►
From YouTube: Network Service Mesh Meeting - 2019-06-26
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
B
B
Cool,
so
shall
we
get
going
we're
about
ten
minutes
into
the
call
I?
Don't
know
we
mean
that
being
a
short
call
today,
yeah
I
know,
for
example,
like
the
the
one
thing
I'm
aware
of
is
the
desire
to
go
and
turn
the
technology
tree
into
more
of
a
roadmap
and
I
think
that
that
is
gonna
require
a
little
bit
of
work
coming
up
to
the
next
meeting
before
it
actually
is
useful
to
go
over
it
I.
B
So
yeah
definitely
I
mean
that's.
Gonna
require
a
little
bit
of
work,
because
I
basically
just
have
the
same
tree.
That
I
had
last
week,
which
is
good
at
all,
but
the
other
one
is
that
it's
sort
of
rapidly
becoming
the
world
is
moving
on.
So
things
like
I
got
word
yesterday
from
the
from
one
of
the
folks,
is
working
on
it
or
domain
that
they
they
sort
of.
Basically
have
it
working
now
and
now:
they're
cleaning
things
up
and
testing
across
different
plows
and
so
forth
before
they
go
ahead
and
push
their.
B
Do
we
do
it's?
It's
really
well
done
as
far
as
it
goes,
but
there
are
still
some
questions
that
is
asking
that
I
think
need
to
be
resolved
before
you
get
there.
If
I'm
reading
it
correctly,
the
current
implementation
will
do
the
fan-out
quite
correctly,
but
all
the
right
pieces
are
in
place,
which
means
that
once
we
get
the
fan-out
functionality,
there
I
think
it's
in
the
right
place.
B
A
A
A
B
A
Yeah,
my
my
understanding
is
like
we've
wear
it
we're
no
longer
a
listing.
The
clustering
has
a
main
item
on
the
agenda,
so
I
think
it's
pretty
much
done,
except
for
except
for
minor
improvements
as
as
I
had
mentioned.
So
if
you,
if
you
want
to
use
that
or
you
want
to
share
it
around,
but
that
should
be,
that
should
not
be
a
problem.
A
A
B
C
C
Looking
at
these
this
terminology
and
then
I
go
cross-reference
to
a
networking
book
and
I'm,
saying:
okay,
I'm
not
able
to
cross-reference
here,
I,
see
some
words,
but
where
is
this
this
model
and
then,
when
you're
in
it
you're
in
a
dialog
of
six
or
seven
people,
group
and
they're,
talking
and
they're,
the
networking
people
are
talking
about
layer,
1,
layer,
2,
layer,
3
and
all
these
things,
and
we
were
like
wait
a
minute.
You
know.
Is
it
that
NSM
kind
of
goes
around
these
things
or
or
what?
C
And
it
seems
to
me
that
sometimes
I
was
thinking.
Maybe
sometimes
some
things
get
lost
when
we
start
talking
about
software
data
planes
and
the
difference
between
that
and
an
ASIC
and
meh
my
kernel
interfaces.
These
are
all
things
that
maybe
I'm
wrong,
but
it
seems
these
things
are
kind
of
looks
like
with
aid,
which
you
would
always
say
they
are
virtual,
represent
it's
kind
of
sometimes
when
we
think
about
it.
Sometimes
it's
a
mistake.
A
B
C
Think
we're
well,
my
point
is,
is
that
there
are
certain
problems
that
are
addressed
at
every
layer
and
that's
where
I,
when
I'm
in
different
discussions,
that's
where
things
get
lost
with
devs
devs,
don't
believe
any.
There
are
any
problems
like
so
like
the
spanning
tree
protocol
like
a
layer,
2,
Ethernet
type
problems.
We
don't
know
anything
about
that,
and
we
don't
care
about.
It
must
be,
it
must
be
done,
must
be
fixed.
You
don't
care
about
that.
We.
B
I'm
well
like
this
is
a
matter
of
personal
opinion.
For
me,
I
I
think
it
is
ludicrous
how
much
we
just
still
care
about
layer
two
and
this
mutt
day
and
age
I
you
you
you
can
make.
You
can
make
a
really
strong
case
that
the
two
central
sins
of
networking,
the
two
things
that
have
brought
the
most
mystery
to
the
world-
well
I,
actually
put
three
in.
There
would
be
number
one.
B
The
ridiculous
ways
that
we
weld
IP
to
Ethernet
right
and
that
we
basic
kept
l2
along
I
mean
it
comes
down
to
shared
media,
was
the
original
sin
and
that
led
to
all
kinds
of
stupidity
and
l2,
and
therefore
all
kinds
of
stupidity,
the
interface
between
l2
and
l3.
