►
From YouTube: Network Service Mesh BoF Meeting - 2019-03-19
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
A
A
A
C
B
All
right,
so,
let's
go
ahead
and
get
started
with
the
main
with
the
main
things
then,
so
we
have
three
recurring
talks
now
one
is
the
innocent.
Well,
this
this
particular
meeting.
We
have
the
network
service
mesh
document
on
Wednesdays
and
the
end
the
same
use
case
on
Fridays,
so
we
also
are
coordinating
with
the
CNF
test
bed,
birds
of
a
feather
so
and
that's
every
other
week.
So
all
of
these
calls
have
been
at
8:00
a.m.
Pacific
time.
B
Up
to
the
very
bottom,
because
it's
otherwise
it's
gonna
get
missed
anyways,
so
we
have
on
March,
28th
and
29th.
We
have
service
mesh
day,
of
which
we
have
a
talk
in
the
in
the
surface
mesh
day,
where
we're
going
to
be
showcasing
envoy
and
I.
Never
service
mesh
working
together,
oh
great,
so
I
have
a
little
bit
of
work
to
finish
on
that.
But
I
have
a
clear
path.
Now
we
also
have
a
Intel
out-of-the-box
network
developer,
meetup,
that's
coming
up
on
April
the
second
the
day
before
ons.
B
Definitely,
if
you're
going
to
be
in
town
on
that
day
and
are
willing
to
come
a
day
earlier,
then
I
would
really
appreciate
having
people
there
to
talk
to
people
about
network
service
mash
and
to
them
to
let
people
know
that
like
what
what
it
is
you're
expecting
to
get
out
of
it
or
help
people
learn.
We
have
ons
coming
up,
of
which
we
have
a
series
of
of
talks
accepted
and
a
panel
discussion
where
I'm
creating
these
topics
will
come
up
based
upon
the
people
there
we
have.
B
We
also.
We
should
also
have
a
booth
demo
as
well
and
permit
I
need
to
coordinate
on
that.
So
Kremlin
you'd
have
a
conversation
soon,
so
we
can
make
sure
that
gets
real,
properly
yep.
We
we
have
MPLS
Sdn
EFI,
coming
up,
no
talks
there
as
far
as
I
know,
but
topic
should
be
interesting.
We
have
container
world
2019
coming
up
with
a
talk
accepted
by
pram
from
Lumina
and
so
that'll
be
in
Santa
Clara
California.
B
B
We
also
have
cube
con
and
Shanghai
coming
up,
of
which
the
call
for
papers
already
closed,
I'm,
not
aware
of,
and
he
talks
to
have
been
submitted.
But
if
anyone,
if
anyone's
attending,
let
us
know
and
we'll
if
we
will
we'll
see
if
we
can
pair
you
up
with
with
interesting
people,
we
have
ons
Europe
and
emp'd
work
Belgium
coming
up,
which
the
call
for
paper
is
currently
open
and
closes
in
June
16th
feel
free
to
talk
in
the
area.
We
also
have
M
EF
in
2019
and
Li,
which
was
very
inconvenient
ly.
B
The
look
in
the
community
located,
but
rather
the
timing
is
inconvenient
because
of
cute
cotton
in
San
Diego,
so
we'll
have
a
little
bit
of
fun
coordinating
between
the
two
of
them,
because
we
we
care
about
both
we
finally
have
and
so
anyways.
If
there's
any
other
events
that
anyone
is
aware
of.
Definitely
let
us
know
added
to
this
list
and
with
that,
let's
head
to
the
CN
CF
project
proposal.
So
can
you
open
the
project
proposal
up?
Please.
C
B
C
C
B
This
proposal
is
different,
so
the
proposal
that's
up
on
the
screen
is
a
is
the
CN
CF
proposal
for
being
part
of
the
TOC
so
for
the
proposal?
I,
don't
I,
don't
believe
I,
don't
know
if
it's
been
submitted
yet
I,
don't
believe
it
has
so
for
this
one.
The
best
way
to
go
to
to
manipulate
this
one
is:
can
you
can
you
go
to
the
very
top,
so
we
get
the
URL
of
it.
So
do
you
see
it's
under
network
service
smash,
slash
network
or
sorry,
it's
under
ed,
Warnecke,
slash
TOC.
