►
From YouTube: Network Service Mesh BoF Meeting - 2019-04-08
Description
Join us for Kubernetes Forums Seoul, Sydney, Bengaluru and Delhi - learn more at kubecon.io
Don't miss KubeCon + CloudNativeCon 2020 events in Amsterdam March 30 - April 2, Shanghai July 28-30 and Boston November 17-20! Learn more at kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects
A
B
At
least
for
me,
too,
it
was
very
useful
to
meet
like
a
good
part
of
the
community
there
at
attornies
and
I
think
that
we
had
a
lot
of
useful
discussions
and
a
lot
of
things
to
follow
up.
So
one
of
the
things
that
actually
was
mentioned
was
I,
think
Jeffrey.
You
were
kind
of
initiating
this
about
the
use
case
with
the
emergency
and
these
things.
A
Yes,
certainly
nice
intro
Nicola,
so
from
my
side
done
so
we
had
the
nice
panel.
Of
course,
besides
Anderson
talks
on
kubernetes
or
the
network
series
Orchestrator
sort
of
a
little
bit
of
up
leveling
of
nsmb,
given
that
as
Haffner
annuity,
never
really
could
get
too
NSM,
but
were
all
it
was
well
attended
and
also
to
add
to
Nikolai's.
So
basically
stress
on
the
use
cases,
especially
coming
from
Jeff.
The
other
one
I
would
like
to
add
is
essentially
around
for
the
t-shirt,
size
infrastructure
profiles
rates.
Basically,
this
is
nothing
new.
A
This
is
very
much
what
covered
us
to
wait.
For
example,
they
are
fixing
the
incisors
and
also
even
hardware
configuration,
but
it
comes
to
networking
our
computer
in
terms
of
this
FPGA
or
you
know,
s-sorry
worried.
So
that
is
an
effort
called
ovp
which,
where
you
may
all
know,
which
is
very
much
pushing
this
for
the
DM
Patrick.
A
I
had
a
big
chat
with
Taylor
I
think
who
had
sever
the
CMC
of
testbed,
calls
and
I
think
he
is
very
well
aware
of
the
topic
and
basically
said
use
could
be
a
very
good
joint
follow
up
on
the
their
cowbell
constitute
service.
We
know
which
occurs
essentially
every
other
Monday,
so
they're,
just
next
Monday,
you
know
so
on
so
forth.
We
don't
go
or
laugh
at
them,
but
sort
of
complement.
D
Yeah
there
was
a
lot
of
interest
with
respect
to
the
demo.
People
were
keen
on
NSM
and
also
they
were
looking
at.
What
is
the
benefit
of
adopting
NS
and,
of
course,
the
common
question
that
was
asked
to
us?
How
is
it
different
from
service
mesh
right,
so
it
just
I
mean
lot
of
discussions
around
innocent
I
would
say
in
all
the
three
days
so
and
the
other
thing
is.
There
are
also
interesting
use
cases.
D
A
A
E
Raymond
romkey
Before,
we
jump
into
the
demo
I
just
had
one
or
two
questions
from
a
use
case.
Point
of
view
is
NSM
worrying
about
vnf
deployments
as
well
or
is
the
use
case
assumes
that
they
are
all
CNS.
There
will
be
a
period
of
time
where
you
will
get
into
vnf
less
CNF,
it's
not
that
the
carriers
or
not
don't
see
the
value
of
CN
F's.
You
know
in
terms
of
the
overall
change
management
and
all
that,
but
from
a
reality,
point
reality
check
of
the
few.
I
see
the
overall
service
mesh.
E
E
A
Truly
believe
so
funny
I
think
it's
a
good
question.
So
the
only
question
is,
as
we
very
well
know,
regarding
VMs
on
kubernetes,
there
are
multiple
options.
Right
I
mean
there
is
Hubert,
there's
also
worthless.
It
is
very
lightweight.
So,
basically
so
now
the
question
is
like:
how
do
we
create
that
alignment
right
and
there
may
be
more
coming
up
right?
So
you
know
so
basically
how
do
we
put
it
all
together?
So
in
that
sense
we
didn't
deem
it
as
like
the
highest
priority
topic.
