►
From YouTube: IETF95-I2RS-20160405-1400
Description
I2RS meeting session at IETF95
2016/04/05 1400
A
If
we
go
to
the
tools,
page
4
hrs
and
go
to
the
minutes,
link
you'll
see
a
link
for
etherpad
95.
This
is
the
one
that
is
highlighted
on
the
screen.
If
we
can
get
a
couple
people
to
go
here
to
help
us
collaboratively
take
minutes
it's
easier
more
than
one
of
you
helps,
and
it's
very
easy
to
know.
Just
add
things
here
and
there
the.
B
B
So
we
have
a
an
agenda
that
short
but
I
have
a
feeling
we're
going
to
have
a
discussion.
That's
long.
So
the
first
thing
we'll
do
is
we'll
walk
through
this
agenda
and
if
you
want
to
add
something,
if
you
have
some
draft,
you
want
to
tell
me
about
it.
The
last
minute
please
do
we'll
have
the
chair
slides,
which
will
probably
catch
us
up
on
the
10
after
and
then
we'll
have
a
hackathon
a
update.
B
B
Okay,
first
of
all,
Jeff
and
I
met,
as
we
came
into
the
ITF
week
to
talk
about
goals
because
I
2
s's,
a
creation
of
a
protocol
interface
and
we've
been
really
go.
Oriented
in
2015
and
I
had
thought
we'd
be
through
a
bit
more
of
the
ISSG
process
by
now,
but
we're
resetting
our
goals.
So
our
version
one
is
to
send
to
the
isg
all
our
requirements
by
May
5th.
B
The
one
requirements
remaining
are
the
ephemeral
state
which
requires
us
to
do
some
work
on
the
protocol,
straw,
man
and
this
management
data
flow
and
then
hopefully,
we'll
work
with
the
net,
come
to
at
least
get
an
initial
protocol
idea
by
July
and
then
work
on
a
hackathon
and
the
hackathons
have
been
very
useful,
so
we'll
be
going
through
that
and
by
July
I.
Think
we'll
have
enough
experience
with
our
protocol
independent
data
models
to
send
most
of
them.
B
The
topology
data
model
has
been
out
and
being
used
for
a
year
and
a
half
is
that
about
right
about
a
year
and
a
half.
The
rib
data
model
will
hope
to
have
some
code
and
open
sourcing
in
between
the
ITF
s
and
the
filter
based
rib.
We
hope
to
have
the
same
thing
in
open
source
and
then
we'd
like
to
ask
for
design
team
rusts.
Let
off
a
really
interesting
discussion
after
the
November
ITF
on
multiple
priorities
and
priority
heads
I
want
to
actually
try
to
pick
up
a
design
team
in
this
next
section.
B
Okay
approved
is
RFC's.
We
finally
got
our
first
to
RFC
CA
and
we
have
three
requirements
at
the
isg.
The
security
requirements
is
a
maybe
going
within
a
day
or
so
so
I
included
that
we
have
the
two
requirements
I
mentioned,
and
we
just
need
to
finalize
things.
Oh
by
the
way,
if
you
haven't
commented
on
the
protocol
security
environments
requirements,
if
you
would
take
a
moment
to
do
that,
what
we
found
in
our
first
review
of
the
architecture
is
they
really
wanted
us
to
describe
for
this
new
environment?
B
The
requirements
and
Miguel
wrote
a
very
nice
document.
If
you
would
just
try
to
look
at
it,
if
you're
into
security,
if
not
we'll,
go
find
some
more
people
from
the
security
environment,
the
security
environment
is
not
a
protocol
but
the
the
environment.
The
protocol
lives
in.
So,
if
you're
running
I
2's
on
a
router
or
on
a
device,
this
tells
you
what
security
you
should
be
concerned
with
on
that
device.
B
I
think
I've
gone
through
the
rest
of
them.
After
that,
we
may
give
some
start
to
people
doing
our
truest
models
for
other
protocols.