A
On
a
producer
it,
the
link
is
here
in
the
in
the
meeting
minutes,
so
the
actually
I
had
a
chat
just
before
the
call
with
with
that
he
saw
that
he
cannot
join,
but
the
thing
is
that
we
have
very
last
week
he
was
at
the
episodes
leadership
summit
and
he
was
able
to
reach
out
to
two
of
the
to
see
members
of
the
CN
CF
TLC
members,
and
we
have
kind
of
secured
their
sponsorship.
They
are
put
their
names
are
put
here.
A
We
have
job
beta
and
Matt
Klein,
so,
based
on
this
and
some
some
other
communications,
the
proposal
is
that
we
do
a
final
kind
of
review
today
to
the
best
of
our
our
possibilities.
And
if
someone
has
I
mean
someone
raises
some
big
concern
and
wants
to
change
something
we
can
postpone.
But
if
we
kind
of
agree
that
this
is
the
text
that
you
want
to
submit,
it
will
push
it
tomorrow.
C
Not
certainly
totally
spot
on
times,
Joe
I
was
also
talking
to
Joe
in
person,
but
one
of
the
general
feedback
I
also
heard,
was
it's
good
to
add
sort
of
the
distributed
cloud
use
cases.
At
least
you
know,
planet
spoke
to
him.
A
broader
picture.
I
do
want
to
make
sure
that
some
of
those
is
reflected.
It's
just
that.
Probably
we
may
need
a
little
bit
of
time,
thus
I'm
requesting
it's
a
little.
Probably
I
may
need
more
than
a
week.
Sorry
more
than
a
couple
of
days.
C
B
The
the
only
problem
we're
gonna
run
into
potentially
is
that
we're
we're
hoping
that
we
can
get
at
this
part
of
the
CN
CF
before
cube
con,
and
so,
if
we
get
it
approved,
then
that'll
just
bring
like
that'll
just
be
a
lot
of
good
things
around
that
overall.
So
the
problem
we're
running
into
is
because
the
TOC
had
an
election,
they
have
a
backlog
of
agenda
items
and
projects
and
so
for
us
to
to
join
in
we're.
B
C
Great,
how
good
this
give
me
24
hours,
let
me
take
a
stab
I,
can
understand
the
urgency.
So
let
me
let
me
take
a
stab
at
that,
but
I
know
it
took
the
a,
but
it's
just
being
crazy.
So
let
me
take
a
stab
at
that
part
and
then
begin
finalized
tomorrow.
It
will
not
be
a
big
addition,
but
you
know
the
use
case
is
next
space
yeah.
A
So
the
way
that
that
we
were
that
we
were
doing
it
up
to
now
is
that
you
just
go
and
add
your
comments.
So
I
think
that
my
comments
were
some
somewhere
here
in
the
in
document,
and
everyone
else
is
just
saying:
okay,
can
you
change
these?
Can
you
change
that,
and
here
I
have,
for
example,
put
my
comments
about
the
release
process
and
then
it
incorporated
it's
all.
D
A
B
B
B
Actually,
well,
we
should
probably
as
well
or
request,
comes
in
if
there's
any
changes
that
are
there
are
necessary,
then
they
no
longer
go
to
add
Warnecke,
slash
toc
for
the
pool,
requests
and
or
comments.
Instead,
they
will
go
to
the
to
the
pool
request
from
from
the
cnce
singing
ncf,
slash
TOC,
so
I.
So
my
guess
is,
we
could
probably
still
add
some
stuff
through
that
particular
process,
but
we
probably
want
to
keep
those
two
minor
edits
as
much
as
possible
unless
they
flag
something.
That's
really
urgent.
Yeah.
C
So
let
me
do
this
trade,
then.
So
what
I'll
do
is
I'll
work
with
cream
on
the
changes
right
on
Esther
day.
I'll
also
make
sure
that
I
have
a
quick
sync
up
cause
at
the
end
on
the
change
is
proposed,
so
that
I
mean
basically,
you
know
we
can
get
to
it
much
quickly
and
then
target
submission
tomorrow.
We
can
all
review,
probably
offline
and
make
a
submission
tomorrow.
C
It
just
that
just
didn't
get
the
time
to
actually
make
that
nice,
edit
they're,
looking
to
sort
of
get
the
use
case
to
a
point
where
we
have
some
good
clarity.
I
think
it's
parity
is
that
we
just
want
to
reflect
it.
