►
From YouTube: Network Service Mesh WG Meeting - 2018-09-21
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
We've
got
a
lot
of
folks
who
are
new
to
the
call
today
and
I'm
delighted
to
have
you
normally
Ingrid
counts,
runs
the
meetings
and
he's
very
good
at
it.
So
if
it
seems
a
little
bit
like
I'm
catching
up,
it's
because
usually
we
have
really
good
meeting
runners
running
the
meetings
so
one
second,
while
I
share.
A
So
this
is
I
just
found
out
just
before
this
meeting
that
Kyle
wasn't
going
to
be
her
I
thought
he
was
running
the
meeting,
so
the
agenda
may
be
a
little
bit
like
today.
We
may
have
to
make
it
up
a
bit
as
we
go
along
so
quick
agenda.
Bashing.
First,
are
there
other
things
that
people
feel
we
should
actually
have
on
the
agenda?
Currently.
B
C
A
A
So
ons
Europe
is
going
on
next
week
and
we
do
have
a
talk
about
Tyler
I
also
expect
there
will
be
happening
that
involves
every
service.
Mesh
and
rumors
are
as
soon
as
Frederick
hits
the
ground
and
finds
a
decent
are
that
there
will
be
something
put
out
about
the
control
network
service.
Mesh
happy
are
so.
If
any
of
you
guys
are
going
to
ons
Europe,
there
should
definitely
be
lots
of
network
service
vegetables
to
engage
with
and
the
place
to
go
for.
That
is.
A
C
So
do
you
expect
to
have
you
know,
I'm
writing
there
on
Tuesday
I
have
a
talk
on
Thursday
on
on
at
the
I/o
and
kubernetes.
Is
there
they
expect
the
guys
to
actually
have
some
sort
of
meeting
point
that
will
be
live
all
the
time
or
to
hook
up
apart
from
in
the
IRC,
but
will
there
be
some
sort
of
continues
and
as
I
am
playing
going
on
or
or
periodic
daily
or
twice
daily
or
something
I
know
what
the
most
discussed
of
three
I.
A
C
A
A
That's
my
expectation.
Okay,
thank
you.
So
there
are
other
events
coming
up
we'll
be
doing
a
lot
at
cube
con
for
network
service,
Manish
I
know
a
lot
of
people
submitted.
Talks
and
I
know
that
one
of
our
major
goals
is
to
be
able
to
demo
network
service
mesh
at
UConn,
and
there
are
several
booths,
both
project
and
corporate.
That
will
be
very
interested
in
having
posting
such
demo
in
their
boots
cool
so
really
quickly.
We
do
actually
have
a
github
project
that
we
will
typically
use
to
sort
of
comb
through
the
issues.
A
So
let's
go
take
a
quick
look
at
that
and
see
where
we
stand
with
issues.
So
on
the
to
dues.
We
have
issues
so
we've
got
the
traditional
X,
Factor
CNF,
headed
hominid
folks
are
familiar
with
a
12o
factor
apps,
but
one
of
the
things
we've
identified
is
as
we
gain
operational
experience.
We
need
something
like
that
for
CN
apps,
that
sort
of
describes.
Okay,
if
you
want
to
make
a
cloud
native
CNF.
This
is
what
it
looks
like
the
migrating
barriers
to
go
to
errors.
Frederic
has
got
that
in
progress.
A
We
have
an
ongoing
conversation
about
becoming
a
kubernetes
working
group.
That's
been
stalling
a
little
bit
as
we've
been
sort
of
rushing
forward
with
development,
but
one
of
the
things
that
we've
got
going
on
is
trying
to
figure
out
what
is
the
proper
formal
home
when
we,
when
we
ask
kubernetes
Signet
working,
they
suggested
we
become
a
kubernetes
working
group,
so
ed.
E
A
Comes
back
to
we're
trying
to
draft
the
proper
set
of
collateral
in
the
course
of
primes
one
of
the
things
that's
come
back,
and
this
is
weird
is
to
be
a
little
interesting.
Is
sig
network
would
like
this
to
be
a
community's
working
group,
kubernetes
likes
us
but
they're,
trying
to
make
working
groups
just
to
be
about
producing
spec,
Docs
and,
of
course,
we're
producing
code.
So
there's
a
little
bit
of
confusion
as
to
how
that
sorts
out.
B
A
A
So
there's
also
in
this
gets
back
to
and
I
think
we're
missing
a
lot
of
the
volca
folks
because
they're
in
rap
to
OH&S
as
well.
Actually
we
have
Watson
here
so
Watson.
Do
you
know
what's
going
on
with
the
supports
Yuna
CN
CF
CNF
project?
