►
From YouTube: Summit 2022: Edgestack and Singulus
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
B
Okay
and
thank
you
for
your
introduction,
hello,
everybody.
B
My
name
is
sango
shin
and
from
the
sk
telecom
in
korea,
and
I
have
been
working
in
skate
telechrome
for
like
a
few
five
or
six
years
and
the
last
few
years
I
have
been
developing
the
mec
platform
in
skt
lockman
at
sk
telecom,
and
I'm
going
to
in
this
talk,
I'm
going
to
give
a
presentation
of
like
the
edgestag
and
the
singulars
with
my
co-worker
gianli,
and
I'm
going
to
cover
a
little
bit
about
like
the
background
of
this
project
and
I'm
going
to
explain
only
because
the
less
technical
part
and
then
after
that
gen
lee
is
going
to
cover
a
little
bit
more
like
the
technical
part,
but
like
yeah
more
about
like
the
iqbal.cni
yeah,
and
I
think
yes
from
the
name
like
the
edge
stack.
B
Hysterect
is
currently
the
the
edge
specific
cloud
solution
and
the
singles
is
called
the
other
network
stack
of
the
h
stack.
So
first,
let
me
explain
a
little
bit
about
the
mec,
because
the
lg
stick
is
called
component
of
the
abc.
So
I
guess
everybody
already
knows
the
white
same
ec.
But
let
me
let
me
explain
very
briefly
so
why
usually
the
mec
is
a
multi-access
edge
computing,
but
in
sk
telecom,
because
it's
sk
telecom,
it's
a
mobile
service
operator,
so
we
call
it
like
the
mobile
edge
cloud.
B
So
originally
without
the
msc.
The
all
of
our
services
are
running
on
the
cloud
and,
as
you
can
see
in
the
this
picture,
all
the
service
comes
through
the
internet,
best
care
for
traffic.
So,
even
though
we
have
5g
network
because
of
this
kind
of
the
best
internet
investor
for
traffic,
the
latency
is
very
large.
B
And
now
when
we
introduce
the
mec
between
the
the
service
and
the
5g,
and
then
the
service
can
be
provided
very
close
to
the
user
user,
which
means
they
like
the
in
the
taco
the
center.
So
they're
like
the
latency,
can
be
reduced
dramatically
like
the
around,
like
yeah
a
few
hundred
milliseconds
to
the
now.
We
are
targeting
about
10
milliseconds,
and
not
only
that
like
because
all
our
services
are
located
in
inside
of
taco,
so
security
gets
much
better
and
also
the
ascii
idea.
B
Some
telco,
like
the
sk
telecom,
can
provide
like
some
very
specific.
The
some
api,
like
the
network
assurance
like
the
user
locations
like
user
identity
and
mobility,
features
now
that
and
features
have
been,
never
been
provided
in
the
cloud
yeah.
So
this
can
be
a
concept
of
the
msc
and
then
now
sk
telecom
has
developed
the
mvc
platform
for
like
the
last
few
years,
and
then
let's
get
it
is
a
commercialized
msc
platform
for
the
last
three
last
three
years
ago,
yeah
and
next.
B
Okay,
so
this
screen
shows
called
the
the
user
interface
of
the
sktmc
solution.
As
you
can
see,
we
are
using
the
mec,
we
can
manage
the
multiple
sites,
and
this
is
called
the
the
screenshot
of
the
dashboard
and
using
dashboard.
We
also
the
operator
can
manage
all
the
workloads
and
all
the
services
of
the
mec
in
a
single
view,
and
next.
B
And
this
is
the
customer
portal,
the
mvc
platform
manager,
the
previous
one
was
for
operators,
but
now
this
custom
portal
is
for,
like
the
developers
like
the
enterprise
customers,
like
the-
maybe
you
can
think
of
this
screen
this.
The
portal
is
called
the
aws
console,
but
using
this
console,
the
the
msc
service
provider
can
deploy
their
service
very
easily
and
any
user
can
request
kind
of
the
any
brilliant
resource
container
resource,
like
the
some
kubernetes
releases.
B
Using
this
screen,
okay
and
next
and
extra
explain
before,
like
the
main
feature
of
the
mvc,
is
to
provide
the
idea
vm
resource
to
any
mec
service.
So
I
we
really
have
to
provide
the
real
resource
and
also
like
these
days,
the
the
mec
services
content.
Well,
my
msc
service
and
also
the
attacker
service
have
been
content
containerized
so
also
we
have
to
provide
the
container
service
yeah.