It's
probably
an
hour
to
work,
but
you
want
to
get
some
crisp
content
here
back
from
the
use
case
and
the
architecture
impact
were
able
to
create.
That's
it
don't.
A
C
B
So
so
a
couple
of
things
on
this
as
well,
so
please,
please
make
sure
that
you
go
over
the
the
contents,
for
those
of
you
are
who
are
listening
in
and
make
sure
that
make
sure
and
any
type
of
help
bonus
would
be,
would
be
appreciated,
whether
whether
it's
some
substantial
addition
or
even
just
a
minor
typo
or
wording,
if
you
could,
if
they're
working
to
be
more
clear,
like
anything,
is
helpful
in
the
space
and
so
in
terms.
A
few
couple
things
I
want
to
point
out
as
well.
B
So
the
current
sponsors
that
we
have
from
from
the
list
are
very
strong,
so
we
have
Jo
beta
I
listed
as
a
sponsor,
who
is
the
one
of
the
founders
of
minetti's
and
that
client
who's,
the
founder
of
NY,
so
I
think
we
have
some
pretty
good
backing
on
the
for
the
initial
for
the
initial
entry,
and
so
it's
primarily
up
to
us
in
order
to
in
order
to
hit
it
out
of
the
park.
So
if
and
we
also
have
as
existing
sponsors,
Bell
Canada,
Cisco
and
da
ki.
B
Think
we
should
I
think
we
should,
let's,
let's
let's
get
the
the
Edition
from
from
romkey,
so
you
get
that
in
as
soon
as
possible
and
I
think
the
rest
of
it.
We
can
probably
we
can
probably
do
online,
but
let's,
let's
make
sure
that
we
at
the
very
minimum
commit
to
some
times
and
to
to
get
to
get
this
thing
in
and
of
course,.
C
B
Yeah
I
think
I
think
that
sounds
good.
So,
let's
so,
let's
go
ahead,
and
so
ed
is
out
for
the
morning.
So
I
don't
know
if
he's
gonna
be
around
in
the
afternoon
or
not,
but
yeah.
If
you
want
I
can
I
can
provide
some
feedback
as
well.
You
also
and
I'm
pretty
sure
ed
is
getting
a
lot
of
rest
right
now,
so
he
can
be
back
in
good
form.
B
C
B
B
A
At
least
name
poll,
so
essentially
because
they
were
not
really
many
feedback
on
naming
I
effectively
chosen
propose-
and
this
is
already
merged
in
the
in
the
main
report
that
we
use
the
the
names
of
the
constellations
like
the
Latin
names.
So
I
chose
three
names
here
proposed,
like
you
know,
start
with
a
so
my
proposal
here
would
be,
if
anyone
has
any
preference,
just
put
your
names
be
decides.
A
We
probably
do
this
for
the
next
couple
of
meetings,
just
to
gather
some
names
and
then
we'll
just
select
one
that
that's
that's
the
best
I
can
think
of
other.
The
other
thing
is
to
just
send
an
email
there
to
the
Trudeau
to
the
group,
but
I.
Don't
think
that
this
works
very
well
or
maybe
I
can
create
a
doodle
poll.
I,
don't
know
what
what
would
be
the
best
I
think
that
the
me
doing
the
meeting
minutes
should
be
should
be
good
enough
to
just
gather
this.
Like
type
of
voting.
E
A
B
A
A
A
B
D
A
D
B
Ok
yeah,
so
it's
something
that
would
be
good
to
do
would
be
to
add
that
to
pull
requests.
So
we
can.
We
can
set
up
a
little
checker
that
just
does
a
go
mod
vendor
so
because
then
that'll
download
all
of
the
dependencies
that
are
listed
there
and
then
and
then
will
then
run
the
license.
Scanner
on
that
vendor
list.
A
A
Yeah
I
think
that
that's
all
I
wanted
to
do
to
ask
here
and
to
just
discuss
a
little
bit
wrong
key
commentator
plane.
Okay,
do
you
need
to
preserve
something?
Is
there
a
link
that
I
can
open
for
you?
No.
C
It's
more
of
a
discussion,
in
fact,
so
maybe
I
would
like
to
actually
take
some
I'm
looking
at
the
document
and
doing
the
update
given
the
urgency.