So
sure
media
is
the
first
in
the
second
sin
was
having
IP
addresses
identified,
both
the
location
used
as
both
identifiers
and
locators,
which
means
now
your
identity
that
the
IP
layer
is
tied
to
location
in
a
way.
B
That's
really
unhelpful
and
leads
to
all
kinds
of
crazy
and
then
the
the
third
sin
is
tying
TCP.
Connections
to
IP
addresses
the
way
we
currently
do.
Id
I
P
addresses
imports
in
such
a
way
that
they
sort
of
mushed
together
in
a
fortunate
way
that
makes
the
transport
layer
a
little
bit
screwed
up
and
the
good
news
is
the
you
know.
Kubernetes
is
pure
l3,
which
means
it's
done
away
with
the
shared
media
myth
you
know,
locator
it
at
a
fire.
B
C
A
C
B
C
B
The
way
I
usually
describe
to
trying
to
explain
this
to
network
people
right
because
network
people
are
like.
Oh,
you
know
they
have
to
do
blah
blah
blah
because
of
performance
and
scale,
and
my
response
is
to
turn
there
are
people
would
say:
look
that
the
cloud
native
people
will
decide
what
they
want.
They
will
find
a
tool
that
solves
the
problem
in
the
most
straightforward
way
and
they
will
call
it
good
up
until
the
point
they
hit
the
wall
on
scale
and
performance.
B
That's
a
losing
argument,
generally
speaking
and
part
of
why
it's
losing
argument
is
that,
quite
honestly,
if
you're
sitting
on
the
other
table
with
the
cloud
native
people
at
least
half
the
time
and
probably
closer
to
80%
of
the
time
when
someone
tries
to
Buffalo
you
with
that
argument,
it's
you
don't
care
right.
So,
where
I
see
the
people
actually
starting
to
care,
is
you
know
when
they
hit
the
wall
on
scale
and
performance
right
so
table
stuff?
B
You
know
you
came
in
literally
18
months
ago
and
there
were
lots
of
people
already
hitting
the
wall
on
that,
where
it
simply
did
not
scale,
and
so
you
know
they
came
back.
It's
like
well
we'll
do
this
I
PDS
well,
which
was
a
little
bit
better
but
still
doesn't
really
scale
or
perform
when
you
actually
get
a
scale
off.
So
you've
got
a
bunch
of
people.
B
Who've
hit
the
wall
there,
and
now
people
are
saying
what
we'll
do
with
EVPs
and
it'll
be
interesting
to
see
to
what
degree
that
actually
scales
and
performs
it'll
clearly
be
better
than
what
we're
doing
right
now
with
ipbs.
But
my
guess
is
that
once
you
get
a
critical
mass
of
people
hitting
the
wall.
B
A
Right,
yeah
I
think
there's
there's
a
few
things
going
on
as
well.
So
first
I
think
people
are
more
comfortable
talking
to
you
than
they
are
talking
to
us
like
with
us.
They'll
say
they
obviously
know
what
they're
doing.
There
must
be
something
right
there,
I,
don't
I,
don't
really
see
it,
but
I
don't
want
to,
but
but
I
don't
want
to
look
stupid
by
asking
a
question,
and
so
that's
where
our
messaging
becomes
very
important
to
to
help
with,
with
with
them
to
understand,
like
not
only
like.
A
So
if,
if
you
talk
to
someone
like
I
use
a
Ian
as
an
example,
so
you
talk
to
someone
like
Ian
who's,
very
invest
in
the
telecommunications
industry,
and
you
know
he's
looking
for
like
that
that
srl
B
support
and
how
do
you
maximize
that
overall
performance,
it's
like
you
talk
to
and
you
talk
to
the
people
who
are
looking
for
his
type
of
services.
There
they're
gonna,
be
all
about.
Let's
maximize
speed:
let's
get
throughput,
you
know,
while
controlling
complexity
developers,
they
couldn't
care
less
or
they
so
what
they
end
up
doing.
A
Is
they
end
up
that?
That's
part
of
the
reason
why
we
we
focus
on
be
on
the
Sarah.
A
narrative
is
that
you
know
and
when
I
talk
with
the
developer
and
say
well,
why
do
I
need
this
and
says?
Well,
usually
you
don't?
But
what,
if
you,
but
let's
say
that
you
had
a
workflow
where
you
were
connecting
a
pod
to
a
financial
BPM,
and
you
know
want
to
expose
your
entire
cluster
to
it.
A
You
want
to
make
sure
that
that
specific
workload
was
the
only
thing
that
was
able
to
connect
to
it
because
of
your
business
security
policies.
How
would
you
do
it
and
then,
usually
by
then
they
they
say:
oh
okay,
I
see
I
can
see
where
I
would
use
MSM
for
that,
and
so
so
it
really.