A
They
basically
get
our
CNF
story
straight,
you
know
and
then,
as
we
progress
make
sure
that
we
are
aligned
completely
on
the
BNF
side.
Also,
you
know
kubernetes
perspective,
but
but
definitely
if
you
have
something
right
away
with
something
like
a
common
aspect
which
can
be
used
across
these
different
realizations
I
mean,
after
all,
all
of
the
muesli
built
it.
But
let
us
you
know,
so
we
should
then
definitely
look
at
it
because
at
the
same,
the
same
observation
about
interrupts
infrastructure
profiles
right
so
basically
they're
coming
from
the
VM
lens.
A
So
now
so
there's
a
nice
GSM,
expec
I'll,
send
it
to
everybody
and,
in
fact
put
a
link
to
it
in
you
know,
never
document.
So
basically,
it's
funny,
though
it's
coming
from
adding
the
EPC
table.
Vendor
started
it
in
GSM
a
but
it's
very
gentles.
It's
nothing.
So
is
this
PTC
alongside
those
infrastructure
profiles,
but
now
yeah
we
do
want
to
make
sure
it's
uplink
to
CNS,
but
now
we're
thinking
the
others
like
whatever
you
for
doing
for
CA
next
should
apply
to
VN
s.
Also
right.
E
Yeah
we,
the
service
level,
aspects
the
service,
chaining
aspects.
You
know
all
those
are
independent
of
the
virtual
compute,
whether
it's
a
VM
or
a
container
yeah
and
and
and
that's
fine.
You
know
if
we
take
it
in
a
phased
approach.
Let's
get
get
the
details
about
Darfur,
a
pure
CNF,
and
then
we
can
as
a
second
step,
look
into
that.
That's
also
fine.
You
know,
I
was
just
wondering
if
the
topic
has
already
been
discussed.
You
know.
A
No,
we
haven't
gone
through
that
detail,
so
basically
at
least
near-term
I
mean
just
immediately.
The
answer
is
our
answer
in
any
of
these
use
cases:
ENS
M,
so
basically
saying
hey,
you
know
we
can
inter
work
with,
for
example,
of
OpenStack
based
solution,
great
for
VF
or
via
net
or
P&S,
like
a
audio
read
switch
with
crux
of
the
demo,
from
frame
rate
so
basically
will
create
the
interworking
asperity
and
to
a
point
of
funny
like
yeah,
the
build
starts
looking
at
us.
You
know
from
a
common
perspective,
this
I'm
thinking
more
like
this.
A
Let
be
differentiated,
settle
down
like
you
know,
but
let
I
think
have
been
support.
I've
heard
support
especially
may
not
have
started
with
more
than
varietals
in
Telus,
like
a
big
fan
of
hood
from
you
know,
I'm
a
lightweight
perspective
with
extremely
lightweight
to
Burt
is
a
little
heavy
Abele,
but
it's
very
rich
in
functionality.
So
yeah.
D
Yeah
I
just
try
to
attack
Lee
one
other
thing
is
at
least
the
non-si
enough
functions
are
looked
at
as
an
SM
entity
within
an
ism
so
and
what
we
are
trying
to
do
is
we
are
basically
trying
to
ensure
that
the
path,
speed
and
all
the
translations
needed
from
that
of
the
inocent
to
that
of
the
non
andersonville
gets
done
by
a
via
the
en
SM.
So
that's
the
whole
crux.
D
E
D
You
so
let
me
know
if
you're
able
to
see
my
screen,
so
what
we
have
done
is
we
essentially
developed
a
shim
layer
which
can
be
thought
of
as
a
DNS
sub
to
begin
with,
between
open,
daylight
and
Tennison
I
believe
all
of
you
would
be
familiar
with
open
daylight,
it's
basically
open
source,
Sdn
controller
and
what
it
does
is.
It
provides
the
ability
to
support
multiple
solve
bones,
for
example,
open
flow
net
conf,
BG,
pls,
pc
and
any
other
thing
that
you
can
add
right
at
Illumina.
D
What
we
have
done
is
we
have
also
extended
it
the
framework
to
support
physical
network
functions.