C
The
key
was
essentially
one
of
the
discussions
that
came
up
and
use
kids
come
discussions
was
essentially
a
common
data
plane
API
around
abstracting
features
such
as
you
know,
fate,
limiting
or
other
aspects
across
software
and
hardware
implementation,
your
SRO,
your
smart
mix.
You
know
you
have
these
different
implementations,
then
the
idea
was:
how
can
innocent
provide
value
through
a
common
data
plane
API?
That
was
the
essence
of
set,
but
we
didn't
get
into
details.
C
So
that's
what
I
thought
we
could
deep
dive,
but
my
mind
is
now
on
the
the
absolute
near-term
target
and
have
several
things
to
take
care
of
today.
So
if
the
team
wants
to
I
mean
discuss,
continue
discussing
perfect
time,
but
I
may
not
be
able
to
fully
driver,
I
mean
able
to
drive.
I
can
listen
in,
but
not
drive
this
today.
Sorry
about
that
yeah.
So.
A
F
You
want
to
pull
up
the
the
glossary
real,
quick
just
so.
A
F
Still
need
like
ed
Frederick
and
you
to
give
it
one
more
glance
over,
but
the
data
plane
has
been
like
this
super
confusing
like
point
of
contention
in
a
lot
of
the
document
calls,
and
so
what
I
did
was
I
went
through
a
lot
of
the
specs
and
I
broke
down
like
what
we
were
already
working
on
all
those
little
individual
components
and
then
from
an
innocent
perspective,
we're
kind
of
going
with
you
know,
assuming
that
it
we
all
agree
on
it.
F
Tomorrow's
call
the
the
data
plan
is
basically
anything
that
provides
you
all
of
these
things.
So
I
know
we
were
talking
about
like
the
whole
concept
of
like
a
mechanism
or
a
channel,
but
actually
found
that
and
I've
been
updating
it
in
the
the
PowerPoint
stuff.
So
I'll
add
this
stuff
to
this
document
later,
but
local
mechanism
is
actually
like
the
name
of
something
in
the
code
and
it's
actually
90%
of
the
time.
What
we're
talking
about
when
we
talk
about
a
micro,
wire
or
all
this
stuff,
do
you
have
yeah.
B
F
We
go
instant
like
the
thing
is:
is
whether
I'm
hitting
like
an
end
API
service,
which
is
in
reality
from
a
network
standpoint,
a
control
plane.
You
know
it
doesn't
matter
as
long
as
an
SM
can
reach
it
to
something
whether
it's
a
physical
device,
whether
it's
a
kernel,
whether
it's
an
SDN
abstraction.
If
it
gives
me
the
connections,
the
wires
and
the
network
service
as
a
whole,
then
it's
kind
of
the
data
plane.
F
Basically,
it's
going
to
make
a
lot
of
networking
people's
heads
explode,
but
in
reality,
in
sm,
is
going
to
go
and
make
a
request
to
something
and
basically
what
it's
going
to
request,
connections
and
wires,
which
will
ultimately
turn
into
a
service,
and
we
don't
care
how
we
get
that
theoretically,
as
long
as
whatever
we're
making
the
request,
who
provides
that
for
us.
Does
that
make
sense
so.
G
Jeff,
this
is
a
funny
here
from
snap
crowd.
My
previous
life
I've
worked
on
a
lot
of
vien
apps
for
the
telcos
at
the
high/low
liability.
Jeff,
no
disagreements
at
all
but
I
think
what
me
and
romkey
and
am,
and
others
for
discussing
was
how
to
create
that
abstraction
layer
which
is
as
simple
as
the
goal
you
stated
and
still
exploit
the
the
functional
advantages
of
a
let's
say:
ADP
DK
or
a
VPP
or
smiling
no.
F
You
have
to
ask,
though,
other
than
the
fact
that
you're
just
gonna
request
a
network
service.
We
have
to
kind
of
start
to
delineate,
and
this
is
something
I've
whined
about
to
IDI
and
Frederick
is
I,
feel
like
NSM
is
trying
to
solve
some
of
kubernetes
data
playing
challenges
as
opposed
to
making
kubernetes
solve
its
own
data
playing
challenges.
But
my
answer
to
your
exact
question
would
be
device
plugins
like
whether
it's
I'm
requesting
it
from
kubernetes
I'm,
requesting
it
from
OpenStack
or
I'm,
requesting
it
from
odl
right.
