►
From YouTube: Kubernetes WG IoT Edge 20230614
Description
June 14, 2023 meeting of the CNCF IoT Edge working group. Open discussion: Understanding “Where the Edge is?” and key implications, what are roles and interests of the people applying edge native?. Review (3rd, near final) of draft of Edge Native Principals Paper
A
Hi
welcome
to
the
June
14th
meeting
of
the
cncf
iot
edge
working
group
on
the
agenda
today,
but
probably
in
the
second
half
of
this
meeting,
we
have
a
discussion
slated
on
this
Edge
native
principles
paper
the
agenda
it's
described
as
part
three,
and
this
is
a
supplemental
or
different
paper
from
the
first
one.
We
were
working
on
starting
over
a
year
ago,
but
addresses
slightly
different
topics
when
we
get
to
that
I.
A
So
as
usual
in
this
group,
if
we
don't
have
agenda
items,
we
open
it
up
for
birds
of
a
feather
discussion
or
anybody
who
joined
is
free
to
add
things
to
the
agenda,
even
late
like
at
the
start
of
the
meeting
and
I'll
drop
by
a
second
link
to
that
agenda.
Notes
document
in
the
chat
before
we
formally
started.
A
Somebody
here
brought
up
a
question
for
the
group
as
to
the
actual
roles
and
background
of
the
people
participating
in
the
call,
so
I'll
invite
you
to
go
and
now
that
we're
officially
started
and
recorded.
Why
don't
you
go
put
that
question
on
the
table
again
and
we'll
see
what
we
get.
B
Yeah
thanks
Steve
first,
so
this
stems
from
a
very
simple
logic
that
what
what
is
generally
the
profile
for
someone
who's
working
on
and
on
it
within
the
edge
native
space?
Is
it
purely
someone
who's
purely
development
development,
engineer
software
engineer
or
is
it
a
nishmash
of
lot
of
devops
things
as
well?
Maybe
an
operations
engineer
or
even
some
bits
of
industrial
engineering
comes
into
play,
because
it
might
be
quite
interesting
to
see
how
this
Edge
native
landscape
pans
out
through
different
sectors.
These
days,
foreign.
B
So
in
general,
I
said:
I
could
start
from
myself.
I
generally
am
more
into
software
architecture,
most
of
the
times.
That
also
leads
to
a
lot
of
I
would
say
more
or
less
system
operations,
because
a
lot
of
tools
that
are
being
used
in
the
industrial
iot
space
are
much
more,
let's
say,
low
code
or
no
code.
So
it's
very
rare
that
you
get
to
write
code
these
days,
because
everything
is
pretty
much
standardized
from
the
operations
technology.
B
A
How
about
if
we
just
go
around
the
table
and
pull
people
for
whatever
it's
worth,
and
if
you
feel
that
you're
just
a
lurker
and
not
prepared
to
answer
you
can
abstain,
but
I'm
just
going
to
go
in
the
list
of
what
shows
up
on
my
participants
list.
So
Brandon.
Do
you
want
to
throw
something
in
there
either
your
personal
experience
or
what
you've
observed
as
a
vendor
or
somebody
in
the
field.
C
Sure
so,
I'm
Brandon
Wick
I
head
up
marketing
for
RNA
networks,
we're
a
software
startup
in
the
edge
orchestration
and
5G
space,
and
our
interest
in
participation
is
just
because
we
see
a
tremendous
amount
of
growth
and
development
with
the
edge
and
Edge
infrastructure.
And
we
want
to
be
a
leading
contributors
and
participants
in
that
we
have
a
team
of
software
Engineers,
Architects
and
developers
that
follow
this
group
kind
of
through
me.
C
So
I'm
kind
of
acting
as
a
bridge
between
those
between
as
representative
of
the
company
and
the
cncf
community,
and
it's
also
I
think
just
in
our
interest
to
produce
documents
like
we're
working
on
now
that
establishes
baselines
for
the
industry
around
important
Concepts,
like
Cloud
native,
so
sort
of
a
project
manager.
D
E
C
Help
facilitate
the
development
but
as
a
long
time,
open
source
Community
participant
definitely
see
that
communities
Thrive
best
when
there
is
a
diversity
of
different
types
of
roles
and
and
of
companies
in
the
ecosystem
that
all
sort
of
get
a
voice.
So
one.
E
C
F
Yeah
yeah,
so
yeah
I
consider
myself
a
software
engineer
right,
but
looking
at
threat,
it
is
an
organization
I
see
yeah,
mostly
software
engineers
and
and
people
already
involved
into
all
kind
of
infrastructure
engineering,
so
working
on
on
Rail
and
openshift,
mostly
evolving
into
this
I'm,
also
as
I'm
coming
from
like
a
middleware
side
of
things,
also
interested
in
in
the
like
event,
driven
architectures
and
and
generally
application,
architectures
related
to
to
Edge
and
and
what
new
type
of
architectures
and
particles
we
need
to
to
solve.
G
S,
it
sounds
big
proponent
of
Open
Source,
my
I'm
in
a
post
sales
or
a
customer
experience,
part
of
Cisco
semi-focused
day-to-day
with
customers
is
predominantly
I,
guess
what
we
call
a
devops
or
infrastructure
up
to
the
NSS
admin
for
the
compute
that
gets
to
the
actual
application,
so
I'd
say
devops
infrastructure
and
connectivity.
G
E
G
Have
that
experience
the
day-to-day
I'm
more
focused
on
the
conductivity
application,
networking
and
devops.
A
Hey
George
I
haven't
seen
you
in
a
long
time,
but
why
don't
you
go
ahead
and
since
I
think
you
have
either
new
to
this
group
or
haven't
been
active
for
a
while,
introduce
yourself
as
well.
H
Oh
hi,
everyone,
I'm
George,
Castro
I,
recently
joined
the
cncf
as
a
developer.
Advocate
and
I
am
literally
going
to
every
tag
meeting
this
week
to
meet
people
see
what
groups
people
are
doing
and
kind
of
be
available
for
projects,
and
things
like
that.
It's
literally
my
third
day
so
I'm,
mostly
just
here,
to
learn
and
listen
to
see
what
what
people
are
into
and
see
how
I
can
help.