Okay-
and
I
do
want
to
list
the
interims-
you
know
we
have
a
practice
of
interims
to
have
that
deep
discussion
will
have
an
interim
on
the
427
518.
You
can
read
it
here.
The
idea
is
to
try
to
get
through
the
last
pieces
of
the
protocols
strawman
and
get
to
the
hackathon
with
a
good,
solid
implementations.
B
B
D
D
Yeah,
so
at
the
at
the
hackathon,
we
had
a
desire
to
add
more
code
to
the
opendaylight
implementation
of
yang
push
and
push
currently
in
open
daylight
is
able
to
create
a
subscription
from
open
daylight
on
to
a
network
element.
In
this
case,
an
XR
box,
most
of
the
code,
was
written
by
ambika
Tripathi
who's.
Coming
with
me
right
now,
and
I
got
some
help
from
ye
from
lead
from
from
Nokia
and
doing
some
the
net
cough
notification
stuff
next
slide.
D
D
If
you
look
at
the
set
up,
you
can
see
a
terminal
where
we
have
rest
api
is
going
from
yang
push
in
the
opendaylight
down
through
a
subscriber,
a
yang
push
subscriber
that
goes
to
the
service,
abstraction
layer
and
down
to
an
XR
box
on
the
bottom.
You
see
three
sets
of
information
and
yang
models.
We
were
able
to
create
and
delete
subscriptions
for
interface
state,
in
other
words,
we're
allowed
to
create
interface
going
up
and
down
we're
able
to
create
and
delete
an
interface
sre,
a
a
CL.
D
So
you
could
do
config
change
and
watch
the
ACLS
be
changed
on
the
box
and,
finally,
we
did
routes
were
able
to
outer
add
a
route
to
the
box
and
watch
leave.
Ralph
information
be
pushed,
so
he
did
add
and
delete
subscriptions
for
three
types,
and
we
also
were
able
to
do
and
show
it
via
net
comp
directly
without
having
to
go
through
up
in
daylight
as
a
as
a
as
an
intermediary.
So
both
were
both
were
shown
and
let's
talk
a
little
bit
more
about
that
next
slide.
D
So,
as
I
mentioned
that
we
had
a
goal
of
making
the
the
delete
subscription
work,
opendaylight
had
only
periodic
subscriptions
available.
However,
in
the
demo
and
in
the
hackathon
we
were
able
to
do
not
just
a
periodic
subscription.
We
also
got
working
an
unchanged
subscription
where
we
get
information
sent
from
the
XR
a
box
as
something
is
changing
in
the
network
element
itself.
B
With
we
had
Eric
doing
one,
and
then
we
planned
to
work
on
an
environment
that
Edwin
Go
Daddy
go,
did
I
pronounce
that
right,
if
you're,
more
Spanish
and
enabled
than
I
am
and
he
sent
a
tutorial
and
he
sent
a
vm
and
what
we
discovered
is.
The
vm
was
a
little
bit
heavy
for
our
laptops
and
then
we
hope
to
load
IRS
data
models
into
the
OTL
library
and
connect.
B
Well
after
we
found
that
the
it
was
a
little
heavy
to
get
this
edwin's
development
environment
by
the
way
I
should
take
a
step
back.
The
reason
Edwin
send
us
a
development
environment
is
one
of
the
problems
we
had
in
hackathon
80
ITF
94
is.
It
was
very
difficult
for
us
to
set
up
the
development
environment.
B
What
I
found
is
that
between
lithium
and
beryllium
did
I
pronounce
that
right,
odl
had
really
improved
it
set
up.
So
I
took
me,
maybe
half
an
hour
to
download
the
ubuntu
and
download
the
OTL.
Suddenly
I
had
the
environment
we
couldn't
set
up
with
with
him.
So
that's
why
Edwin
had
taken
his
work.