B
F
If
I
have
SRV
capable
compute
like
I've
got
Nix
with
you
know
in
virtual
devices,
or
can
you
know
containerize
devices
that
have
the
drivers
in
them
and
stuff
a
device
plug-in
should
say
I
have
this
resource
available
in
SM?
When
you
come
in
with
your
NSC
or
your
NSC,
you
should
say:
I
need
to
deploy
this
network
service
and
it
should
make
a
request.
F
G
Yeah,
the
devil
is
going
to
be
in
the
details.
How
do
how
do
you
make
sure
that
network
service
mesh
is
able
to
run
on
all
kinds
of
hardware,
accelerated
or
ADP
DK
accelerated
implementations?
You
know
if
you
go
back
up
meh,
my
F
was
as
we
host
user
and
all
that
you
know
that's
where
the
details
are
going
to
be
how
the
CN
f's
or
the
containers
or
the
parts
are
going
to
interface
with
the
data
plane
and
still
extract.
The
hardware
acceleration
that
are
present
downstairs
is
going
to
the
come
so.
B
What
so,
at
this
particular
moment,
we
focused
primarily
on
enablement
and
not
trying
to
make
them
have
a
uniform
path
once
you
perform
that
enablement.
So
if
you
have
something
that
speaks
DVD
K
or
you
know
hey,
you
have
something
that
knows,
how
do
we
interact
with
SRO
V
or
a
kernel
interface
as
an
example
like
it
has
to
know
how
to
actually
properly
interact
with
those
like
it
doesn't
obviate
the
need
for,
for
that
and
I
feel
like
there
is.
B
There
is
a
good
opportunity,
though,
for
a
for
a
simple
CNF
library
that
knows
how
to
deal
with
these
type
of
these
type
of
things.
So,
in
other
words,
you
mask
for
a
connection
and
then
it
provides
you
with
the
ability
to
to
work
with
whatever
it
is
that
you
that
your
hardware,
the
courts
and
ends
up
abstracting
it
out
but
I
I,
think
that
that
would
be
something
that
network
service
mesh
would
enable
I'm
not
entirely
sure
that
would
be
something
that
a
network
service
mesh
itself,
at
least
not
there,
the
the
main
component.
B
It
could
be
a
side
project
on
the
end,
the
Sam
that's
attached
to
it
as
a
library.
That's
that
could
be
something
that
that
could
that
could
support,
but
the
core
get
network,
service,
mesh
or
sorry
get
network
service
or
connect
me
to
this
data
plane
or
so
on.
I
think
I
think
we
should
be
careful
to
not
try
to
to
merge
these
all
into
one
and
and
rather
force
all
the
users
into
into
one
format,
format
or
one
mode,
because
even
with
a
nice
client
library
that
does
99%
of
what
you
want.
B
F
Other
thing
is
I,
don't
know
if
everybody
knows,
but
I
am
I,
have
contacts
and
Intel
being
the
big
service
provider.
Guy
I
have
access
to
their
little
brain
trust,
and
when
we,
you
know,
we
mentioned
DP
DK
a
couple
times.
So
the
problem
is:
does
kubernetes
doesn't
give
a
crap
about
Numa
zoning
right
and
so
say.
F
Turn
on
huge
pages,
I
deploy
a
container
I
get
you
know
a
PCIe
Lane
here,
I
get
a
node
over
here
and
I've.
Get
the
memory
Channel
over
here,
and
none
of
them
line
up
well,
see
BTK,
says
I'm,
not
passing
any
of
your
packets
because
you
don't
meet
my
Neumann
restrictions,
so
there's
even
like
with
these
API
acknowledges
and
stuff
with
they're
quickly
solved.
F
So
we
are
trying
to
work
with,
like
some
of
the
you
know,
upstream
contributors
on
fixing
some
of
these
problems,
or
at
least
letting
us
toggle
some
of
the
restrictions
off.
But
the
fast
data
plane
stuff
in
kubernetes
is
still
a
challenge
just
because
of
how
kubernetes
wants
to
place.
Pods
and
namespaces.
G
Yeah
Jeff
I
think
I
kind
of
agree
most
of
your
stuff.
What
I
was
trying
to
point
out
was
it's
probably
a
subtle
point.
It's
about
the
TX
and
rx
part
of
a
container
and
in
the
last
meeting,
I
was
highlighting
that
trying
to
marry
yourself
into
let's
say
a
Saraiva
based
data
plane
has
its
implications
and
so
we'll
be
with
DP
DK
or
the
upcoming
smart
Nix,
and
so
on.