So,
okay,.
A
A
Historically,
it
started
as
the
CNC
or
the
kubernetes
iot
edge
working
group,
and
it
was
people
trying
to
address
Edge
use
cases
which
can
be
all
over
the
map
as
we're
discussing
now
with
kubernetes
it
kind
of
morphed
to
where
the
observation
was
that
people
were
starting
to
include
that
in
many
of
these
edge
cases,
maybe
kubernetes
wasn't
the
best
or
most
appropriate
tool,
or
it
was
a
collateral
thing
with
mostly
other
things
being
the
things
that
were
more
worthy
of
being
talked
about
and
discussed
and
evaluated,
and
it's
kind
of,
oh
because
it
had
that
more
open
end.
A
The
nature
it
transitioned
to
the
cncf,
where
we're
free
to
talk
about
even
pretty
much
any
open
source
project
in
the
cncf
landscape,
as
well
as
LF
Edge
projects,
Eclipse
Foundation
projects.
Sure
rule
here
is
that
we
don't
want
it
to
be
commercial
promos.
But
if
it's
open
source,
it's
fair
game
if
if
it
applies
at
the
edge,
so
that's
what
we're
aspiring
to
be
about
here.
Awesome.
A
Okay,
next
Julio,
maybe
that's
not
how
you
pronounce
it
if
so
correct
me,
but.
I
I
I
I
am
an
architect
for
Dish
Wireless
and
my
role
is
some
responsible
for
the
design
of
the
virtualization
and
kubernetes
platforms
for
the
5G
Network
that
we
are
deploying.
I
A
D
J
I
created
one
of
the
first
kubernetes
companies
in
2014
and
I.
Really,
all
of
my
you
know.
Work
has
been
around
open
course,
since
the
90s
open
sourcing
stuff
for
Novell
312
networks
and
I
really
work
on
the
Better
Together
story
with
kubernetes
and
webassembly.
J
A
Okay,
thanks
through
my
name
list
here,
Philip.
K
Hi
everyone
so
Philip,
Griffis,
I,
look
after
outbound
product
and
evangelism
for
net
Foundry
Company
behind
open
ZT,
which
allows
us
to
apply
zero
trust
Network
into
basically
any
use
case
being
open
source,
whether
it's
Edge
iot,
you
know
multi-cloud
Etc,
so
we
can
completely
abstract
away
complexity
of
underlying
networks
and
not
worry
about.
You
know:
public
DNS,
vpns,
inbound
ports,
Etc.
K
My
interest
is
actually
working
on
iteration
three
of
the
weight
paper
and
how
we
can
apply
zero
trust,
networking
Concepts
to
Edge
native
environments,
in
the
same
way
that
we
are
at
the
moment
with
Ajax,
which
is
one
of
the
large
projects
within
the
the
LF
Edge,
so
no
foundation
for
the
edge
yeah.
That's
me.
L
L
Hi
everyone
prakash
here
I'm
at
San,
Francisco
area,
Bay,
Area
and
I-
am
part
of
the
HTTP
age
working
group
coach
here
and
I
have
been
in
the
age
for
almost
a
dozen
years
now,
and
currently,
I
am
a
co-founder
and
for
emerging,
open,
Tech,
Foundation
out
of
Mumbai
India
and
stay
here
in
Bay
Area
in
Silicon
Valley.
So
we
also
I'm
part
of
the
ccici
cloud
Council
of
India,
where
we
had
a
zero
trust,
a
security
for
the
cloud
and
so
conversion
with
lot
of
areas.
L
I
also
contributed
earlier
and
verified
various
the
Linux
Foundation,
as
well
as
openstack
and
various
places.
So
at
this
time,
I'm
just
coming
back
after
a
long
time.
So
I
don't
know
what
was
happening
here
so
apology
for
that
and
just
try
to
review
what
is
going
on
and
hopefully,
if
I
can,
we
have
any
help
to
the
organization.
Thank.
M
Yeah
so
hi
everyone,
so
I
am
done
edge
with
Leota
project.
There
was
from
VMware
and
World
Service,
so
I
basically
work
as
a
software
engineer.
Currently
I'm
a
bit
I
have
not
worked
recently
on
the
edge
groups
and
also
joined
the
group
too,
because
it's
an
interest
platform
when
I
have
to
so
that's
why
I
joined
the
group
to
follow
along
and
learn
from
your
guys
and
also
contribute
to
whatever
the
card.
So
thanks.
Okay,.
A
N
Sure
it's
my
pleasure.
My
name
is
Rob
bully
I
work
in
the
city
of
office
at
Wind
River,
some
of
the
things
that
I
do
involve
a
lot
of
PowerPoint
slides,
but
also
doing
technical,
perfect
Concepts,
so
that
could
involve
easing
small
microcontrollers
with
Zephyr
and
other
rtos's
in
using
kubernetes
to
orchestrate
workloads
to
those
microcontrollers.
O
I'm
like
true
about
actually
a
database
consultant
for
20
plus
years,
so
I've
been
actively
in
the
open
source
Community
for
the
past
several
months.
What
area
I'm
focusing
is
security,
however,
I
believe
the
the
I'm
interested
in
Edge,
because
just
there's
a
lot
of
things
going
on
in
Edge
and
I
believe
Edge
is
more
of
a
for
me
at
least
it's
kind
of
a
more
of
a
location
of
the
application.
O
Instead
of
you
know,
what
do
you
do
there
so
before
I
jump
into
database,
I
was
actually
a
physicist
in
from
a
National
Lab
doing
900
product
physics,
and
there
we
actually,
whenever
there's
a,
we
actually
work
on
every
aspect
from
Hardware
software
physics
analysis.
So
to
me
it's
yeah,
I
guess:
I
have
the
luxury
of
do
not
have
to
make
a
living
on
edge
Computing.
So
so
that's
for
me
I'm
trying
to
learn
from
every
aspect
of
it,
and
you
know
whatever
is
interesting
again.
O
A
Okay,
I
guess
I'll
go
there
and
I
guess
maybe
I'll
even
segue
from
your
question
Victor
of
what
is
Edge-
and
this
is
just
based
on
my
observations.