What
does
it
mean,
as
we
have
basically
a
JSON
RPC
endpoint
and
you
can
define
an
open,
config
yang,
which
represents
that
device
and
you
can
perform
the
operations
on
the
open
config
on
and
behind
the
scenes
it
gets
translated
into
the
CLI.
D
D
What
it
means
is
I
was
mentioning
that
open
conflict
for
physical
devices,
but
you
can
bring
in
any
yang
model
and
you
can
define
the
operations
that
you
want
to
perform
on
it
and
we
have
the
translators
that
would
essentially
translate
it
right.
So
similarly,
what
what
has
been
done
here
is
we
have
basically
the
ENS.
Some
part
of
the
opendaylight
integration
has
two
parts
to
it:
one
us
into
the
NSM:
it
can
talk,
G
R,
we
see.
On
the
other
side,
it
talks
JSON
RPC,
so
that's
the
whole
essence
of
it.
D
So
what
you
see
is
essentially
the
yang.
So
all
the
yang
gets
mounted
as
a
yang
in
point,
and
here
you
see
an
SM
to
be
an
endpoint
and
from
a
network
function
perspective
or
a
network
service
perspective.
We
have
taken
the
sample
Network
function,
which
is
the
ICMP
responder
that
has
been
created
by
the
team
and
what
this
particular
ICMP
responder
does.
It
basically
creates
a
pod,
injects
an
NSF
interface
to
it,
and
then
it
provides
is
the
ability
to
ping
those
interfaces.
D
So
that
you
know
the
interface,
the
NSM
interface
that
was
been
created
in
this
part
is
active.
So
that's
the
intent
of
this
demo,
so
we
have
created
four
operations.
One
is
create
interface,
which
would
essentially
be
invoking
the
create
part
of
this
whole
ICMP
responder
showed
a
essentially
show
the
interfaces,
which
means
the
parts
and
the
respective
interfaces
and
the
ping
would
essentially
being
those
with
that.
Let
me
get
into
it.
D
A
D
D
D
A
D
Let
me
take
an
example:
okay,
so,
for
example,
let's
assume
you
have
an
VPN,
it's
executing
in
the
NSM
context.
Vpn
would
essentially
expose
those
operations
to
the
outside
world,
which
is
essentially
the
endpoint,
and
then,
if
it's
a
unless
some
client,
you
would
essentially
follow
the
same
mechanism
of
trying
to
look
for
the
particular
endpoint.
Get
that
end
point
and
then
invoke
that
endpoint.
D
So
similarly,
here
what
happens
is
there
has
been
a
client
and
a
some
client
that
has
been
developed,
which
would
essentially
be
triggered
by
the
en
SM
right,
and
it
would
talk
natively
to
this
in
the
NSM
context,
but
to
the
outside
world.
It
would
essentially
translate
and
give
it
in
the
format
whatever
you
want.
Okay
right
yeah,
so
let's
get
into
the
demo,
so
to
begin
with
I'm
going
to
just
show
the
interfaces.
D
So,
as
you
see,
basically
what
it
has
done
us
it
has
essentially
he
executed
that
call,
and
then
it
shows
what
are
the
end
points
or
what
are
the
parts
that
has
been
created.
So
there
are
like
these
three
pots
that
have
been
created:
NSA
VPP
means
it
is
basically
by
a
socket
and
VPP
agent
means
it's
via
the
memory
interface
and
the
IP.
What
you
see
is
not
the
cni
IP
of
the
part,
but
the
NSM
IP
okay,
so
each
of
this
parts
would
essentially
have
the
MSM
interface
created
in
it.
D
A
So
my
question
is
on:
is
it
really
a
point-to-point
interface
between
the
unless
important
heroes?
More
of
a
just,
you
are
just
create
trigger
trading,
as
any
interface
which
is
open,
trim
is
not
naturally
point-to-point.
A
D
E
D
D
D
Is
a
requirement
so,
for
example,
okay,
let
me
probably
give
a
bit
of
background
about
NSM.