So
the
critical
path,
the
TX
and
rx.
Let's
say
it's
a
going
to
be
a
mem
I
for
a
V
host
user.
G
Let's
try
to
create
a
abstract
interface
so
that
the
containers,
really
the
pods,
really
don't
care.
What
kind
of
a
underlying
data
path
they
are
dealing
with.
I'm,
not
talking
about
get
me
a
connection.
That's
a
control,
plane
function,
I
might
talking
about
the
tx/rx
and
the
in
the
packet
interface
that
the
parts
will
will
have.
You
know
right
now
with
the
evolving
various
flavors
of
data
planes.
G
Even
on
the
vm
side,
I
see
there's
so
many
ways
to
do
the
data
plane
inter
interaction
and
that
is
causing
lot
of
vendor
lock-in
and
few
other
issues.
So
that's
where
I
was
trying
to
draw
a
caution,
but
I
can
of
agree.
You
know
it
has
to
be
at
an
abstract
level
and
the
network
services
shouldn't
care
about
the
lower
level
details.
B
E
F
E
B
B
Odb
DK
is
fundamentally
broken
in
this
environment,
but
I
have
access
to
srl
B,
so
will
will
give
the
user
access
to
SRS
r
io
v
as
part
of
that
and
then
work
work
with
kubernetes
and
the
device
bug
in
and
so
on.
In
order
to
drop
in
that
that
device
into
into
your
pod
and
then
the
the
pod,
the
software
in
the
pod
is
responsible.
It'll
be
told
this
is
an
SR
le
device,
so
the
software
in
the
pod
is
then
responsible
for
interacting
with
that
device
in
the
correct
way.
B
That
would
not
rely
on
an
Asst
on
NSM
itself
to
provide
that
that
functionality
other
than
perhaps
being
a
client
or
being
able
to
to
provide
the
the
generator
on
the
endpoint
and
and
so
on,
but
but
effectively
it
would.
It
would
not
be
part
of
the
the
core
and
a
SEM
server
aspect
of
it
if
that
makes
sense,
make.
F
A
suggestion
because
I
mean
this
is
something
that
like
ian,
has
been
like
losing
his
mind
over
as
well
like
this
exact
same
thing.
Is
we
kind
of
glossed
over
some
of
the
details
on
how
like
from
a
networking,
sneaky
networking
person's
perspective?
The
data
planes
supposed
to
work
like
the
actual
forwarding
of
packets
like
I,
was
lured
in
you
know
when
I
saw
this
at
cube
con
with,
like
these
delusions
of
grandeur
for
CNS
and
I,
think
that's
what
kind
of
drew
a
lot
of
us
in
here.
F
I
think
this
may
be
part
of
the
use
case
goal
and
romkey
thanks
for
shifting
it
to
Monday
I
think
I'll
be
able
to
start
attending.
Maybe
for
now
we
can
start
creating
a
list
so
we're
at
least
identifying
what
challenges
are
going
to
arise
around
like
data
in
the
forwarding
sense
of
the
term
and
not
from
the
network
sense
of
the
term
network
service
missions
and
just
start
chronicling
that
and
stuff,
because
I
think
part
of
the
problem
is
a
lot
of
times
the
more
application
you
people
are
like
yeah
yeah.
F
We
get
it
that
this
is
a
concern,
but
we
don't
care
about
it,
which
is
fine,
but
I.
Think
those
of
us
that
care
about
our
packets,
getting
in
and
out
of
these
servers
really
really
fast.
Just
we
get
like
the
nervous
tic
when
we're
not
addressing
it.
So
maybe,
as
part
of
this
use
case,
we
start
looking
at
like
what
these
local
mechanisms
are
and,
like
you
know,
it'd
be
nice.
F
If
this
library
did
exist
because
then
you
could
just
drop
it
in
when
you're,
creating
an
NSC
like
all
this
library
get
them
mif,
get
an
SRV
virtual
fort
or
you
know,
injected
into
this
namespace,
etc,
etc.
Right,
but
I,
think
part
of
the
problem
is,
is
we
just
keep
glossing
over
it
and
we
told
people
that
this
was
going
to
help
them
solve
CNF
challenges?
And
so,
if
we're
gonna
solve
CNF,
this
you
know
to
the
use
case.