Working
in
the
field
that
is
self-described
as
Edge.
First
of
all,
I
work
for
VMware
and
I'm.
Also
I've
been
a
co-chair
of
this
group
for
years
and
prior
to
VMware,
I
worked
for
what
was
then
a
startup.
A
I
was
one
of
the
founders
of
the
company
wonderwear
that
did
a
toolkit
for
Industrial
Automation,
originally
mainly
user
interface
stuff,
but
it
morphed
and
did
Acquisitions
getting
into
control
historical
data
logging
and
you
name
it.
My
observation
of
edge
is
that
it's
always
been
a
conflicted
self-description
of
some
of
them
determined
by
vocation
some
of
it.
The
location
is
based
on
physical
location,
some
based
on
network
latencies
and
connectivity.
There
are
some
definitions
that
declare
it
to
be
a
space
with
low
resource
and
I.
A
Think
the
common
thing
is
it's
stuff
not
running
in
a
public.
Cloud
I
think
that
there's
pretty
much
Universal
agreement
that
that
description
would
apply
to
Edge
that
it's
not
running
in
a
public
Cloud.
But
at
that
point-
and
it
probably
also
isn't
the
mezzanine
level
if
you've
got
a
big
enough
organization
that
you're
running
tears,
it's
sort
of
the
stuff
that
Network
wise
is
out
on
Leaf
nodes
and
beyond
that
no
one's
ever
really
came
come
up
with
a
universal
description.
A
If
anything,
it's
forked
into
things
like
you
know,
device,
Edge,
retail,
Edge
and
I
think
that
kind
of
working
is
fair,
because
people
in
these
different
categories
have
different
priorities
of
what's
important
to
them
in
terms
of
Shan's
original
question
of
who
what
is
typical
of
the
players?
A
I
think
you
know
whether
they
be
actual
Hands-On
software,
Engineers
or
more
Ops
people
I
think
it
varies
dramatically
by
use
case
in
my
observation
and
from
a
project
perspective
I've
over
decades
seen
a
lot
of
organizations
that
really
rely
on
outside
vendors,
or
you
know
whether
they
be
shrink,
wrap
software,
vendors
or
consulting
firms
to
do
kind
of
the
initial
architecture
and
planning
design
or
at
least
contribute
to
it,
because
those
are
short-term
roles
and
often
the
life
cycle
of
these
things
is
measured
in
decades,
so
that
unless
they're,
a
really
large
organization,
that's
continually
deploying
a
growing
number
of
edge
outposts.
A
That's
a
short-term
need
and
they
choose
to
Outsource
it.
So
they
don't
necessarily
have
people
permanently
on
staff
to
do
some
of
these
roles
and
over
time
you
know
they
have
internal
people
who
operate
these
things,
but
they
weren't
necessarily
the
ones
who
did
the
original
planning
and
organizations
that
point
to
the
extent
possible
people
often
try
to
rely
on
I'll
call
them
shrink,
wrap,
Solutions,
but
shrink
wrap.
A
Solutions
are
tough
in
this
field,
because
every
Edge
thing
tends
to
be
a
little
different
so
that
you
do
need
somebody
to
get
involved
with
those
kind
of
customization
aspects
and
in
terms
of
what
percentage
is
custom
versus
done
by
your
off-the-shelf
stuff.
Sometimes
the
custom
part
is
10
times
the
size
in
terms
of
engineering
hours
involved
of
what
you
were
able
to
get
from
shrink
wrap
anyway,
I'll.
O
Hand,
that's
me
again
so
yeah
I
still
so
so.
For
me,
it's
still
it's
the
location
that
I'd
like
like
to
understand
it.
For
example,
if
you're
in
an
AWS
data
center,
you
know
you
even
reading
the
data
center,
but
you're
using
5G
technology,
that's
probably
still
considered
Edge
right.
So,
in
the
other
hand,
if
you're
having
a
full
power
cloud
data
center
on
the
moon
on
Mars,
isn't
that
that
also
Edge
Computing
yeah.
A
I
when
I
said
Cloud
I
said
public
Cloud,
meaning
you're,
like
you
rent
it,
you
know
there.
There
is
an
aspect,
arguably,
if
you
can
put
it
out
a
cluster
of
servers,
maybe
as
small
as
three
you
know
three
Intel
nooks
and
turn
them
into
a
kubernetes
cluster,
and
that
is
a
cloud.
So
I
specifically
put
the
adage
public
Cloud
there,
as.
L
I,
don't
know
yes,
so
let
me
let
me
take
this
one.
It
is
depends
on
the
priority
of
what
you
are
getting
out
of
the
age.
If
you
are
talking
about
iot,
that's
one
thing:
IIT
can
happen
for
weather
bureaus.
Also,
it's
not
necessarily
that
it
is
private.
It
can
be
public.
Also,
so
I
don't
think
that
it's
just
private.
L
It
depends
on.
So
some
of
the
priorities
of
some
of
the
applications
are
latencies
and
if
the
latency
is
the
criteria
then
closer
to
the
user,
that
is
the
definition
for
it
for
those
priorities
and
other
definitions
are
like
example:
if
I
have
a
car
and
I
have
to
automatic
driving,
that's
an
edge,
it's
a
moving
Edge
same
way.
L
We
can
have
a
satellite
and
if
you
are
having
a
what
you
call
satellite
moving
and
then
you
are
trying
to
have
something
related
to
satellite
for
communication
terrestrial
to
satellite
that
might
be
Edge
for
them
for
drones.
Etc
drones
are
even
high
performance,
you
call
it
as
Aviation
vehicles
and
all
that,
so
it's
ages
very
relative
term
for
iot.
Of
course
we
try
to
have
a
Gateway,
but
there
is
no
need
for
a
Gateway.
It
can
be
direct.
Also,
it
is
all
pervading
you
can
Define
as
you
want.
There
is
no.
L
As
long
as
you
know.
What
is
the
use
case
we
can
say
this
is
the
HR
not
the
edge
in
automation,
automated
training?
Of
course
automation.
You
want
that
car
is
The
Edge.