Did
you
guys
get
together
the
kinds
of
stuff
that
you
needed
us
to
do
or
I.
G
A
F
Yeah
I
can
said
they're
those
I
think
there's
three
audiences
we're
looking
at
one
is
people
developing
an
NSF
framework
could
and
ap
is
then
people
developing.
You
know
plugins
ie
here
is
a
network
service
for
IPSec
or
disappeared
switch
and
then,
as
people
just
use
this,
we
say
uh-oh,
the
disabuse
switch,
yeah
they're,
good
I
can
just
plug
it
in
and
I
goes
to
somebody
that
some
of
your
presentations
about
the
the
end
user
says
oh
I,
this
and
I
don't
want
to
know
what
what
and
to
did
to
make
this
happen.
F
Just
give
me
enough
information
to
do
this
and
I
think
those
three
ordinances
need
to
figure
out
how
to
speak
them
separately.
I
think
yeah.
We
had
some
discussions
before
about
the
current
code
base.
The
way
this
played
out
it
mixes
it
all
together.
You
know
you
can
piece
it.
You
can
piece
it
together.
If
I
get
on
IRC,
it
ask
you
or
Kyle
or
Fredrik
or
Serge
A,
to
point
B
to
where
to
look.
A
F
A
F
Would
be
nice
to
have
a
different
way
so
I
when
I
enter
the
code
base
and
I
drop
a
PR
say
here?
Yes,
new,
PR,
here's,
you
know
just
be
a
switch.
Everybody
knows
it's
not
affected
any
of
the
existing
code
and
then
I
also
drop
in
to
say.
Oh,
if
you
want
to
use
this
just
be
switch
here.
Is
you
know
the
the
ammo
file,
the
animal
documentation
to
go
that
guy.
A
H
H
A
You
know
and
and
and
also
sort
of
clear
traditions
as
to
then
how
those
get
consumed
by
people
so
that
it
gets
to
be
really
really
easy
for
people
to
contribute
those
components,
because
those
should
be
relatively
modular
components.
And
then
you
know,
even
even
beyond
that.
We
may
also
want
to
offer
places
for
people
to
contribute
en
sm-6,
ternal,
NSN's
or
pms
Emma's
proxy
autosomes
to
do
some
of
the
more
sophisticated
things,
but
I
think
some
of
that's
going
to
have
to
emerge
as
we
sort
out
a
little
bit.
A
F
H
One
of
them-
maybe
one
thing
that
could
help
we
have
scripts
to
deploy
and
SM
in
the
CI
environment,
but
we
have
very
little
how
to
deploy
an
SM
in
the
cutter
in
the
product
or
in
the
lab
environment.
So
maybe
something
like
a
helm
chart
which
would
go
and
deploy
old
in
the
required
pieces
could
help
people
to
consume
it
easier.
A
B
A
B
Well,
I
was
also,
and
I
was
not
pushing
semantic
data
m''d,
so
Oliver
saying
was
like
when,
in
the
end,
when
you
have
the
context
of
the
API,
send
people
also
want
to
know
the
bigger
picture.
It's
onm.
What
is
NSM
API
versus
kubernetes
network
policy,
and
you
know
when
we
get
that
I
also
talk
about
the
service
smash
itself.
There
is
a
little
bit
of
overlap
there.
How
this
is
all
coming
together
right.
Four
people
are
interested
when
you're.
H
A
I
I,
don't
have
a
pull
request
yet
on
I
have
have
been
looking
at
the
API
doc
that
sergej,
prepared
and
merged,
and
looking
at
so
I'm
looking
at
two
things
simultaneously.
What
and
this
is
relevant
to
the
discussion
that
we've
just
been
having
the
last
10
minutes?
Well,
what
additions
to
the
API
doc
do
we
need
to
deploy,
if
not
quite
data
plane,
something
focused
on
the
low-level
delivery
of
the
of
the
actual
resource
that
the
MSM
endpoint
needs
to
in
order
to
construct
this
l2
forwarding
plane?
I
And,
secondly-
and
we
have
some
of
that
definition,
but
there's
a
few
things
that
are
missing
or
could
could
be
augmented
and
one
of
the
things
I
was
thinking
about,
was
the
possibility
of
there's
other
communities
working
and
and
with
CR
DS
and
like
Intel
is
developing.
Apparently,
though
they're
not
all
yet
there,
but
I
saw
them,
and
some
referred
to
in
some
preliminary
slides
for
another
project.
I
You
know
propagating
through
NSF,
so
I've
been
kind
of
thinking
about
that
as
well.
So
this
is.
This
is
really
like
to
do.