Probably
Nicolay
is
the
master
in
how
things
work
I
can
give
you
more
from
a
use
case
perspective,
so
the
requirement
what
we,
how
the
whole
Emerson
started
was.
Basically
what
happens
is
in
case
of
a
typical
telco
workload
for
sure
need
multiple
interfaces
right.
D
So
today,
in
case
of
the
kubernetes
part,
you
just
have
one
interface,
which
is
a
CNI
interface
right.
That
is
one
second
thing
is
for
example,
most
many
of
the
telco
routes
would
essentially
need
to
have
their
own
non
IP
channel,
which
is
essentially,
for
example,
you
will
have
MPLS
LSP
or
any
other
signaling
thing
which
cannot
be
done.
Why
are
your
conventional
IP
interface
of
C
night
right
with
that?
D
E
D
E
D
In
the
you
examples,
whatever
you
see
in
their
nests
and
what
happens
is
this
particular
interface
is
the
one
that
would
be
essentially
used,
for
example,
to
connect
with
that
of
your
client
for
the
data
plane
purpose.
What
it
means
is
the
VX
land
would
be
created.
Using
these
interfaces.
There
will
be
similar
injection
that
will
happen
on
the
client,
and
this
is
all
done
by
the
NSM
infrastructure.
I
see.
A
One
request
inserting
the
GS
is
bad
calm,
so
basically
in
so
I
was
thinking.
So
basically
there
any
detailed
questions.
Perhaps
you
should
follow
up
separately.
Address
I
would
like
to
sort
of
sum.
I
mean
Jeff
has
some
interesting
ideas
on
the
near-term
use.
Cases
like
this,
so
basically
gonna
make
some
progress
on
it
and
a
call.
The
next
half
an
hour,
yeah
sure
yeah.
Definitely
yeah.
D
E
D
The
other
thing
is,
the
other
part
is
essentially
create
interface,
whether
it
shows
all
the
interface
but
for
benefit
of
time.
I'll,
probably
take
it
offline
depending
upon
the
interest
and
I
can
showcase
the
pods
and
the
interfaces
within
the
pod
aspects
like
that:
okay
yeah.
So
this
was
the
demo
all
about
yeah.
C
A
C
Emergency
response,
like
little
video
right
like
I,
mean
Cisco,
has
like
this
specialized
tag
team
that
does
this
I'm
Charter.
We
have
this,
but
in
the
event
of
a
natural
disaster
you
know.
Obviously
infrastructure
and
services
is
the
like
key
things
that
has
to
come
up
very
very
quickly
and
it's
incredibly
variable
and
difficult
and
I
was
thinking
a
very
simple
use
case
of
showing
us
like
using
something
like
an
en
SM
sitting
on
some
small
cells.
Or
you
know
some
access
points.
C
You
know
Wi-Fi
access
points
or
whatever
stitching
directly
into
a
provider
network,
so
like
say
like
charters,
access
network
and
did
that
provisioning,
ports
and
scale
for
the
increased
load,
I
think
would
be
super.
Compelling
I
haven't
had
a
chance
to
like
diagram
this
out.
So
it's
kind
of
hard
for
me
to
try
to
just
put
this
into
words.
But
imagine,
like
you
know,
a
tornado
goes
through
rural
Alabama
and
the
town,
in
fact,
I
mean
now.
C
We
can
use
this
right
here
to
buy
just
the
kind
of
thing,
but
imagine
as
opposed
to
this
more
like
static
enterprise.
Use
case
like
what
we're
trying
to
show
is
tornado
hit,
the
town
infrastructures
gone,
you
know,
cell
towers
been
taken
out,
and
you
have
all
these
responders
in
the
area
who
have
no
connectivity.
If
we
can
show
a
use
case
to
where
we
can
bring
devices
online
like
dynamically
and
as
they're
coming
up,
you
know
these
would
be
like
essentially
be
client.
C
There
would
be
pre-established
in
points
in
the
access
network,
and
you
know
both
Iran
or
the.
You
know
the
Metro
Ethernet
side
of
the
house
where
they
are
presenting
themselves
as
endpoints
and
accepting
you
know,
connection
requests,
and
so
you
could
basically
just
literally
plug
devices
in
come
up
on
DHCP
and
then
request
your
network
service
and
suddenly
bring
like
an
emergency
response
network
into
being
I,
like
largely
do
like
I
declared
a
network
service.