So
that's
kind
of
a
background
of
the
hour
and
easy
solution.
Okay,
next.
B
B
The
first
one
is
the
mvc
router,
and
next
one
is
the
mvc
platform
manager
which
manages
the
all
of
the
the
life
cycle,
management
of
the
virtual
resource
like
vm
and
containers,
and
also
I've
shown
some
the
screenshot
orchestrator
and
I've
shown
the
other,
the
screenshot
of
the
other
business
on
the
business
console,
and
on
top
of
that
also,
we
are
trying
to
provide
some
media,
some
specific,
like
yes,
service
or
so
we
are
trying
to
provide
as
a
vision,
ai
the
edge
product
and
also
we
are
trying
to
provide
some
between,
like
the
other
mec
federation
function,
also,
the
marketplace
all
the
mec
services
like
that,
and
the
way,
if
you
look
at
the
the
the
lady
part
is,
is
edgy
stack.
B
So
we
think
that,
like
the
other
core
function
of
the
mvc
flapper
manager
is
how
to
provide
the
the
edge
specific
cloud
solution
to
the
users
very
with
agile
and
then
h
is
very
kind
of
the
the
it's
not
like
the
the
hyperscalers
like
the
cloud.
So
it's
just
like
the
the
edge.
Usually
it
is
deploying
some
kind
of
very,
very
a
smaller
area.
B
So
usually
we
have
to
provide
the
the
discount
solution
with
very
lightweight
this
kind
of
be
a
kind
of
requirement
of
our
edges
next,
so
when
we
designed
our
edge
stack,
there
were
some
challenges
like
the
this
is
kind
of
the
the
evolution
of
the
attack
network
function.
Originally,
the
the
technology
function
was
provided
as
a
country,
the
physical,
it's
called
kf,
and
then
it
has
been
virtualized
and
then
it
has
been
provided
as
a
bm.
But
now
all
of
the
services
has
been
the
containerized
yeah.
B
So
so
our
challenge
was
that
how
to
provide
the
the
vm
service
and
the
container
service
with
a
single
platform
without
like
the
heavy
platform
like
the
some
openstack
right,
you
know
some
dm
errors
like
the
tkgs
and
everything,
because
it's
like
it's
like
friendly
edge.
B
It's
not
like
the
hyperscaler,
so
we
have
to
provide
the
discount
discounted
solution
in
a
very
lightweight
yeah,
and
that
was
kind
of
our
main
challenge
and
next
and
of
course
like
there
are
like
the
openstack,
was
kind
of
very
kind
of
de
facto
of
the
some
cloud
solution
for
for
like
the
last
four
five
years,
but
like
maybe
you
anybody
you
have
who
has
been
deployed,
the
open
stack
will
will
be
realized
that
it
is
very
complex
to
deploy
and
then
also
the
management
of
the
other
open
states.
B
It
is
much
more
complex.
Of
course,
there
are
a
couple
kind
of
efforts
to
make
it
a
little
bit
easier
to
deploy
using
like
the
kubernetes
there.
There
was
kind
of
the
project
was
called
airship
and
using
the
airship
the
openstack
can
be
deployed
very
easily
on
top
of
the
kubernetes.
B
However,
still
we
have
to
deploy
the
kubernetes
on
top
of
that
also
still,
we
have
to
deploy
the
openstack
and
then
that
increases
a
lot
of
capex
and
then
and
then
also
still,
we
have
to
manage
the
both
kubernetes
and
openstack.
That's
going
to
be
requires
a
lot
of
the
management
complex
yeah,
so
we
couldn't
like
you,
choose
this
kind
of
the
solution,
yeah
and
next
and
now
both
now
as
everybody
guessed
it
now,
that's
why
we
have
chosen
to
use
the
kubernetes.
B
As
you
know,
the
the
kubernetes
provides
the
vm
service
using
the
pad,
so
using
the
the
single
kubernetes,
the
the
framework
you
know,
we
can
provide,
of
course,
the
other
container
service
and
at
the
same
time
we
can
provide
the
the
bm
service
yeah
yeah
yeah.
However,
there
are
some
limitations.
The
the
kubernetes
project
is
very
nice.
However,
there's
some
limitations
in
the
network.
It
probably
now
it
only
supports
like
calico,
flannel
and
andrea,
which
is
good,
but
but
we
but
but
we
had
to
use.
B
We
have
to
use
this
kind
of
the
the
cubicle
solution
for
the
edge
solution,
but
that
the
very
the
strict
requirement
of
the
edge
is
low,
latency
yeah.