This
really
requires
both
the
code
and
some
adjustments
of
the
protocol,
and
you
know
ultimately
I'll
I'll
get
this
done,
and
we
may
end
up
being
something
like
what
sir
J
did
and
that
it
would
be
part
of
the
tree
and
not
really
a
clean
division
of
leveraging,
other
CR,
DS
and
so
on
and
so
forth.
You
can
see
I'm
sorry,
go
ahead.
A
You
know
so,
for
example,
when
I
schedule
an
se
and
I
wanted
to
you
know
Corpening.
If
all
I
really
want
to
do
is
Corpening
and
I'm
not
associated
to
any
particular
hardware
right,
then
that's
already
there
if
I
would
like
to
consume
SR
iove
and
therefore
I
want
to
be
in
a
particular
Numa
zone.
There's
active
debate
about
what
the
right
way
to
do.
That
is,
but
that's
going
to
be
worked
out
in
the
context
of
resource
management
as
part
of
how
the
pod
spec
works.
So
that's
not
something
we
would
manage
ourselves
directly.
H
I
I
think
you're
right.
So,
but
what
and
that's
what
I
was
hoping
I
would
hear,
and
there
seems
to
be
general
agreement
on
that
as
that
we
we
leverage
what's
being
done
elsewhere
in
the
Kas
community,
but
I
think
it's
it's
I
I
just
am
trying
to
to
think
this
through
and
how
much
I
need
to
actually
make
a
concrete,
because
some
part
of
this
effort
is
to
make
a
concrete
example
that
will
actually
forward
packets.
I
A
I
I
think
I
think
so
and
I
think
there
may
be
I'd
be
happy
to
do
that
and
I
think
part
of
what
I'm
trying
to
do
naturally
feeds
into
that.
So
it
could
put
a
in
the
chat
and
I
think
there's
at
least
one
person
here:
who's
also
involved
in
the
moat
Malta's
project
and
I.
Don't
know
whether
they're
taking
the
same
approach
or
not,
but
they
put
in
some
ways
they,
maybe
if
anybody
would
want
it
for
no
comment
on
that.
A
B
Income-
this
is
extremely
well
said,
so
in
fact,
I
was
a
participant
and
resource
management
working
group
like
almost
six
months
but
and
I've
gotten
to
other
things,
but
yeah
well
said,
and
in
fact
we
are
looking
at
the
layer.
3
cache
partitioning
as
sort
of
a
advanced
resource
management
constructs,
but
we've
tabled
it
in
a
future.
That
could
also
be
useful
for
network
functions,
but
yes,
I
think.
B
This
is
essentially
a
higher
degree
of
performance
and
isolation
for
network
function.
So
basically,
if
you
have
sort
of
a
network
function
which
wants
some
guarantees
say
as
an
example
5g,
you
know
packet
costing
function
price.
You
know
UPF
right
and
then
you
have
some
other
network
function
and
a
needed
workload
which
is
more
best-effort
right.
So
the
idea
is
basically,
you
can
create
some
partitions
for
the
cache.
So
basically
you
dole
out
only
a
section
of
the
cash
for
the
best
effort
workload
right
there
as
the
the
the
guarantee
network
function
can
see
entire.
B
C
A
I
A
I
would
love
to
have
it
because,
among
other
things,
there
are
some
things
that
your
so
really
the
things
happening
in
the
resource
to
working
group
conversations,
but
some
of
them,
my
gut,
is
telling
me
don't
go
far
enough.
So,
for
example,
around
the
new
manager
they
would,
if
you're,
if
you
were
a
pawn
bid,
is
gonna
grab
this
snare.
I
A
J
A
F
I
Well,
you
know
some
of
the
same
things
that
we're
gonna
have
to
do
like
there's
these
CR
DS,
which
don't
exist
yet
so
I'm
trying
to
look
at
one
level
below
that,
because
I
think
ultimately,
they'll
probably
be
something
of
a
meaning
of
the
minds
between
these
various
projects
that
are
trying
to
enable
a
high
speed
networking
in
kubernetes,
but
but
so
I'm
trying
to
be
aware
of
what
they're
doing
but
but
not
get
bogged
down
in
and
discussions
of
multiple
address
spaces,
because
we
think
we
have
an
alternative
solution
for
that.
That's
that's!
I
F
A
I
I
Right
sergej:
does
this
go
hand-in-hand
with
the
API
document
that
was
merged
I,
don't
remember
the
pull
request
at
the
moment,
yeah.
H
Exactly
so,
basically,
it
implements
that
dog
with
the
slight
modifications.