The
onboard
car
has
got
its
own
processing
to
take
care
of
directions.
L
L
If
you
are
aitka,
you
always
want
a
iot
if
you
are
doing
AIML
and
all
GPT
chats
and
chat
gpts,
and
all
that,
then
you
want
something
so
that
you
can
offload
your
work
from
Edge
to
the
cloud
so
that
computation
is
reduced
at
the.
So
there
is
no
such
thing
as
you
have
to
Define
it.
You
can
go
by
use
cases
whatever
the
use
cases,
if
you
find
that
that
improves
whatever
your
criteria
is
based
on
that
you
can
Define
The
Edge.
L
Now,
then
the
question
is:
what
is
the
difference
between
engine
cloud?
There
are
models,
Cloud
can
control
the
Edge.
Edge
can
be
self
self-managed,
it
can
be
part
of
the
cloud
or
so
there
are
many
many
models
peer-to-peer
different
ways
of
looking
at
it,
but
if,
since
we
I
think
this
was
we
are
dealing
with,
iot
I
would
stick
to
that.
Iot
is
a
gateway
at
the
edge
yeah.
Thank
you,
I
hope.
I
have
made
my
point.
O
It's
skill
set
for
depend
on
the
type
of
location.
The
skill
set
will
be
quite
different,
though
right
just
like
you
said,
if
you,
if
you
develop
application
for
iot
or
for
automobile,
it's
quite
different
for
someone
building
a
data
center
on
the
moon
right,
because
that
that
is
totally
different.
Skill
set.
F
You
have
your
hand
raised
yeah
yeah,
in
my
opinion,
this
example
of
the
you
know
Cloud
on
the
moon.
To
me
it
would
be
more
multi-cloud
or
hybrid,
hybrid
Cloud
use
case
than
the
edge
I
see
that
edge
needs
to
have
some
core
cloud
in
in
in
in
in
in
that
matter,
so
it
is
yeah
distinction
is,
is
is
a
little
bit
blurry,
but,
to
my
opinion,
Edge
is
is
the
edge
of
the
cloud.
So
so
there
needs
to
be
something
yeah
that
score
cloud
in
this
use
case.
I.
L
L
So
if
you
look
with
respect
to
cloud
multi-cloud
and
hybrid
cloud-
and
they
are
the
cloud-
but
when
you
talk
about
the
edge
can
be
any
of
the
any
of
these
operating
Cloud
operating
systems,
so
age
doesn't
is
not
determined
by
either
hybrid
cloud
or
multi-cloud.
L
It's
a
computational
computational
data
center
type
of
place
connected
with
the
network
to
some
either
peer-to-peer
or
whatever
models
you
use
with
cloud
or
with
not
with
Cloud
cell.
It
is
a
smaller
place,
smaller
computational
unit,
which
allows
the
application
or
the
users
to
be
enabled
for
the
criteria
they
are
looking
for.
So
it's
not
necessarily
the
Big
Data
Center
hyper
skill
data
centers,
where
you
have
got
100
thousands
of
device,
compute
servers,
rather
a
small
generally.
L
We
say
it
is
small
means
it
can
be
half
a
server
one
server
to
10
server
type
of
capability.
That
is
what
we
call
as
Edge
by
any
definition.
So
and
and
iot
is
too
small
for
that
right.
Even
we
don't
even
go
to
that
level.
We
just
say
it
is
aggregation
point
whereby
you
are
able
to
deliver
this
Services
of
the
IIT
at
the
age.
A
It's
just
they've
narrowed
it
down,
but
there
are
categories
and
subcategories
and
maybe
we'll
leave
it
at
that
for
now
Brandon
do
you
want
to
go
and
be
host
of
the
the
yeah
white
paper
craft
discussion,
yeah.
D
C
I'd
welcome
you
to
open
the
dock
in
the
chat
to
have
it
as
a
reference
and
I'll
also
share
my
screen.
So
a
couple
of
the
authors
couldn't
make
it
today,
I
think
we're
missing
Frank
and
her.
We
do
have
Andy
and
Joel
here.
So
we
went
through
this
paper
on
the
previous
call.
We've
now
touched
on
it
a
couple
times,
I
think
our
goal
here
is
to
try
to
bring
this
to
completion
and
to
finalize
this
draft.
For
now
we
can
leave
some
time.
C
You
know
after
we
round
up
here
for
any
final
edits
or
comments
from
the
group,
but
we're
trying
to
bring
this
to
a
close
here
soon.
So
we
can
move
this
into
the
design
phase.
There
are
a
couple
of
comments
outstanding.
C
We
can
talk
through
those,
but
first
I
think
we
want
to
talk
about
what
has
changed
from
the
last
version
that
we
went
through
and
I
think
at
least
stroll
has
made
a
contribution
around
diagram.
So
Joel
do
you
want
to
maybe
start,
and
then
we
can
go
to
Andy.
E
G
E
G
So
this,
if
you
stay
right
here
for
a
moment
Brandon,
what
I've
done
is
this
section
here
almost
gets
into
the
deployments.
G
So,
at
the
at
the
end,
I've
taken
there
there's
the
table
that
gets
into
tiers
and
layers.
So
now
Brandon.
If
you
go
down
to
the
bottom,
I,
simply
added
diagrams
that
are
additional
slides
in
the
edge
native
sketches.
G
Further
up
there's
a
bold
text,
it
says
deployment
scenarios
they're
right
there,
so
that
right
there
was
so
this
this
could
be
earlier
in
the
document.
This
could
go
to
the
end
or
we
could
choose
not
to
use.
It
depends
on
you.
G
An
element
scope,
creepier,
based
on
the
spirit
of
behaviors
for
this
document,
but
this
I
reinserted
the
for
the
previous
discussion
or
for
the
purpose
of
this
paper,
and
this
discussion.
E
G
K
G
Edge
SP
centralized
Data
Center
I
and
if
you
can
scroll
down.
G
E
G
H
G
Camera
a
smart
device
or
on-prem
Edge
with
a
sample
app
and
then
a
centralized
data
center.
So
that's
how
that
sample
application
Maps
to
those
definitions
and
then,
if
you
go
down
further,
the
progression
here
is
here's
a
single
tier.