So
we
require
the
very
high
speed
and
low
retention
network
and
that's
why
we
have
to
design
the
our
own
clever
cni,
which
will
be
explained
in
more
detail
by
the
gm
later.
B
Okay,
this
is
kind
of
the
brief,
the
the
the
artistic
architecture,
but,
as
you
can
see,
they're
using
the
kubernetes
and
the
concrete.
Now
we
can
provide
the
the
bm
service
and
also
we
can.
We
can
provide
the
manager
kubernetes,
and
the
funny
thing
is
that
now
we
can
provide
the
the
container
using
vm,
and
now
we
are
now
the
other
we
are.
B
We
are
providing
vm
using
the
containers
yeah
and
except
that
also
we
have
some
kind
of
models
called
edge
stage
disc
at
this
provides
the
the
edge
specific
the
story
solution
and
the
edge
trunk
provides
some
api
to
interface.
All
of
these
kind
of
fantastic
the
features
and
also
there's
a
component
fully
fully
provides
assurance.
The
it
collects
all
the
data
from
the
some
kubernetes,
the
the
kubernetes,
also
from
the
network
side,
and
provides
all
the
the
stat
resource
statistics
to
the
the
mxc
platform
manager.
B
And
also
we
are
because
the
our
ht
stack
is
composed
of
the
mostly
likely
with
open
sources.
So
now
we
are
using
the
open
interfaces
yeah,
for
example,
yeah,
where
we
provide
the
manage
kubernetes
using
the
vm.
We
are
using
cncf
plus
api
yeah,
so
they're
like
the
on
any
any
other
implementation
that
follows
the
cluster
api
can
be
used
for
as
an
adapter
to
the
our
the
the
the
histox,
the
managed
against
managed
kubernetes
kubernetes
provider
yeah.
B
Okay,
this
is
my
part,
and
then
I'm
going
to
introduce
generally
and
he's
going
to
talk
about
the
more
technical
detail
about
the
kubrick
sinai
and
gm
and
please
okay,.
C
Yes,
okay,
okay,
let
me
present
the
cuber
cni
part.
So
let
me
first
provide
some
motivations
of
this
cuber's
cni.
So.
C
So
yeah,
so
this
is
a
cuber
cnn.
What
we
are
going
to
introduce
today.
So
when
you
come
up
with
the
integrated
keyboard
with
the
existing,
I
try
to
migrate
the
existing
vm
services
into
the
culprit.
We
encourage
that
the
networking
party
is
not
fully
compatible
with
the
existing
openstack
and
which
one
so
so.
C
Ip
address
management
is
not
fully
supported
with
the
existing
cni,
but
the
good
news
is
that
overlay
and
some
other
parts
like
stateful,
bible
and
multicasting
load
balancing
those
type
of
things
are
sort
of
supported,
they're,
partially
supported
and
there's
some
gpu
like
accelerations,
but
also
supported
and
and
those
cns
are
sort
of
a
compatible
with
the
kubert.
But
but
again
it
doesn't
really
support
vpc
interview
model.
So
if
you
go
to
this
openstack
neutron,
it's
just
really
for
vm
networking
model.
C
So
it
is
provided
with
this
enterprise
model
and
the
implementation
interfaces
of
ml2,
and
they
support
a
bunch
of
like,
like
vm
native
networking
like
subnet
isolation,
underlay
and
cpip
assignment
and
security
group
and
so
on.
C
C
So
now
we
introduce
kubernetes,
actually
singular,
is
a
set
of
like
cni
plugins
that
realize
both
vm
and
control.
Networking
stack
and
the
cuber
cni
is
one
of
the
singular
cnn
plumbing
which
is
dedicated
for
supporting
the
vp
syncing
model,
and
we
have
another
cnn
called
current
cnn
and
support.
The
currency
model
is
very
similar
to
entry
and
this
model
leverages
obs
as
data
plane,
and
we
use
almost
as
in
controller
as
a
networking
platform.
C
So
this
is
one
of
the
deployment
examples
that
we
deploy
both
management
network
as
well
as
obs.
As
a
service
or
maybe
external
network,
and
then
we
use
acupress
cni
to
control
the
service
network
in
this
use
case,
and
so
so
now
we
introduce
this
previous
cni.
This
is
for
a
bunch
of
like
a
bbc
network
model
that
has
not
been
easily
realized
by
using
existing
cni's
and
most.