Does
that
make
sense,
yeah.
A
Yeah,
that's
a
very
good
use
case
here.
Ya
know
make
sense
yeah.
This
is
sort
of
a
generalized
depiction
of
the
I
mean
H
computing
use
case
didn't
get
into
specific
details
on
emergency
response,
but
tender
ith
computing.
You
know
how
everything
can
happen
in
the
edge.
So
basically
there
are
two
clouds.
You
know
B,
you
have
the
ran
close
on
the
left,
and
then
you
have
the
H
computing
cloud
on
the
right.
A
C
And
it's
not
just
the
not
just
the
you
know
the
multi-tenancy
aspect
of
you
know
bringing
more
and
more
consumers
online,
but
you
using
something
like
an
en
SM
short
term
such
as,
like
friends,
odl
model
or,
yes,
writing.
You
know
the
abstraction
for
network
service
managers.
You
live
on
a
PMF
itself,
but
I'm
DM,
the
you
know
QoS
and
the
bandwidth
provisioning
I
think
is
where
it
could
be
really
really
dynamic
right,
like
if
an
emergencies
happened
and
we
start
putting
up
these
small
cells
or
these
wireless
access
points
out
in
the
field.
A
C
That's
exactly
what
I'm
thinking
is
like
basically
using
an
SM
to
create
a
a
more
dynamic,
SD,
LAN
right,
like
SD
way
and
still
like
largely
at
least
all
the
major
vendors,
whether
it's
a
low
clouds
atella.
You
know
it.
Basically
everything
but
Meraki
is
still
a
very
traditional
top-down.
You
know,
centralized
orchestration
model
versus
going
with
a
more
distributed,
declarative
model,
and
it's
exactly
I
use
the
Frederic
I.
Think
I,
you
and
I
talked
about
this.
That's
exactly
what
I
was.
A
A
A
A
So
do
we
think
you
wanted
to
just
double
click
on
this,
so
basically
I
have
the
PowerPoint
for
this.
Do
you
wanna
take
it,
take
it
and
basically
just
craft
into
an
emergency
responder
use
case
and
say:
hey
here:
are
the
network
functions
and
basically
a
double
click
on
it?
You
know
so
thank
you.
C
Sure
yeah
I'll
make
like
a
copy
of
this
and
work
with
some
Nikolai
and
Frederick
on
kind
of
expanding
out
like
what
we
were
discussing
last
week,
yeah
I
mean
I.
Guess
it's
basically
like
if
I
was
going
to
sum
it
up
to
a
single
sentence.
It's
a
declarative
radio
access
network
with
an
SD
web
component.
A
C
Think
long
term,
that's
where,
like
the
real
value,
is
I,
think
starting
small
first
and
getting
the
SDM
component
up
and
then
I
mean
or
we
can
go
in
reverse
and
just
show
the
dynamically
provisioned
way
and
I
sorry
I
ran
either/or,
but
I
think
the
story
is
not
complete
unless
you
have
both
ends
of
it
right,
like
you
could
set
up
a
small
cell
and
instead
of
everybody
trying
to
individually
go
to
their
phones.
C
We
have
you
know
like
some
type
of
5g
CPE,
that's
SDA,
I'm
capable
that's
you
know
reaching
into
the
metro
and
asking
for
you
know,
QoS
and
in
capacity
and
it's
doing
internet
offload
right
like
it's
basically
tunneling
straight
to
like
one
of
the
pops
and
then
it's
out
into
the
great
blue
yonder.
You
know
in
whatever
emergency
services
the
on-site
personnel
need
I,
think
is
kind
of
like
in
my
mind
what
we'd
want
to
show
right
and
like
to
Frederick's
point
I
like
there's,
there's
so
many
different
components.
C
You
know
like
tons
and
tons
of
data
needs
to
be
shipped.
You
know
people
who
are
doing
triage
and
stuff
for
probably
sending
photos
in
real
time
to
emergency
professionals
who
are
not
on
site.