It
could
be
a
standalone
app
on
an
edge
device
and
then
it
just
builds
from
there
there's
an
example
of
a
single
tier
or
a
single
layer.
G
When
you
go
down
further
that
just
builds
to
okay,
you
have
a
single
edge,
but
the
antennas,
you're
gonna
have
tens
hundreds
or
many,
and
then
that's
a
two-tier
or
two
layer
that
was
in
that
table
and
then
below.
That
is
a
three
tier
which
generally,
if
you're
gonna
scale
large
enough,
you
may
need
to
have
an
SP
Edge
or
an
intermediary
to
be
able
to
support
that.
E
E
G
Far,
the
visualization
I
didn't
put
what
a
Content
or
description
of
that
because
I
don't
know
if
that
goes
into
a
follow-on
paper.
Or
do
we
want
to
add
that
here.
P
Yeah,
thanks
for
adding
this
in
here,
Joel,
so
I
think
you're,
right,
I.
Think
in
terms
of
the
scope
of
the
paper,
as
is
I,
mean
I,
think
alluding
to
the
idea
that
there
could
be
intermediary
or
more
than
just
Hub
and
spoke
and
introducing
the
idea
of
a
hierarchy
here
might
be
useful,
but
not
necessarily
giving
any
more
of
the
the
details
necessarily
other
than
to
say
that
it
could
be
an
option.
P
I
think
I
think
it
does
probably
fall
outside
the
scope
here,
because
it
can
start
to
bring
in
other
I,
think
ramifications
and
other
concerns,
like
things
like
how
to
secure
the
and
make
sure
that
the
multi-tenancy
and
the
isolation
of
the
workloads
ETC
are
are
insured,
so
I
think
a
lot
of
those
kinds
of
larger
concerns
start
to
creep
in
once.
You
start
talking
about
more
than
just
Hub
and
scope.
P
This
is
more
of
a
I
think
this
for
what
it's
worth
and
it
being
an
edge
native
principles
or
application
best
best
practices
how
to
design
and
Implement
one
I
think
this
is
good
the
way
it
is
now.
You
know
without
getting
too
much
further
into
the
details
of
of
what
it
would
be
like
to
introduce
more
than
just
Hub
and
spoke,
but
that's
just
my
opinion.
C
Andy
can
I
ask
if
you
think
it's
more
appropriate
to
list
these
diagrams
sort
of
at
the
bottom
as
a
sort
of
an
appendix
as
it
is
or
up
in
the
body
text?
Are
they
too
long?
Do
they?
Would
it
break
up
the
flow?
Should
we
keep
them
down
here
separately,
yeah
I'm.
P
Looking
at
the
at
it
online-
and
you
know
it
just
flashed
away,
how
did
that
happen?
Zoom
is
taken
over
control
of
my
desktop
here.
Okay,
yeah
I'm,
looking
at
these
that
are
more
towards
the
bottom
I
think
we
we
got
in
past
that
vcat.
The
vcat
probably
is
the
last
scenario
that
that
one
there
that
has
the.
P
G
P
G
C
G
With
that,
it
could
be
your
next
white
paper.
D
G
G
Limbs
of
hey
here's
user
St,
centralized
Data
Center
and
he
had
diagrams
that
should
hey
here's
where
kubernetes
here's,
where
k3s
here's
Blossom
I,
think
that
would
be
great.
You
know
it's
like
we
could.
We
could
end
it
at
the
sample
app
and
then
just
go
right
into
that's
one
Edge
deployment
scenarios
and
then
the
other
paper
is
that
you
alluded
to
is
zero
trust,
Edge,
but
I
think
Phillip
had
to
drop.
G
The
only
thing
I
would
say:
Andy
potentially
is
a
suggestion
is
take
that
the
bottom
there,
the
LF
Edge
Continuum
I,
think
is
the
title
in
the
original
paper.
If
we
just
inserted
that
we
had
hey
here,
are
considerations:
hey
here's,
the
LF,
Edge
definition,
your.
P
I
think
that's
right,
I
think
he's
showing
how
the
deployment
maps
on
top
of
the
of
the
suggested
you
know
how
so
you're
giving
you
know
when
you.
P
If
we
scroll
up
a
little
bit
here,
you're
giving
out
you
know
what
are
the,
how
did
the
principles
map
in
to
a
subscribed
architecture,
and
then
you
make
it
a
little
bit
I
think
you
connect
it
back
to
what
the
terminology
was
where
this
was
already
linked
to
from
the
previous
paper
and
I
think
that
that
gives
the
the
reader
the
connective
tissue
that
they
need
to
say.
P
You
know
this
is
a
coordinated
effort
from
The
iot
Edge
community
and
that
the
the
documents
align
and
that
this
is
a
natural
extension
I
think
going
any
further
than
that
right
starts
to
bleed
into
you,
know,
new
territory
and
something
we
would
have
to
spend
more
time
linking
it
back.
So
I
tend
to
agree
that
I
think
that
once
you
get
done
with
that
last
diagram
on
the
LF
Edge,
the
next,
the
the
the
the
tiering
discussions
single
multi
Etc,
start
to
get
into
new
territory
and
I
think
they.
P
E
C
B
Brendan
Andy
I
was
just
curious
about
the
the
specification
of
the
word
volume.
Would
it
is
it
very
specific
to
the
like
some
of
the
tools
that
we
use
I
mean?
On
the
other
hand,
you
have
something
like
a
persistent
storage
service.
What
what
difference
would
it
make
when
you
have
a
mention
of
volumes
in
the
diagram
and
as
far
as
I
just
did
a
quick
search?
There's
only
one
mention
of
what
a
volume
is
over
here.
P
P
Yeah,
if
you
look
up
a
little
further
in
the
principles
and
I'm
just
scrolling
just
to
there
is
a
discussion
point
about
data
storage
right
is
that
you
know
the
the
addition
of
either
the
having.
If
you
have
constraints
on
the
size
of
of
whatever
the
volume
is
in
this
case,
so
I
guess
you
could
say,
volume
is
interchangeable,
with
data
storage
and
and
the
size
of
the
volume
makes
the
differences
to
you
know
whether
metrics
or
logs
will
be
kept
locally,
whether
to
be
rotated
on
some
regular
Cadence
or
they'll.