C
C
The
and
btk
and
the
dpo
hydra
upholding
is
also
in
our
world,
and
the
good
news
is
this:
cni
is
fully
compatible
with
the
covert
and
now
I
would
like
to
briefly
introduce
with
owners,
because
it's
our
network
platform-
and
I
think
maybe
some
people
may
hear
about
owners-
and
maybe
some
are
not
so.
C
I
will
provide
some
brief
introduction
of
donald's
own
says:
open
source,
skin
controller,
it's
just
for
fusion
for
a
scanner,
free
solutions
cases
and
they
provide
high
availability
through
the
series
control
plane,
and
this
is
for
carrier
grade
scene
controller
and
the
support
various
sales
funnel
protocols
like
openflow,
obviously,
b
and
p4
and
etc,
and
by
itself
is
provide
high
valuability
by
using
the
distributed
core.
C
C
So
so
the
high
availability
can
be
achieved
by
using
this
mechanism,
and
so
the
application
developer
no
need
to
care
about
these
availability
issues
and
then
another
irrational
that
that
we
choose
almost
is
like
it,
since
it
supports
various
sales
fund
api
and
there
are
a
lot
of
bunch
of
applications
available
on
top
of
owners
and
the
one
of
the
applications
called
charles.
C
It's
a
multi-purpose
sleep,
spine
fabric
management
solution
and
if
you
come
up
with
with
the
deploying
this
vm
networking
solution
platform,
somehow
you
need
to
configure
underlay
network
as
well
as
overlay
and
typically
those
configurations
supposed
to
be
done
in
different
platforms,
a
different
manner
and
different
like
the
control
panel.
But
using
by,
but
by
adopting
these
owners,
then
we
can
configure
and
control
overlay
the
undulator
just
in
one
like
a
network
platform,
so
no
need
to
switching
between
the
different,
the
control
panel
in
this
case.
C
So
if
your
fingers
work
by
trials,
you
can
visit
that
site,
and
this
is
the
diagram
that
shows
how
we
integrate
this
accuracy
and
now
with
ovs.
So
we
there
are
twofold:
to
integrate
this
cura
cni
one
is
we
interact
with
the
kubernetes
api
server
from
the
keyword
cni
perspective?
Typically,
we
receive
a
bunch
of
events
from
the
api
server
like
network
creation
and
volte
creation
of
security,
recreation.
C
The
type
of
event
can
be
received,
cni
and
then
by
by
using
that
event
to
be
calculate
the
flow
rules
and
eventually
install
that
rules
using
donors
and
almost
actually
in
charging.
It
interact
with
the
open
b
switch
using
the
obs
db
protocol
as
with
open
flow
protocol.
So
it's
in
charging
of
creating
bridges
and
attached
port
and
the
install
flows,
and
so
on
so
once
vm
has
been
created,
then
the
tubular
will
invoke
cni
commands.
C
I
think
that
we
are
using
obs
cni
from
keyword
convert
to
community
and
to
create
to
attach
the
port
to
the
to
the
obs,
and
this
is
example
of
using
both
colico
and
ovs
to
provide
the
management
azure's
service
network,
and
this
is
the
basic
tool
and
then
we
typically
deploy
the
appeal
verb
as
a
type
of
bootstrap
site
and
there
inside
there
there
is
a
leader,
elector
and
the
computer,
like
sidecar,
are
running
together
with
the
trooper
siemen.
C
So
this
is
summary
of
the
security
center
feature.
So
definitely
we
provide
ltl
fake
connectivity
by
using
a
bunch
of
custom
resource
definitions.
Actually
we
try
to
not
introduce
many
crds
and
we
try
to
reuse
existing
crds
in
from
the
community.
I
like
networking
definitions
from
motors
for
layer
three,
we
introduced
a
virtual
error
crd
and
the
faulty
ipcrt.
C
Also,
another
feature
is
that
the
ipad
address
preservation
so
once
so
once
we
so
if
we
spawn
the
vm
and
the
restart
vm,
typically,
it
gets
assigned
another
ip
address
because
it
runs
inside
part.
But
by
using
this
ip
address,
preservation
assignment
the
ipa
address
for
vm
will
be
statically
assigned.
So
even
if,
when
we
restart
vn,
the
vm
will
always
get
assigned
the
same
ip
address
and
we
realized
by
using
the
htcp
and
the
cloud
in
it
like
approaches,
and
this
is
the
source
for
the
crds.