Iron
III
I
hit
some
issues
and
they
were
required
very
minor
changes
in
the
dock,
which
is
already
at
it's
already
in
the
updated
revision
of
that
dog
in
the
same
PR.
So
you
can
see
what
was
changed.
It's
very
minor
in
general,
its
it
just
implements
what
we
discussed.
They
agreed
and
NSM
underscore
AP
IMG
file.
F
F
A
We're
actually
really
fun,
but
but
you
you're
eventually
going
to
discover
that
you
need
to
do
things
like
reach
out
and
talk
to
external
components
that
are
outside
of
your
cluster
as
a
network
service,
endpoints
and
so
I
sort
of
talked
a
little
bit
about
this
in
the
generic
right.
So
you
look
me:
you
look
at
the
network
service
mission
components
in
the
abstract
and
then
we
go
down
to
how
we
look
at
them
in
the
cluster
right
and
then
the
abstract.
A
You
just
have
some
network
service
client
that
that's
an
l2
or
l3
connection
to
an
every
service.
Endpoint
right-
and
you
know
you
talk
about
what
is
the
network
service
right?
It's
the
abstract
representation
is
something
that
you
want.
You
know
it
could
include
all
kinds
of
things
like
isolate
a
resource,
access,
protection
for
threats,
guaranteed
bandwidth,
load,
balancing,
proxying
right
and
then
it's
very
focused
on
the
payload
right,
so
you're
actually
focusing
on
the
payload,
not
the
interconnect.
A
When
you
talk
about
the
network
service
and
then
the
network
service
client
does
anything
that
wants
to
connect
to
that.
So
an
example
would
be
a
pod
which
wants
to
connect
to
a
network
service.
You
know
and
they're
talking
about
all
these
things
sort
of
in
the
abstract
and
the
discus
you're
put
about
the
network
service
endpoint
right
in
the
abstract.
A
network
service
endpoint
is
just
something
that
provides
a
network
service
to
which
you
can
get
at
l2
l3
connection
from
whatever
the
client
is,
whether
it's
a
pot
or
something
else.
F
Yes,
I
make
my
stuff
to
make
sense.
I
mean
it's
I'll.
Just
try
to
think
of
implementation
here.
I
will
just
doodling
think
of
visit.
There's
a
lot
of
discussions
about
how
to
manage
network
service
endpoints.
But
if
you
make
them
part
of
the
pod,
the
NS
MD
diamond
set.
Imagine
becomes
easier,
that's
a
more
complexity
but
also
make
sure
I
did
something.
B
A
The
abstract
description,
because
you
know,
because
we
have
been
focusing
on
a
particular
instantiation
of
these
things
very
hard
I-
think
it's
good
to
have
focus
because
we're
trying
to
get
into
a
place
where
we
could
show
things
that
are
working
to
people
and
get
more
people
involved.
You
know
it,
but
and
and
in
different
ways
that
you
may
realize
that
are
going
to
have
different
pluses
and
minuses,
and
you
just
have
to
sort
of
figure
out.
A
Are
there
use
cases
that
make
sense,
given
that
so,
for
example,
you
you
be
one
example
of
running
a
network
service
in
point
in
the
in
the
network
service,
client,
pod
right?
There
are
some
advantages
to
that.
If
you
have
one
service
endpoint
per
pot
right,
because
you
like
a
live
gig
out
there
that
that's
convenient.
A
If
you
have
one,
if
you
have
an
every
service
in
point
that
serves
many
many
network
service
clients
right
many
many
pods,
then
that
of
course,
is
a
terrible
solution
right
and
it
really
depends
on
the
problem
you're
trying
to
solve,
but
I
think
what,
for
the
nice
things
is.
The
framework
is
actually
sufficiently
flexible
that
you
could
explore
these
possible
solutions,
and
some
of
them
will
make
sense,
and
some
of
them
won't,
and
most
of
them
will
make
sense
within
a
particular
problem
domain
you're
trying
to
solve.
F
A
H
You
go
to
the
NSC
go
quote,
which
is
kind
of
a
sample
and
I
see
provided
with
the
repo
I
mean.
That
gives
you
an
idea
of
the
interaction
between
the
NSC
and
NSM,
because
you
will
in
either
way
you
will
have
to
your
NAC
needs
to
talk
to
an
SM.
So
the
rest
of
the
cluster
would
know
that
you
exist
and
what
service
you
provide
and
that's
pretty
much
it.
H
A
This
VR,
the
stuff
in
the
list
of
components,
is
actually
pretty
well
done
so
far,
but
when
you
start
getting
into
the
stuffer
we're
talking
about
like
the
NSM
to
NSM
api
in
this,
dr,
that's
still
somewhat
of
a
work
in
progress.