You
know
helping
them
assess
injuries.
C
You
know
you're
gonna,
take
pictures
of
structural
damage
and
you
have
people
there
on
site
who
don't
necessarily
know
if
they
can
like
pick
XYZ
components
from
a
fallen
building
off
of
somebody,
but
without
you
know
causing
more
structural
damage
like
I,
don't
know
it
just
it's
a
very,
very
chaotic
and
dynamic
thing,
and
if
we
could
show
we
could
make
the
networks
of
that
less
chaotic
and
I.
You
know
and
more
dynamic.
Then
you
know
just
like
Lily.
All
you
do
is
like
say
plug
this
device
and
come
on
the
network.
C
A
A
C
A
E
So
Jeff
funny
here
I
actually
like
the
use
case,
the
way
you
have
defined
it
in
it's
a
very
valid
one,
just
varying
the
technical
hat.
Does
it
make
sense
to
obviously
it's
two
pieces
right?
One
side
is
the
SD
van
piece.
The
other
side
is
the
radio
part
or
the
the
radio
ran
part
of
it.
You
know
if
you
look
at
the
3gpp
spec.
Also
they
clearly
separate
the
radio
access
from
the
packet
access
networks
right.
There
is
even
kind
of
starts
at
the
packet
side
of
the
overall
and.
A
E
It
work
right
so
even
from
a
use
case,
point
of
view
would
you
agree.
It
is
like
bringing
on
board
a
new
SD
van
site,
but
but
from
a
use
case
point
of
view.
You
bring
some
good
points
as
to
what
kind
of
dynamism
and
special
requirements
that
particular
SD
van
site
need
to
support,
and
and
and
how
can
we
make
that
integration
with
the
4G
or
5g
LTE
network,
more
tighter
and
so
on?
C
Yeah
I
mean
I
you're
gonna,
have
you
know,
network
service
managers
that
are
unique
to
each
section
of
this
I
I
think
from
a
development
standpoint
you
can
develop
them
independently
of
each
other
I.
Think
for
the
end-to-end
like
showcasing
of
it.
Like
you
said,
it
would
basically
be
like
a
new
side
activation,
but
specifically,
as
opposed
to
a
wired
connection.
C
It
would
be
a
4G
or
5g
enabled
on
CPE
right,
and
so
you
drop
a
small
cell
into
the
disaster,
location
or
multiple
small
cells
and
those
are
dynamically
provisioned
into
the
ran
and
then
simultaneously
like
once
that's
come
up.
You
bring
up
a
you
know,
LTE
or
5g,
capable
on
CPE,
because
the
thing
about
Sdn
is
it's
really
hard
and
like,
despite
the
fact
that
lots
of
really
smart
people
have
been
designing
solutions
for
a
long
time,
I
mean
I.
C
Like
having
like
a
land
that
you
create,
that's
tying
back
into
both
the
radio
access
network
via
the
LTE
antenna
on
the
CPE,
which
is
then
helping
you
back,
all
of
your
traffic
to
the
actual
Internet
itself,
I
think
is,
is
cool
so
to
your
question,
because
I
extend
to
all
the
talkin
ever
bro
stand.
Roundabout
manner
is
from
a
technical
standpoint.
Yes,
I
think
that
they
can
largely
be
you
know
their
own
standalone
use
cases,
but
then
also
showing
the
integration
between
the
two
is
like
we're.
You
know
the
powerful
message
comes
in.
E
A
I'm
just
to
Isis
that,
actually,
if
you
look
at
this
use
case
it
it's
almost
like
the
left
side
like
the
BBU
and
then
they
would
be
if
they
simply
give
you
this
is
your
your
ran
and
then
so,
basically
you
see,
then
you
do
a
handoff
day
to
the
packet
networks.
At
this
point
you
do
a
handoff
and
you
can
do
basically
the
coming.
You
can
directly
hand
out
with
edge
computing
application
here
right.
You
know.
Maybe
there
could
be
a
first
tender
guess
that
that
is
computing.
A
E
A
D
D
D
A
A
Actually
nicely
ties
with.