C
So
this
is
a
currency
and
a
bit
of
vanilla
ovs,
and
this
is
the
example
of
constructing
a
tenant
network
and,
as
you
can
see,
there
are
two
types
of
content,
but
one
is
we
excellent
re-examine
is
denoted
as
a
green,
another
one
is
booth
and
those
network
will
be
served
by
using
the
integration
bridge
network
and
each
teletunnel
will
be
served
by
a
separated
integration
bridge,
as
you
can
see
here,
and
those
bridge
will
be
attached
to
the
bridge
tunnel
tunnel
bridge
and
for
external
connectivity.
C
We
also
introduced
another
concept
called
the
lightweight
hkwa
and
inside
gateway.
We
have
three
bridges:
one
is
the
bridge
visna
another
bridge
tall
tune,
another
bridge
and
the
all
gateway
alerted.
The
flow
rules
will
be
installed
in
the
bridge
it
and
the
nat
related
rules
will
be
installed
in
a
bridge
if
this
map,
and
so
on,
okay.
So
this
is
an
example
of
creating
a
prior
network.
This
is
we
we.
We
use
the
natural
attachment
definition,
crd
and,
as
you
can
see,
we
added
a
bunch
of
annotations
for
creating
the
network.
C
So
definitely
we
need
to
add
the
slider
and
the
gateway
ip
and
that
the
tool
and
the
whole
12-
and
here
this
is
very
similar
to
the
concept
from
the
openstack.
So
we
just
brought
all
the
like
networking
concept
from
from
the
slide
and
you
you
can
similarly
add
those
compensation
in
kybusine
as
well,
and
this
is
for
proactive
network.
C
C
You
also
need
to
specify
the
segment
id
and
so
on,
so
the
original
part
mode
is
similar,
but
since
the
row
of
each
bridge
is
different,
so
in
this
case
you
have
to
this,
so
this
type
of
prototype
is
associated
with
pr8,
but
while
the
title
is
associated
with
plus
the
segment
id
here
in
this
case
for
the
security
we
implemented
a
bunch
of
the
pipeline
and
they
support
various
like
a
full
base
policy
enforcement.
C
So
it's
for
source
and
destination,
iep
address,
exact
match
and
longest
briefing
match
and
the
protocol.
We
so
far
supported
four
types
and
directions
ingress
in
the
u.s
and
we
also
support
poor
range
and
I
think
code
match
as
well.
So
this
is
a
detailed
pipeline
that
how
we
implemented
this
is
the
security
be
using
obs
in
this
case,
and
in
order
to
do
that,
we
leverage
the
contract
of
this
module.
C
This
is
the
example
of
creating
sqlcr,
and
so
one
security.
You
can
contain
the
set
of
skill
group
rules
and
you
can
specify
details
of
these
rules
here.
So
in
this
example,
we
just
allow
ingress
by
traffic.
So
you
need
to
specify
the
pro
for
range
max
and
then
at
this
manner,
and
then
we
allow
all
of
remote
like
access
in
this
case.
C
However,
if
you're,
if
we're
using
vanilla,
obvious
one
of
the
pinpoint,
is
the
performance,
so
we
we
are
now
moving
forward
to
adopt
obviously
ptk
in
acute
cni
and
the
architectural
why
there
are
not
big
changes.
One
thing
is
we
have
we
are.
We
are
considering
to
use
the
user's
bcni
to
adjust
path
to
obs
bridge
with
the
rehost
user
interface
and
the
the
data
type
supposed
to
be
configuration
of
that
and
observe
that
configuration
needs
to
be
enforced.
C
Okay,
okay-
and
this
is
the
revised
architecture
for
the
obs
dpdk
and
the
several
differences
has
been
so
we
have
to
introduce
another
breach,
called
the
vr
tune,
tpdk,
and
we
have
to
configure
ip
address
for
this
bridge
in
this
case,
and
so
for
tenants,
look
is
fairly
simple:
just
go
for
this
tool
and
for
cross-tenant
communication.
C
C
So
underlay
network
is
very
simple:
there
is
no
you
you,
you
just
need
to
directly
connect
it
to
the
iot,
and
there
is
your
observing
the
network
like
a
public,
interconnectivity
and
so
on.
The
summary.
C
The
cuber
cni
to
provide
vm
networking
stack
specifically
for,
and
we
have
already
contributed
to
the
owner's
open
source
community
in
a
way
to
help
users
to
ease
the
migration
process
from
openstack
to
kubernetes
and
those
are
source
codes.