It's
sort
of
firming
up
a
bit,
but
it's
dealing
with
some
of
these
things.
Like
oh
wait.
We
have
to
negotiate
tunnel
parameters
back
and
forth
between
zones,
yeah.
A
Because
you
know,
speaking
as
a
network,
I
can
back
me
on
this.
When
you
talk
about
SLA
across
the
entire
network,
that's
a
very
hard
problem.
You
know
a
lot
of
people
when
we
talk
about
bandwidth,
reservations
and
trip
identities.
They
just
mean
you're,
guaranteed
a
slice
for
the
NIC
that
goes
out
of
the
node
and
that's
not
so
hard.
But
if
you
want
to
know,
I've
got
a
guaranteed
bandwidth
between
point
a
and
point
B
that
gets
to
eat
right
here.
A
B
It
was
quick
question
here
so
wondering
I
mean
if
somebody's
looking
at
it
slightly
in
a
different
way.
In
a
sense,
hey
I
have
some
have
already
display
decided
in
my
overlay
communication
mechanism.
Hey
you
know,
I,
don't
want
all
this
negotiation,
I
just
decided
to
all
be
exclam.
So
do
we
have
a
way
of
I
mean
and-
and
there
is
already
an
implementation
around
it,
do
you
have
a
way
of
plugging
into
it?.
A
The
network
service
manager
actually
doesn't
do
the
data
playing
bits
right.
It
doesn't
or
more
Network
Service
data
blames
yo
thigh
that
are
available
to
it
and
those
network
service
vegetative
planes
will
have
capabilities
right.
So
if
you
have
a
network
service
mesh
data
plane
that
can
only
do
VX
lab,
then
no
state
say
on
the
left.
Most.
You
know
network
service,
the
left
most
node
here.
Basically,
it
only
does
VX
lab.
A
Well,
then,
when
it
goes
to
trying
to
request
a
connection,
the
only
mechanism
that
it's
great
advertising
by
it
is
going
to
be
the
X
lab,
and
then
the
network
service
way
to
be
able
to
do
the
X
plane
with
it
or
not.
You
can't
ask
people
to
do
things.
We
won't
do
so.
You
can
bring
whatever
in
a
plane
accident
industry,
it's
just
going
to
new.
Whatever.
B
B
B
A
A
From
my
perspective
and
I'm
curious
other
people's
opinion,
so
this
stuff
all
the
way
down
through
the
section
about
sort
of
the
network
service
components
in
kubernetes
that
I
feel
pretty
solid
about
right.
That
stuff
we've
talked
about
a
lot
when
you
get
into
the
network
service,
miss
you
PI's
be
MSM
to
NSF
stuff.
It's
not
it's
quite
a
bit
more
solid,
but
it's
not
quite
fully
solid
yet
so
for
just
as
an
example,
one
of
the
things
that
isn't
here
yet
is
the
negotiation
of
addressing
and
routes
isn't
in
here.
Yet.
J
Go
through
it
as
well
to
make
sure
like
that.
It's
that
I
think
it's
it's
clear
as
well
before
I
shared
with
anyone.
I
just
wanted
like
be
able
to
point
people
just
towards
something
if,
if
like,
if
they
wanna
get
really
into
the
weeds
or
something
like
that,
then
being
able
to
point
them
towards
this
is
I
here,
start
with
this
and
then
come
and
talk
to
me
and
we'll
talk
about
how
to
get
you
more
heavily
involved.
I
may
run
to
a
couple
of
people
like
that.
A
K
A
I
A
C
A
C
Yes,
so
the
most
of
us
who
are
attending
like
me,
you
know
I
schedule
in
advance,
as
the
calendar
is
filling
up
quickly.
So
is
there
any
chance
to
work
out
that
the
time
and
space
for
MSM
interactions
I'm
very
interested
to
talk
about
you
know
anything
everything
that
I
care
about
in
a
center
for
cloud
negative
networking,
the
current
state
of
the
work,
generic
general
discussions
and
testing
benchmarking
comparisons
and
says
so
on?
Would
that
be
possible
next
week?
C
J
J
J
J
So
we
don't
have
a
time
for
them
that
just
yet
and
that's
still,
that's
allowed
to
be
accepted
and
they
work
out.
The
schedule
during
the
conference
itself
is
my
understanding,
but
ideally
we
should
get
enough
people
wanting
to
join
in
that
we
should
get
the
unconference
session
on
Thursday,
so
we'll
also
post
that
on
to
the
on
to
the
website
at
the
events
page
once
we
once.
We
know
that
time
that.