P
Be
you
know
if
they
can't
be
shipped
off
and
removed?
Entirely
same
is
true
for
things
like
considerations
for
bandwidth,
caching
or
buffering
Etc
et
cetera.
So
the
idea
of
having
the
volume
located
inside
of
the
smart
device
Edge
is
the
is
just
to
give
you
the
play
that
it
could
be
if
we
have
it
on
the
edge.
P
You
have
considerations
that
have
to
be
taken
into
account
and
if
you'd,
rather
than
be
on
the
cloud
in
a
persistence
and
persistent
storage
device,
then
there
are,
you
know
and
you're
not
going
to
use
a
volume
locally.
Then
there's
other
ramifications
for
that
right,
so
you
have
you
can't,
have
you
have
to
have
constant
connectivity?
So
if
we're
talking
about
things
like
Edge
devices
that
can't
always
have
connectivity,
they
have
to
do
some
degree
of
storing
forward
and
so
store
and
forward.
Then
begs
the
question:
well:
where
is
that
being
stored
before
us
forward.
A
Yeah
can
I
suggest,
maybe
in
that
diagram,
I
I
think
it's
a
perfectly
valid
point
that
volume's
not
defined
for
context.
Kubernetes
happens
to
use
the
term
volumes
for
what
they
call
persistent
or
even
non-persistent
volumes,
but
I
think
we
maybe
want
to
be
more
open
here
in
that
these
Solutions
at
the
edge
nodes
might
be
kubernetes,
but
perhaps
not
so
labeling.
That
storage
would
that
be
yeah.
A
Well,
it
could
be
ephemeral
or
not
I
think
so.
Whether
I
think
I
still
have
a
comment
there
on
the
storage
section
that
Andy
mentioned
that
I,
don't
feel
is
fully
resolved.
Yet
as
to
whether
this
is
a
blanket
prohibition
on
use
of
persistent
storage
at
the
edge
node
or
not,
but
anyway,
I
think
storage.
A
There
should
be
no
debate
on
that's
appropriate
in
the
diagram
and
whether
you
choose
to
call
out
that
that
storage
happens
to
be
a
cache
only
or
not,
I
think
it's
something
an
issue
still
on
the
table.
B
How
about
labeling
it
something
like
application,
specific
storage,
so,
depending
on
how
you,
if
it's
a
buffer,
if
it's
a
cache,
if
it
has
retentions,
if
it
has,
if
it's
meant
to
be
persistent
too,
like
there
can
be
cases
where
people
would
love
to
have
things
persisted
on
the
edge
itself,
maybe
some
Edge
devices
might
be.
We
are
capable
enough
to
have
a
terabyte
disk,
I,
don't
know
or
nvme
stuff
on
it.
So
just
a
suggestion,
maybe
yeah.
J
Is
Liam,
I
I
think
the
Instinct
to
take
all
the
implementation
details
and
specific
attributes
out
of
the
discussion
is
the
right
one.
We
face
this
issue
in
webassembly,
for
example,
to
take
like
blob
store.
For
example,
there
are
many
different
implementations
of
a
blob
store,
whether
it's
a
physical
disk,
whether
it's
S3
or
something
along
those
lines,
and
to
the
to
the
various
points
shared
around
the
room.
J
Those
could
be
spooled
to
physical
disk
spool
to
memory
the
implementation
details
really
don't
particularly
matter
and
I
think
it
makes
the
the
diagram
and
definitions
it
Waters
them
down.
If
you
try
to
pin
them
to
a
specific
type.
J
At
this
point,
I
think
we
want
to
pull
them
up
really
high
to
some
sort
of
generally
accepted
definition
like
a
like
a
blob
store
or
just
a
loose
definition
of
a
volume
to
allow
specific
momentations
to
to
fall
in
the
second
thing
I
was
going
to
mention
is
yes,
the
I
did
show
the
graphic.
Last
week
we
made
some
edits
to
it,
based
on
the
feedback
we
got.
We
pulled
the
kubernetes
down
a
little
bit.
J
I'll
talk
to
my
graphics
person
and
I'll
share
that,
in
the
in
the
spirit,
slack
Channel
I
apologize,
I'm
mobile
today,
I'm
about
to
hop
on
a
flight
to
Ireland
and
then
the
the
third
item
I
just
wanted
to
touch
on
was
on
the
zero
trust
piece.
I
agree.
That's
super
interesting.
It's
probably
a
separate
full
topic.
Capital
One
just
did
an
awesome
talk
on
zero
trust
microservices,
where
they
did
zero
truck
networking,
plus
zero
trust,
microservices
and
webassembly.
Super
interesting,
stuff,
Jordan
rash.
Just
gave
that
talk.
I
Yeah
sorry
I
just
want
to
share
my
our
experience
on
on
chair
storage
or
share
PV
at
the
edge.
It
is
quite
a
challenge
because
it's
very
hard
to
have
a
storage
solution
with
low
overhead.
It
ends
up
that
to
have
share
storage
share.
Pv
requires
a
massive
amount
of
resources
just
to
have
the
storage
solution
at
the
edge,
so
that
I
agree
with
the
comments
on
addressing.
That
is
say,
it's
a
major
challenge.
I,
don't
know
if
that
needs
to
be
mentioned
or
needs
to
be
addressed
in
a
different
document.
A
D
A
Think
I
added
a
note
in
the
chat
too
saying
that
this
this
goes
all
the
way
back
to
early
days
of
Docker
and
I,
actually
love
to
comment.
Somebody
made
on
Twitter,
saying
stateless
is
a
fraud,
and
it
was
a
comment
on
these.
The
12
factors
adage
saying
that
all
applications
should
be
stateless
and
it's
an
observation
that
declaring
that
all
apps
are
stateless,
isn't
real.
A
You
have
to
have
stayed
somewhere
and
what
you're
doing
is
just
saying
I'm,
going
to
throw
the
hard
Parts
over
the
fence
to
somebody
else,
and
they
don't
go
away
now.