So,
if
you're
interested
about
contributing
to
the
community,
you
can
visit
your
sites
and
make
contributions
so
future
work.
C
What
we
have
country
planning
is
that
we
want
to
support
network
telemetry,
and
we
want
to
provide
some
reports
of
the
network
statistics
to
differ
from
users
and
then
another
thing
is:
we
want
to
provide
data
planning
saturation
and
for
this
year
q2.
We
want
to
integrate
that
with
the
obvious
dpdk
and
for
at
the
end
of
this
year.
C
We
want
to
upload
it
to
eventually
to
the
hardware
and
the
those
are
areas
that
need
to
have
from
the
community,
and
so
one
is
accelerable
balancer
for
10th
convention
cluster
and
the
recently
we
found
that
when
it
at
the
curb
committee
are
developing
the
cluster
api
provider
trooper.
So
we
hopefully
want
to
integrate
that
with
the
with
this
one
and
one.
C
We
also
want
to
enhance
the
security
by
introducing
the
ipsec
terminal
and
the
pro
security
and
the
vpn
service
and
the
firewall
service,
and
we
also
want
to
support
the
ipv3
from
the
visitors
textbook
as
well.
So
this
is
the
end
of
my
talk
and
if
you
have
any
questions,
feel
free
feel
free
to
ask.
Thank
you.
A
Thank
you
already.
I
mean
that's
amazing,
of
what
you
weaved
around
keyboard.
So
I
see
three
questions
in
here
is
the
cni
open
source?
If,
if
so,
could
you
please
provide
the
link.
C
Okay,
so
this
is
the
link,
I
think,
the
end
of
fiat
print.
My
slide,
you
can
find
the
link,
and
actually
this
has
been
including
that
almost
project.
So
in
order
to
use
this,
you
know
you
have
to
compile
the
and
we
actually
upstream
all
the
source
code
to
the
unknown's
master
branch
and
yeah.
You
can
just
directly
put
another
current
and
the
compiler
and
they
can
deploy
that
on
top
of
kubernetes.
A
C
So
do
you
mean
attaching
that
to
the
part
and
providing
some
part
communication
as
well?
Okay,
actually,
that
is
possible.
Yeah,
that's
that's
possible,
and
actually
you
can
attach
it
both
to
pad
or
vm.
So
we
actually
do
not
really
care
about
the
vocal.
So
you
also
yeah
you
can
use
that.
As
for
power
communication
as
well.
C
So
for
with
the
vanilla
ovs,
based
on
our
like
the
preliminary
performance
evaluation,
we
found
that
with
the
10
g
d
card,
it
only
support
like
seven
around
seven
g,
like
a
circuit
using
vxlan
tunnel.
C
If
you
are
using
vlan,
then
the
performance
will
be
increased
and
if
you
are
using
the
provider
network,
I
think
the
performance
would
be
more
than
similar
to
the
parameter
like
case
so.
But
if
we
move
to
the
obs
dpdk,
then
it
can
show
a
lot
of
cpu,
but
it
almost
preserved
like
like
a
guaranteed
the
performance
like,
oh
so
wrong,
so
for
10
gd
car
it
can
provide
almost
like
line
rate
like
from
maybe
not
99.5
gigabyte,
but
one
limitation
is
that
it
has
it
gonna
intentionally
use
cpu.
C
So
if
you
are
lacking
of
cpu
resource,
our
recommendation
is
try
to
upload
these
to
the
dpu.
So
if
you're
using
dpu,
then
all
the
traffic
will
probably
go
through
directly
through
the
smart
link.
Then
it's
going
to
free
up
a
lot
of
like
the
cpu
resources
and
that
the
the
performance
can
also
be
guaranteed.
A
All
right
there's
the
last
question,
but
we're
a
little
bit
short
on
time.
So
one
thing
I
want
to
call
out
is
it's
great
to
see
your
work.
I
would
encourage
you
to
add
yourself
to
the
kubert
adopters
file.
If
you
go
to
the
keyword
kubert
repository,
if
you're
open
to
it,
then
you
can
open
a
pr
to
the
adopters.md
file
to
add
yourself
to
it,
because
you
seem
to
adopt
the
keyboard.
C
Okay,
can
you
answer
the
question
because
it's
or
do
I
need
to
answer
yeah?
I
I
think
we
I
need
to
discuss
it
with
the
hour
management
and
to
decide
whether
to
open
it
to
the
to
the
adapter,
and
I
will
come
back
too
soon.
Once
I
get
answered
awesome.