It
really
depends
on
on
the
on
the
user
where,
like
they,
they
know
that
there's
some
interesting
amazing
magic
that
goes
on,
but
they
really
don't
want
to
care
about
it.
A
It's
literally,
it
is
the
embodiment
of
the
situation
and
Hitchhiker's
Guide
to
the
galaxy
and
I.
Don't
think
they're
wrong
about
about
that,
because,
even
if
they
knew
about
the
problems
that
occurred
in
the
telco
space,
what
could
they
do
about
it?
And
so
it's
it's
good
for
them
to
be
able
to
focus
on
their
on
their
specific
problems
that
they
have
in
need
and
make
sure
that
they
build
something
that
can
make
use
of
new
new
improvements
going
on
on
the
infrastructure,
yeah.
C
They
just
aren't
they
what
we
were
saying
this
fourth
profile,
we're
talking
about
profiles,
maybe
there's
a
fourth
profile:
it's
like
an
infrastructure
developer
and
a
data
center
type
person
he's
setting
up
the
networks
and
stuff
in
a
larger
data
center.
Let's
say
a
government
type
data
center
or
something
where
there's
going
to
be.
You
have
to
care
about
performance
at
some,
so
at
some
level
me
not
a
telco
but
you're
you're
hitting
that
you
have
to
be
mindful.
You
just
can't
set
up
the
infrastructure.
B
C
B
You
want
high
performance,
then
you
have
to
do
things
right,
so
they
could
do
just
willy-nilly
anything
and
looking
for
a
service
carefully
designed
to
keep
working.
But
you
know
if
they
want
all
kinds
of
cool
tricks,
neither
physical
network
could
do
for
them.
They
may
have
to
do
something
to
expose
that
via
Network
Service
mash.
A
So
when
I
was
describing
every
service
to
him,
one
of
the
things
that
I
that
I
realized
was
that
the
data
center
has
a
couple
of
problems
that
they're
running
into
so
the
first
one
is
the
they
have
no
insight
into
the
workloads
that
are
running
on
top
of
them,
which
means
that
they
cannot
adjust
for
the
type
of
workloads
that
they
that
they
need.
So,
if
you
just
need
standard
networking,
okay,
that's
fine!
A
So
when
we
were
talking
about
never
service,
my
shows,
like
you
know,
well
what
additional
services
or
things
that
they
provide
access
to
and
potentially
bundle
up
as
a
service.
So
if
you
wanted
something
that
that,
provided
you
know,
even
something
is
like
it
could
be
something
at
a
higher
level
if
we
make
a
VPN
or
might
be
something
of
a
lower
level.
A
That
does
something
special
in
their
infrastructure
to
use
to
do
some
traffic
management
or
to
or
to
guarantee
some
type
of
quality,
then
being
able
to
to
request
for
those
things
with
in
a
declarative
way
with
labels
and
so
on.
It's
something
that
they
could
build
towards.
That
would
give
them
a
unified
interface
for
requesting
those
type
of
those
type
of
things,
and
so
so
he
was
quite
excited
with
the
concept
of
network
service
mesh
from
from
the
purposes
of
being
able
to
expose
out
and
refine
the
abilities
of
the
dentist
data
centers
fabric.
C
A
But
that's
a
very
specific
type
of
infrastructure
person.
I.
Think
your
average
infrastructure
person
that
we're
probably
gonna
run
into
in
the
beginning
is
going
to
be
enterprise
infrastructure.
So
not
the
data
center
person,
but
like
the
person
who
runs
the
the
clusters
and
to
be
honest,
I
think
the
biggest
use
case
we're
gonna
have
for
them.
In
the
beginning,
it's
probably
going
to
be
Sara's
Sara's
use
case.
You
know
literally
hooking
up
VPNs
from
one
organization
to
another
and
I
yeah.
A
It
turns
out
that's
a
crazy,
difficult
problem
even
today,
and
it's
not
uncommon
for
that
scenario-
to
take
several
months,
sometimes
up
to
six
months
to
to
establish
connectivity
Wow.
So
if
you
can
go
and
say
hey,
we
can
do
this.
You
know
once
you've
signed
the
paperwork,
so
you
can
doesn't
fix
the
legal
side
entirely,
but
once
once
they've
signed
the
paperwork
and
have
agreed
to
share
to
share
a
connection,
you
know
the
my
hope
is
that
they
can.
They
can
set
it
up
in
five
minutes.
That's
my
that's
my
hope.
A
You
know
you're
sure
you
share
the
key,
that's
their
the
secrets
and
then
you
used
to
use
load
of
network
service
message
and
say
you
make
me
do
this
thing
with
these
secrets
and
then
you
know
there.
You
have
a
yellow
pod
loaded
that
that's
or
the
VM
loaded
in
your
in
your
over
structure.