There
is
a
very
nice
sort
of
a
presentation
from
Verizon
an
Intel
on
sort
of
a
CBE.
You
know,
convert
CPE,
so
I
think
this
falls
very
well
in
place.
So
if
you
know
CD
in
a
box
which
that's
both
the
natural
function,
processing
and
also
application,
so
maybe
sort
of
first
responder
in
a
box
sort
of
you
know
we
could
visualize
it.
E
E
A
Be
it
could
be
container
a
DM
at
least
on
each
of
each
components
would
be
29
a
VM
at
least
I
have
seen
more
of
being
based.
It
depends,
I
think
it.
It
all
depends
on
sort
of
the
latency
requirements,
so
some
are
not
virtualized
at
all.
So
basically,
if
you
want
like
ultra-low
latency,
then
you
don't
virtualize
the
BBL.
So
basically.
F
E
That
first
responder
use
case
Jeff.
The
other
interesting
scenario,
F
I've,
seen
as
part
of
my
yoke
note
here
days,
is
that,
like
an
emergency
ambulance,
you
know
who
is
picking
up
some
patient
and
you
need
to
the
only
reachable
do
you
have
is
through
4G
Phi
G,
it's
an
LTE
and,
and
it
means
that
the
lower
latency
you
have
the
better
when
you
are
transferring
all
the
data
from
the
ambulance
to
the
hospital
that
you're
driving
towards.
F
Yeah
I
think
that
one
of
the
nice
things
that
we
have
one
of
the
nice
advantages
is
we
can
establish
certain
types
of
connections
and
facilitate
them.
So
there's
two
things
I'm
thinking
of
first
one
is
there'll,
be
a
large
number
of
different
types
of
systems
that
need
to
be
interconnected.
We
can
help
facilitate
those
interconnections
and
the
second
one
is.
F
Those
connections
like
you're,
correct,
like
some
of
them,
are
like
I,
have
to
get
the
message
through
as
fast
as
possible
in
order
to
save
in
order
to
save
someone,
there's
also
another
use
case
we're
the
speed
of
the
message
is
not
as
important,
but
what
does
matter
is
if
this
message
has
very
strong
possibility
of
getting
through
under
editorial
conditions,
so
those
ones
it's
okay,
to
be
slower,
it's
okay,
to
bring
in
things
that
are
that
have
higher
error,
correction,
recent
properties
and
so
one.
A
really
good
example
is
I.
F
Think
there
was
a
like
1989.
They
had
this
hearing,
California
or
San
Francisco
Bay
Area,
and
they
had
trouble
getting
messages.
Saying
like
you
know,
Frederick
counts
is
okay.
He's
he's
he's
at
that
friend's
house
or
something
like
that
and
then
said
they
send
it
all
off
to
Dallas
actually
went
over
the
ham
radio
groups
and
they
then
created
a
database
there.
That
people
could
call
in
to
to
ask
hey.
Is
this
person?
F
E
Yeah,
that's
interesting.
You
escaped
the
point,
the
other
interesting
fact
about
the
emergency
ambulance
network.
You
know
it's
like
a
disaster
is
happening
every
day.
You
know,
and
these
guys
are
first
responders
who
need
to
I,
mean
I,
think
highlighting
a
distributer
edge
that
serves
a
every
day.
E
Disasters
which
is
like
a
bunch
of
ambulances
are
networked
together
and
you
know
they
are
able
to
as
you
as
you
point
out,
you
know
it's
not
just
about
latency
it's
about
guaranteeing
the
delivery,
so
the
QoS
components
across
from
the
edge
to
the
core
to
the
final
hospital
network.
I
think
that
kind
of
a
service
also
could
be
potentially
highlighted.
C
To
the
trusted
edge
compute,
but
but
yeah
so
like
the
Sdn
would
be
its
own
workflow
right
and
it
would
be
all
based
on.
You
know:
I
feat,
raffiq,
but
it
would
use
the
radio
access
network
to
tunnel
into
the
core
network.
That's
on
the
right
hand,
side
yeah.
E
And
it's
it's
almost
like
romkey
each
ambulance
is
a
is
a
branch
by
itself.