B
And
and
I
want
to
be
careful
with
my
wording
on
this,
like
I,
think
it's.
What
you
described
is
an
incredibly
important
problem.
So
the
question
is:
is
that
something
that
NSM
as
it
is
now
should
solve,
or
is
out
of
something
that
NSF
should
work
with
something
which
could
be
an
NSF
library,
a
client
library
that
that
solves
that
problem?
So
so
why?
B
That
knows
how
to
to
solve
these
these
type
of
problems
and
not
not
make
the
NSM
core
try
to
solve
everything,
because
and
now
we're
talking
about
overcomplicating
lee
in
the
system,
but
I
do
think
it's
an
incredibly
incredibly
important
problem
to
solve,
because
even
the
rubber
has
to
hit
the
road
somewhere
and
that's
it.
That's
that's
the
that's
the
spot
in
terms
of
the
in
terms
of
the
CNF.
So
so
it's
definitely
something
that
we
we
need
to
do.
We
definitely
need
to
address.
A
E
Yes,
sorry,
one
extract
I
will
I
really
like
Frederick's
point
of
view.
One
thing
I
would
like
to
refer
about
how
to
integrate
all
those
technologies
and
the
data
plane
inside
is
a
lieutenant
Lee
Gatto.
There
was
a
people
proposed
in
BP
pmmif
and
the
understood
that
sometimes
the
pod
doesn't
know
me
my
F,
and
you
need
to
find
a
way
to
make
it
work,
and
he
actually
said-
and
let's
build
an
abstraction
that
you
add
to
the
path
that
makes
it
ma'am.
E
I
have
compatible
way
down,
I
think
to
build
a
completely
to
team
within
a
part
that
part
that
was
upstream
part
of
the
negative
project.
I
see
the
same
as
an
SM.
So
anybody
else,
because
there's
something
else
but
right
now
in
that
deaf
people
are
proposing
AF
grass
or
you
don't
even
have
a
networks
back
and
be
in
the
container
that
will
come
up
over
all
of
its
committees.
E
At
some
point
there
should
be
er
her
own
backward
for
those
kind
of
technologies
that
the
day
you
have
them
how
to
make
them
supportable
in
MSM.
But
that's
not
part
on
an
essential
work
should
should
focus
on
the
trick
is
really
getting
access
from
the
external
world
into
a
community
spot
where
you
have
CNS
and
then
get
out
and
how
you
can
change
the
NS
within
the
community's
path
or
multiple
clusters
and
out
to
bring
that
out,
build
that
I.
E
B
B
Okay,
yeah,
so
I
think
I.
Think
a
important
part
in
this
is
that
we
definitely
a
lot
of
this
stuff
is
still
uncharted
territory
in
the
CNF
world.
So,
for
example,
there
are,
there
are
people
who
have
loaded
the
DBK
within
a
within
a
container
and
they
prototype
that
and
we've
had
some
prototypes
with
loading
these
some
of
these
things
within
within
pods
but
with,
but
with
some
very
severe
workarounds
Taylor
can
tell
you
all
about
it
later
on.
If
you
ask
them,
but
a
large
part
of
this
is
like
how
do
we?
B
How
do
we
land
ourselves
to
a
point
where
we
can
start
to
productize
and
start
to
work
out
where
the
limitations
are
and
we're
trying
to
address
some
of
them
with
Intel
as
an
example
with
DB
DK,
where
and
with
Cooper
and
with
kubernetes,
where
you
have
like
Numa
alignment
problems
and
Mike?
How
do
you
get
new
alignment
or
simultaneously?
B
How
do
you
make
sure
the
DB
DK
works,
even
if
there
isn't
an
alignment
problem,
because
you
can
still
get
a
significant
performance
increase,
even
if
the
alignment
isn't
there
so
there's
a
lot
of
questions
that
are
that
are
starting
to
come
up
with
with
challenging
the
initial
assumptions
of
each
of
these
projects,
and
so
we
so
we
have.
We
have
a
lot
of
work
to
go
through
in
order
to
in
order
to
make
some
of
these
particular
use.
Cases
works.
B
That's
going
to
involve
the
broader
community,
so
what's
great
about
this,
though,
is
that
we
have
in
this
call
we
have
the
people,
the
experts
necessary
and
the
people
with
the
right
leverage
that
I
think
we
can
impact
and
enable
these
changes
and
provide
guidance
to
the
rest
of
the
community
as
well.