Just
it
just
does
its
thing.
B
It
turns
out
that
there's
only
really
one
feature
that
people
are
looking
for
in
layer.
One
and
that's
things
you
shopping
wouldn't
come
out
the
other.
Then
you
go
to
layer
2.
The
number
exists
that
people
want
and
what
features
they
want,
and
the
fact
that
the
features
conflict
with
each
other
gets
to
be
really
gnarly.
D
A
And
the
way
that
I
try
to
think
about
it
is
so
also
have
you
heard
of
the
term
link-local
I
am
so
so
what
I
try
to
think
of
is
the
NS
MV
wires
as
we're
calling
them
is
trying
to
keep
link
local,
actually
local,
between
just
two
connections.
So
that
way
and
and
above
that
use
you
can
have
crowding
and
so
on,
but
it
literally
tries
to
constrain
as
much
as
possible
the
link
local
to
a
to
a
very
small,
very
small
domain.
C
Well,
I
have
a
bunch
of
things
that
I
was
going
to
try
to
push
into
the
dock,
that
Jeffrey
is
doing
and
I
guess
there's
the
other
dock.
That
is
the
more
I
guess
on
the
vendor
side.
So
I
guess
there's
some
things
there.
We
should
probably
I,
don't
know
in
the
end,
merge
everything
but
I'm
trying
to
trying
to,
as
far
as
on
the
CI
CD
side
and
trying
to
pull
the
best
practices
on
over
and
try
to
look
at
some
of
the
things
that
MSM
bring
to
the
table.
C
D
C
Outside
of
right
so,
and
then
say,
this
is
one
of
the
drivers
for
using
the
NSM
in
the
testbed.
It's
because,
okay,
now
we're
installing
things
using
helm
and
getting
away
from
the
imperative
file
scripts
for
installing
and
trying
to
get
closer
to
declarative
and
these
types
of
things
and
trying
to
say:
okay.
This
is
property.
Icd
nicer,.
C
C
Why
something
is
even
cloud
native?
You
know
having
things
like
what
you
would
say,
added
loosely
coupled
and
other
things
why
you
can't
have
everything
all
call
together,
most
likely
all
in
one
big,
BM,
three
different
concerns
or,
however
many
and
then
call
it
cloud
native
I.
Think
that
that's
maybe
some
of
Jeffrey's
concern
and
I
know
that.
B
B
C
B
That's
right
silly
right
there
I'd
say:
look,
you
know
you
guys
went
in
totally
eyes
open.
You
wanted
to
be
able
to
get
performance
numbers.
You
make
very,
very
well
reasoned
choices
around
the
out
of
bounced
off,
but
that
can't
be
the
way
the
world
works
in
the
in
the
limit
of
actual
appointments.
Yeah.
C
C
B
C
A
B
C
Yeah,
so
my
goal
is
to
not
something
it's
really
to
say.
This
is
what
it
means,
and
here
the
argument,
if
you're
not
doing
these
things,
then
you're,
not
cloud
native
but
I'm
not
going
to
on
cloud
maybe
and
for
CIT
D.
These
are
the
goals
for
it
to
be
cloud
native,
so
it
can
be
orchestrated
on
things.
Otherwise,
I
don't
get
the
benefit.
A
Actually,
instead
of
bronze
CNF
like
perhaps
the
browser
needs
to
be
a
term
towards
has
I
mean
the
entire
concept
on
that
stuff
quotes
to
provide
a
progression
so
as
a
first
is
to
provide
a
progression.
So
if
you're
trying
to
become
a
best
practices
on
native
network
function,
here's
a
here's,
a
reasonable
path
that
things
you
can
do
to
move
up
the
to
move
up
the
chain.
The
second
thing
is:
is
risk
mitigation
for
the
operators,
because
your
risk
profiles
will
look
very
different
than
you
take
up
a
bronze
CNF.
A
Our
main
goal,
see
enough
and
so
I
think
I.
Think
that's
part
of
what's
been
end
up
happening
is
my
hope
is
that
what
happens?
Is
that
the
operators?
They
they
don't
look
at
it
as
a
rubber
stamping
of
a
CNF
where
you
just
had
to
lift
and
shift
things
that
require
privilege
and
stick
it
in
a
container
and
instead
look
at
it
as
an
a
well.
A
B
A
C
Yeah,
my
the
rule
of
thumb,
I'm
saying,
is:
if
I
hear
something
you
know
50
times,
they'll
make
a
video
explaining
things
have
been
changing
a
lot
lately,
but
we
keep
saying
things
over
and
over
then
you
know
we
can
deal
five
three
five
minute.
Video
on
I
was
hoping
this
glossary
or
something
along
these
lines.
It's
you
know
we
keep
using
this
terminology,
then
that
might
be
it
so
see.