You
could
come
up
with
a
you
know,
a
blanket
statement
saying
that
when
you
run
an
edge
you're
not
allowed
to
have
any
state
at
Edge,
but
I
personally
think
that
in
many
cases,
Julio
is
right
and
it's
really
hard
to
have
reliable,
persistent
storage
and
Edge.
A
But
in
some
cases
you
have
no
choice,
I
mean
if,
if
you
are
faced
with
an
application
that
has
data
sovereignty,
regulations
saying
that
this
information
isn't
allowed
to
leave
this
site.
You've
got
to
deal
with
it
and
you
might
be
in
a
situation
where
you
have
to
deal
with
persistent
storage
and
Edge.
So
I.
Don't
think
that
it's
fair
to
make
a
statement
that
it's
outright
band.
P
No
well
I,
so
Steve
we
get
into
like
some
interesting
spaces
and
conversations
when
it
when
it
comes
to
persistence
and
non-persistence
at
the
edge
I
I
can't
say
that
there's
got
to
be
Fiat
right,
but
but
at
the
same
time,
what
I
would
say
is
this:
is
that
Mo,
when
we're
talking
about
I,
have
to
wear
two
hats?
When
we
talk
about
this,
the
first
is,
is
if
I
want
the
truest
sense
of
what
an
edge
is
as
defined
by
the
iot
working
group
and
by
others
around
right.
P
P
So
we'll
be
thinking
through
those
parameters
and
through
that
lens,
then
it's
reasonable
to
think
that,
and
this
is
the
reason
why
the
tiering
was
introduced
in
this
document,
but
yet
not
fully
explained
if
you
have
another
tier
for
which
you
can
offload
some
of
that
responsibility.
The
one
we
believe
one
of
the
best
practices
is
to
do
so
right
is
to
is
to
give
it
to
it,
doesn't
have
to
be
the
Hub.
P
It
can
be
an
intermediary
that
exists
just
North,
let's
just
say,
without
getting
too
specific
on
proximity
north
of
of
what
that
Edge
device
is
that
is
responsible
for
things
like
storage
or
other
other
functions
yeah.
So
it
it's
not
Edge,
but
it's
intermediary.
It's
a
service
provider
of
some
sort.
A
Yeah
I
agree,
certainly
that
if
you've
got
it
and
the
options
available
that
generally,
if
you
don't
have
constraints
on
bandwidth
and
it'll
work
for
you
fine,
you
should
use
it,
but
I
think
that
there's
also
situations
where
that
won't
be
an
option
so
that
you
saying
that
you
would
recommend
it
or
provide
features
that
support.
It
is
fine,
but
you
know
going
as
far
as
the
12
factors
of
all
applications
shall
be
stateless,
I.
Think
no.
P
P
You're
right
I
think
we
can
back
away
from
there
being
Fiat
I
totally
agree
with
you.
There
I
I
just
think
it's
it's
you're
you're
right
in
in
the
way,
you're
softening
the
approach
and.
A
I
think
there's
a
even
a
Continuum
where
you've
got
the
storage
north
of
the
edge
node,
where
you
could
have
it
be
a
caching
situation
that
attempts
to
be
one
for
one
sort
of
a
stretch,
source
storage
scenario
where
you
know
well,
you
know
on
this
Continuum
there
could
be
one,
that's
synchronous
where,
if
you've
had
transaction
like
processing,
the
actual
right
isn't
confirmed
until
you
know
for
a
fact
that
you
have
made
a
duplicate
copy
north
of
your
Edge
node
and
it's
guaranteed
to
come
back.
A
Should
you
suffer
outages
or
failures
at
the
edge
site?
Another
one
is
kind
of
best
efforts
where
you've
kind
of
put
it
in
a
pipe
to
be
replicated
elsewhere.
But
it's
subject
to
loss
and
kind
of
the
bottom
of
that,
where
you're
trying
to
do
your
best
best.
One
might
be
something
like
you
have
storage
at
Edge,
but
you
do
daily
backups
on
a
24-hour
basis.
If
you
lose
that
edge
well,
you
might
have
lost
any
data
recorded
in
the
last
day,
but
you
didn't
lose
everything
and
it's
better
than
nothing
and
I'm
reminded.
P
I
remember
as
you're
talking,
I'm
reminded
and
I
get
Twitches
right
and
nightmares
about
these
types
of
things
where
my
my
file
system
is
full
and
I
can't
do
anything
anymore,
right,
I
can't
I,
can't
start
I
can't
execute
applications,
I
can't
make
changes
to
the
operating
system
or
the
system
configuration
I,
get
all
kinds
of
restraints
and
there's
nothing.
That
alarms
me
immediately.
That
says:
hey
it's,
because
your
your
file
system's
full
dummy.
P
So
the
thing
is,
is
that
you
know
when
we
talk
about
small
devices,
we
want
to
kind
of
stay
away
from
there.
Being
this
you
know
hey,
it
could
be
anything
you
want.
We
want
to
kind
of
give
some
recommendation
here
so,
but
I
agree
that
it
shouldn't
be
as
strict
as
we
might
be.
Projecting
it
right
now,
I
see
hands
up,
so
I
wanted
to
defer.
J
I
I
think
this
discussion
is
exactly
why
we
can't
be
too
prescriptive
and
I
think
we
did
an
awesome
job
but
I
think
maybe
if
we
took
the
other
approach
and
we
instead
of
talking
about
the
implementation
details
or
that
would
show
up
in
a
particular
solution.
If
we
talked
about
some
of
the
requirements
on
the
front
side,
it
would
help
I
think
to
make
room
for
lots
of
different
implementations.
J
Let
me
give
you
an
example
of
what
I
mean
if
we
called
out
instead,
some
of
the
edge
requirements
that
may
show
up
like,
for
example,
limited
and
deliberate
autonomy,
privacy,
security
performance
constraints
and
so
forth.
We
could
essentially-
and
that's
me,
that's-
might
me
with
all
the
pains
I
apologize.
J
We
could
I
think
make
room
for
lots
of
of
different
implementations
to
show
Downstream
and
still
give
people
the
it's
going
to
give
the
understanding
that
hey.