So
I
think
that
it's
gonna
be
very
important
for
us
to
have
these
type
of
discussions.
How
do
we?
How
do
we
you
do
as
early?
How
do
you
do
DB,
DK
and
drive
these
type
of
things?
B
Have
someone
have
a
have
a
person
or
a
small
group
basically
become
the
champions?
That
of
that
topic?
Go
and
and
cover
the
entire
end-to-end
use
case
not
just
be
like.
This
is
just
a
client
side
but
say
this
is
how
we
do
it
from
from
start
to
end
production
and
and
so
on,
with
the
idea
of
being
to
provide
guidance
for
for
that
level
and
I.
B
Think
if
we,
if
we
take
this
approach
a
few
times,
you
know
specifically
with
with
mi
FDB
TK
s,
rlv
and
and
so
on,
I
think
that
will
start
to
land
into
a
into
a
point.
Where
we'll
see
those
commonalities
and
then
we'll
have
the
right,
then
we'll
have
the
right
set
of
knowledge
centralized
enough,
then
we
can
effectively
go
and
build
such
a
library
or
even
work
out.
B
Does
it
make
sense
to
build
such
a
library
but
I
think
it
does
and
and
go
and
build
that
thing
and
and
solve
the
problem
across
the
entire
board?
Alright,
so
I
think
we
have
a
real
opportunity
here,
so
so
I
think
we
where
we
definitely
need
to
start
focusing
on
some
of
these
questions,
especially
now
that
we're
starting
to
get
close
to
providing
the
initial
did
we
call
it
an
alpha
or
a
beta,
we'll
call
it
the
equip
D
a
release.
F
I
point
out
one
thing
to
you
on
the
concept
of
like
a
library
right,
it's
kind
of
like
the
methodology
that
the
DB
DK
community
took
at
is
since
we're
pushing
in
SM
as
a
cloud
native
networking
solution
right.
Part
of
that
is,
if
you're
someone
who
doesn't
want
mutability
in
your
world,
and
we
make
it
so
that
it's
not
a
core
function
of
innocence,
then
that
abstraction
is
there.
The
simplicity
remains
and
the
standard
applications
people
can
come
in,
get
a
network
service
and
go
about
their
day.
F
For
those
of
us
who
constantly
want
to
break
our
infrastructure,
we
can
import
these
libraries
and
start
adding
more
knobs
to
turn
more.
You
know,
pod
manipulations
to
like
bear,
etc,
etc.
And
then
it's
on
all
of
us
obnoxious
service
providers
to
make
life
more
complicated,
well,
leaving
in
a
sim
and
its
natural
state,
simplistic
for
the
rest
of
the
world.
Yeah.
B
A
A
G
Yeah
would
funny
here
again,
I
would
like
to
reiterate
and
I
reiterate
what
Devon
made
the
comment
that
he
made.
You
know
it's
valid,
that
we
need
to
not
be
really
reinventing,
how
deep
it
ek
and
all
those
things
work,
but
I
think
it's
important.
We
define
how
we
are
going
to
use
various
such
projects
that
are
going
on
in
parallel
and,
more
importantly,
are
we
abstracting
enough
so
that
they
see
Annette's
that
are
riding
on
the
network
service?
G
D
Actually,
interestingly,
I
was
starting
to
few
folks
at
Intel
and
they
were
indicating
they
may
join
networks
of
this
mesh,
which
is
a
good
news
for
the
community,
and
one
other
thing
is
probably.
We
can
also
also
invite
MJ
goo
evangelist
SDP
decayed
to
some
of
the
calls
to
get
this
input.
We're
in
yes,
insights
into
how
things
would
look
like,
at
least
from
my
DP
DK
perspective.
You.
G
D
Okay,
I
guess
they're
on
the
top
of
the
hour
already
so
one
quick
question:
sorry,
it
is
not
related
to
the
topic,
but
this
is
related
to
the
CN
CF
proposal.
So
I
just
went
through
the
proposal
and
I
believe
it's
quite
good
and
then
it
addresses
most
of
the
common
issues.
So
the
reason
I'm
asking
this
question
was
before
we
make
any
changes
to
it.
If
you
feel
that
we
can
probably
maintain
at
this
level
of
abstraction
I
think
that
would
be
good.