
track
session,
but
it's
also
for
those
Edge
day
submissions
and
helping
facilitate
those
happen.
So
I
think
we
can
definitely
do
a
better
job
of
that.
I'll
personally
try
and
do
a
better
job
of
that
of
trying
to
create
opportunities
in
this
working
group
to
present
at
Edge
days
and
reminding
people
of
that,
but
yeah.
A
C
A
E
B
I
I
will
also
I'll,
also
say
this
Andy.
It's
it's
that
I
think
that
there
has
been
we're
all
volunteers
and
I.
Don't
think
that
oftentimes
it's
that
we
maybe
get
communicated
the
same
set
of
rules
and
I
I
hear
that
the
frustration
in
that
I
felt
that,
as
a
maintain
as
a
somebody
that
volunteers
to
run
a
day
and
that's
a
different
day
than
the
than
the
day
that
Steve
is
working
on.
B
So
maybe
Steve's
team
was
told
that
there
that
it
does
apply
when
we
were
when
we
asked
we
were
told
they
didn't
think
it
applied.
But
but
I
understand
that
this
is
all
contextual
and
I,
also
always
believe
in
positive
intent
and
I
know
everybody
is
doing
the
best
they
can
to
try
to
embrace
the
Better
Together
positivity
of
the
cncf
and
the
community
aspect
and
move
forward.
E
Misinterpret
my
comments,
my
my
comments
are
are
surely
this
is
that
they're,
just
based
on
you
know,
what's
been
going
on
the
workflow
the
factual
day
by
day
week
by
week,
and
that
it
just
it's
a
retrospect
if
anything
and
listen,
I'm
I
receive
this
all
the
time
from
my
open
source,
Community
right
on
occasion
when
things
happen,
so
all
it
is
is
a
reflection
and
it's
not
I'm,
not
walking
away
from
here
punching
a
wall.
It's
not
happening.
You
know
so
yeah.
A
C
What
would
I
what?
What
I
would
would
appreciate
is
that
we
wrap
up
the
paper
as
the
kind
of
draft
version
one
in
the
next
two
weeks.
So
everybody
please
go
and
throw
in
your
comments.
We'll
go
get
them
in,
so
that
we
have
a
version
for
pre-publishment
ready
for
Brendan
to
process
and
we
can
go
and
progress
that
way
so
I
think
then
at
least
we
have
something
out
and
we
can
at
least
see
like
maybe
in
the
closing
node.
C
A
Here's
an
idea
if
you
wanted
a
forum
at
the
event,
it
is
sort
of
running
a
sidebar
thing,
but
in
the
past,
at
kubecon's
we've
had
physical
working
group
meet
and
greets
where
the
idea
was
that,
if
there's
a
number
of
people
from
this
group
who
are
maybe
even
just
casual
people
interested
in
the
general
topic
that
we
just
declare
a
meeting
place
like
some
bar
or
restaurant.
You
know
in
this
case,
in
Chicago
and
just
meet
there
and
I'd
say
about
two-thirds
of
the
time
we
had.
A
Some
corporate
sponsors
stand
up
and
pick
up
the
tab
or
a
couple
of
them
pick
up
the
tab,
but
sometimes
even
that
didn't
happen
and
everybody's
on
their
own.
But
I
found
those
historically
to
be
quite
useful,
where
you
know
you
sort
of
enjoy
meeting
people
on
these
zooms,
but
it's
a
little
different
from
a
physical
meeting
and
if
a
number
of
us
are
physically
giving
talks
at
kubecon,
North
America,
we
could
have
such
an
event
and
present
on
the
white
paper
at
this
venue.
We've
done
that
before
too
I
think
the
original
one
was.
A
The
original
unveiling
of
few
badge
happened
at
one
of
these
in
at
the
time,
Huawei
was
the
one
who
sponsored
the
event,
but
whether
we
get
a
sponsor
or
not,
I
think
we
could
do
that
and
publish
this
going
into
kubecon
and
maybe
build
a
you
know.
I
we've
we've
had
audiences
that
are
up
in
the
range
of
30
some
people
at
past,
coupons,
which
you
know,
isn't
bad
or
a
specialized
interest
area.
E
A
Be
plenty
of
opportunities
to
just
pick
an
ad
hoc
place
to
have
you
know
an
informal
meeting.
It
wouldn't
appear
on
the
official
kubecon
schedule,
but
I
think
that
we
can
get
our
word
and
we
could
mention
it
during
kubernetes
on
edge
day
and
a
few
other
times,
so
that
people
are
aware
that
it
exists.
Yeah.
E
I
I
was
just
there
four
weeks
ago,
so
I'm
really
familiar
with
the
with
the
area
and
I'm
sure.
A
lot
of
people
here
have
traveled
there.
Recently
too,
so
I
I'd
be
comfortable
with
that
as
an
option.
I
think
it's
also
good.
As
you
know,
honorable
mention
in
the
closing
as
well
I
mean
again.
The
goal
is
to
just
get
the
the
final
polish
on
the
on
the
paper.
I
think,
yeah
and
I
think
I
think
Frank
brings
up.
E
A
Sounds
good
to
me:
okay,
cool,
okay,
we're
at
time.
So
let's
call
this
meeting
to
a
close,
we'll
have
another
meeting
in
two
weeks.
I
think
the
agenda
is
still
well
well,
it
isn't
even
in
the
agenda
node
stock,
but
if
anybody
wants
to
put
something
on
there,
I'll
open
a
space
in
the
agenda,
node
stock
for
it
and,
as
usual
members
can
just
self-nominate
for
topics
they
want
to
cover.
A
If
we're
trying
to
close
out
this
document,
that
sounds
like
a
worthy
thing
to
put
on
the
agenda,
and
maybe
we
could
even
aspire
to
having
it
done
there.
I
won't
be
available
for
the
meeting
in
two
weeks,
but
I
think
Kate
has
said
that
she
thinks
she
will
be
so
we'll
see
you
in
a
couple
of
weeks.
Thanks
for
coming
everybody
thanks.