We
understand
that
you
know
for
a
satellite,
that's
intermittently
connected
around
the
earth
or
a
driverless
car
that
is
sometimes
disconnected
from
the
internet.
There
are
all
sorts
of
requirements.
So
let's
just
talk
about
some
of
the
requirements
that
may
show
up.
I
apologize
I
get
a
slack
message
whenever
someone
logs
into
our
product,
so
they've
just
got
people
a
ton
of
people.
C
No
worries
thanks
for
that,
so
I
think
you're,
saying
if
we
introduce
something
a
little
earlier
in
the
paper
talking
about
these
potential
requirements
on
the
front
side
list
them
out
doesn't
have
to
be
an
exclusive
list,
but
then,
when
we
introduce
these
sections
later
on,
it
sets
the
stage
for
a
little
bit
of
flexibility,
and
we
can
probably
change
the
name
or
change
the
language
here
right
now.
It
does
sound.
Dictatorial
data
only
says
something
like
can
be
dead.
A
Me
throw
a
little
bit
more
color
on
that
storage.
Oh
okay,
on
the
storage
thing,
I
hate
this
maybe
already
is
already
out
of
control,
but
I
think
it's
important
to
realize
that
this
storage
goes
into
multiple
categories
too.
There's
the
storage
used
by
your
app.
You
know
if
you
like,
write
an
application
and
it
tries
to
use
I,
don't
know
database
or
whatever
that
that's
an
implementation
detail,
but
your
app
store
might
have
persistent
storage
needs
and
you
might
aspire
to
keep
that
in
a
cloud
or
need
to
keep
it
local.
A
But
in
addition
to
that,
the
control
plane
itself,
if
you
use
kubernetes
that
might
be
the
FCD
backing
store,
becomes
an
issue
and
there's
things
kind
of
in
a
gray
area
like,
depending
on
your
techniques
of
app
deployment,
you
might
get
involved
with
Git
Ops
and
have
a
git
repo.
You
likely
have
a
container
image
registry,
maybe
a
Helm
chart
registry.
A
If
the
edge
node
needs
to
be
autonomous,
meaning
that
it
it
can
boot
up
when
the
upbound
connection
is
severed,
you
you're
going
to
have
effectively
a
local
registry
in
order
to
pull
that
off.
Potentially
now
maybe
that
gets
replicated
so
that
that
registry,
that's
local
at
the
edge
node,
is
really
a
cache,
but
you
you
should
make
a
conscious
decision
of
that
and
likewise,
if
you're
using
get
Ops,
you
might
have
kind
of
a
local
git
repo
that
is
being
replicated
from
an
upstream
one
in
the
cloud.
L
Yeah
prakashi
I
would
like
to
say
that
private
cloud
and
even
virtual
private
cloud
is
a
cloud
private
Cloud,
even
though
it
may
be
hosted
on
a
public
Cloud.
So
I
don't
think
you
should
put
an
on-prem
there.
Just
leave
it
as
private
cloud
in
the
table.
There.
O
So
I
think
there's
a
lot.
I
mean
the
reason
there
there's
so
many
different
use
cases
and
slash
locations.
Sometimes
it's
really
hard
to
generalize
at
this
point.
For
my
understanding,
for
example,
the
how
assembly
applications
in
the
orchestra
right
now,
it's
not
by
by
kubernetes
right
dogs,
Trader
but
I,
believe
there
was
work
being
done
to
actually
change
that
so
yeah.
So
there's
a
lot
of
scenicity,
evolving
and
and
and
it's
hard
to
generalize
in
a
lot
of
areas.
L
Yeah,
true,
that's
true,
but
at
the
same
time
we
are
only
focused
on
the
weekend
application
right.
So
there
is
a
transformation
enrolled
and
the
transformation
may
be
very
compute
intensive.
So
that
is
why
we
may
have
to
move
it
out
of
the
age
to
any
place.
L
Otherwise,
if
it
was
as
simple
as
just
capturing
the
image
and
applying
it
that's
a
different
story
here,
you
need
sometimes
image
recognition.
You
need
Sometimes
some
other
Transformations,
so
that
is
one
of
the
main
reasons
that
we
have
to
offload
some
of
the
work
and
that
could
be
Beyond
it,
and
if
it
is
everything
it
can
be
done
at
the
age
like.
L
If
you
have
something
like
what
they
call
as
image
processing,
there
is
one
more
right:
I'm,
forgetting
that
name
onni
or
something
which
is
you
can
for
a
machine
learning
and
all
you
can
process
use
those
models
to
do
runtime
there
itself
with
a
reduced
load
workload.
For
that,
then
it
is
possible.
Anyways.
A
A
Rob,
you've
got
your
hand
up
and
we'll
make
you
the
last
comment
here,
because
we're
nearing
our
time
all.
N
Right
thanks,
I
think
one
of
the
challenges
here
is
the
on
the
diagram
where
we
Define
what
is
Edge
and
what
is
cloud.
We
have
a
whole
list
of
constraints
that
people
have
to
consider
and
I.
Don't
think
people
are
going
to
have
all
constraints
in
all
use
cases,
so
people
are
going
to
make
trade-offs
and
they're
going
to
put
more
storage
at
the
edge
in
order
to
overcome
the
problems
with
bandwidth
and
latency.
N
A
I
like
that
idea
anyway,
we
we
are
at
the
top
of
our
brand
and
I
think
you're
driving
this
with
others.
So
I'll
leave
you.
We
can
continue
on
asynchronously
using
the
comments
in
that
doc,
but
yeah
I
I
agree
with
Rob's
last
comment
myself
that
it
that's
an
interesting
perspective
that
in
some
ways,
allocating
more
resource
to
storage
is
perhaps
a
valid
trade-off
for
having
connectivity
constraints.
C
Stephen
to
help
the
folks,
the
authors
who
weren't
here
today
understand
the
discussion.
Could
you
share
with
us
the
recording
yeah.
A
I
can
I'll
I'll
get
it
up
in
the
next
24
hours?
Okay,
thank
you.
Okay,
thanks
everybody
for
attending
and
we'll
see
you
at
the
next
meeting,
bye.