You
know
it's
a
mobile
branch
yeah
and,
as
you
guys
know,
you
know
the
first
one
hour
after
the
accident.
The
relevant
steps
the
first
responders
take,
will
decide
whether
that
patient
will
live
or
die.
You
know
so
that
two-way
communication,
it's
actually
important,
as
you
guys,
probably
already
know,
I
know.
So
there
is.
C
E
It
it
should
be
the
you
and
I
kind
of
a
model
which
is
dynamically
asking
for
what
service
it
needs
absolutely
and
eaten
each
happen
in
real
time.
So
I
think
the
more
we
highlight
such
dynamic
use
case
is
containerized
network
functions.
The
value
of
the
continuous
network
functions
will
be
more
visible
to
the
non-technical
guys
forcefully.
Oh
yeah,.
A
E
I
specifically
paid
extra
attention
to
that
in
mine
was
no
cure,
because
no
I
wanted
to
make
mock
more
marketing
mileage
out
of
it.
My
suggestion
internally
at
that
time,
was
that
let's
take
one
of
the
European
cities
and
you
know,
give
our
SD
van
solution
free
of
cost
and
demonstrate
that
and
it'll
be
on
the
news
all
over,
and
you
know
so
I
think
it
better
connects
with
every
human.
You
know
when
you
highlight
when
you
cut
when
you
relate
it
to
a
real-world
use
case
like
that.
A
So,
in
fact,
of
what
I
can
do
is
I
have
the
slides
for
this.
This
is
so
basically
I
can
distribute
to
the
team,
and
maybe
funny
also
you
may
have
some
looks
like
you
may
also
have
some
interesting
slide
system.
Those
yeah
I
mean
any
shareable,
I
mean
again
I,
don't
know
from
the
past
day.
There
is
anything
which
is
publicly
shareable.
We
can,
you
know,
use
that
locally
I
think
caught
on
right,
so
that
you
know
in
tying
the
story
together.
E
A
What
I
meant
was
if
there
is
anything
like
sort
of
the
pictorial?
Also
we
can
leverage
I
mean
if
there's
something
which
you
can
leverage
from
your
past
work,
because
it
takes
time
to
something
this
pretty
pictures
right.
A
nice
picture
of
yourself
thinking
out
loud
when
you
build
something
caribou
yeah.
E
A
E
A
F
If
we
see
that
the
connection
is
very
resilient
and
we
have
statistics
on
error,
correction
see:
oh
there's
your
errors,
let's
go
ahead
and
reduce
your
correction
and
get
that
message
to
you
faster,
but
with
a
was
still
hitting
the
DSL
a
so
so.
I'm
really
excited
about
this
particular
use
case,
because
it
shows
a
very
complex,
dynamic
situation
and
that
and
the
distributed
nature
of
this
system
will
lens
very
well
for
for
solving
the
problem
at
hand.
So
I
know
I
do.
Is
this
fantastic?
The.
C
Last
little
marketing
bit
to
write
ISM
we're
trying
to
show
like
actual
cloud
native
networking,
because
we
are
not
customizing
the
infrastructure
for
this
use
case
right.
The
whole
concept
is
an
immutable
infrastructure
with
a
common
deployment
model.
So
whether
you
know
I'm
in
Alabama
or
I'm-
and
you
know,
Massachusetts
disaster
strikes,
I
go
there
and
the
network
service
is
the
network
service.
I
put
devices
out
there.
It
makes
the
request
and
I
expect
repeatable
results
every
single
time
without
having
to
change
the
infrastructure
itself.
A
I
know
exactly
this
is
awesome,
so
fantastic
in
the
top
of
the
hour,
so
if
needed,
I
think
we
could
also
meet
on
demand,
so
I
will
send
the
slides
I
mean
me
and
prim
created.
Basically,
this
is
like
a
literally
a
cat
face
on
PowerPoint
right,
so,
basically,
I'll
send
Chad
the
slides
with
the
team
everybody
so
I
mean,
so
we
are
meaning
two
weeks
from
now
right
if
needed.
That
would
be
glad
and
we
should
I
think
meet
earlier
tools
needed.