He
had
set
up
that
and
min
net
and
min
exit
so
but
it
kept
having
too
much
memory
requirements
our
laptop.
B
So
we
took
the
second
goal,
which
was
to
hack
together
an
I
er
to
his
client
and
server
and
get
an
hour
to
his
route.
Add
working.
So
these
are
the
people
who
helped
us,
Sudan,
Jason,
Lucy,
Mamadou
and
we
broke
into
teams
and
stand
up.
Mama
do
momma.
Dee
was
our
first.
It
was
the
first
time
a
participant
in
the
ITF,
and
he
joined
this
bravely
at
hackathon
within
his
first
couple
hours
and
he
actually
helped
us
get
the
environment
up.
B
B
Who
did
that
and
got
it
up
and
going
once
they
worked
out
a
few
bugs
in
virtualbox
and
then
I
in
Anna
and
Jason,
looked
at
loading
gang
CLI,
pro
quagga
and
tail
FD
together,
and
we
got
the
and
Yuma
where's
our
Yuma.
We
got
the
Yuma
pro
loaded
up
the
gang
model,
and
then
we
were
trying
to
connect.
The
two
are
then
next
step
was
to
get
quokka
to
talk
to
Taylor
and
after
we
work
to
this
a
while.
B
We
found
that
this
was
the
commercial
variant
that
tail
F
actually
sells
is
a
Quadra
installation,
so
that
was
probably
pretty
good.
We
did
get
at
least
the
two
coffee
and
yang
pro
up.
It
was
fairly
easy
to
use
both
you
just
unload
them.
Their
command
lines
are
easy
to
work.
Thank
you.
Andy
for
ax
and
humor
works
for
having
a
lovely
yang
pros
gang
CLI
pro
for
us
and
comfy
was
really
easy
to
unload
and
begin
to
connect.
Zebra
has
an
tral
RPC.
B
B
B
Okay,
I
have
the
next
presentation
because
in
as
the
chair
tried
to
bundle
up
all
the
requirements
that
everyone
else
is
done
and
send
them
off
to
the
isg
I
found
a
hole
so
I
need
your
help
to
sort
of
finish
this
off,
and
so
I
can
ship
them
okay,
or
at
least
when
we
didn't
know
about
now.
One
of
the
things
we
talked
about
in
the
architecture
is
that
we
would
not
only
have
a
configuration
and
invent
channel
that
we
would
have
some
other
data
flow
requirements.
B
Okay,
so
I
took
as
the
chair,
and
this
is
sort
of
a
Chairs
draft
trying
to
clean
up
and
put
a
bow
tie
around
all
the
requirements.
I
looked
at
the
use
use
case
summary
and
Joel
carefully
pointed
to
me
that
that's
a
sort
of
old
draft-
and
this
is
a
plea-
if
you
could
read
that
and
tell
me
if
it's
still
reasonable
but
I
looked
at
this
draft
and
I
looked
at
all
the
data
flow
requirements
and
after
having
done
everything
else,
I
sort
of
compiled
a
list
of
requirements,
it
would
have
said
now.
B
I
am
not
saying
these
are
requirements.
We
need
to
do,
but
it
is
something
we
need
to
talk
about
and
can
want
to
include
them
in
the
requirements.
There
are
a
couple
taking
my
chair
head
off.
That
I
would
like
to
see
us
consider
like
what
happens
if
we're
doing
a
management
channel
in
the
middle
of
a
denial
of
service
attack.
Do
we
want
to
try
to
do
something
special
I
would
like
to
try
to.
In
my
personal
hat,
see
what
we
can
do
about
validation?
B
Do
we
do
always
do
netcom
style
validation,
or
do
we
do
maybe
less
to
make
faster
or
maybe
not
at
all,
and
there
are
implementations
which
are
doing
all
free.
So
let
me
start
with
this
and
at
the
end
of
each
one
I'm
going
to
go.
Does
anybody
think
we
really
have
to
have
this
in
will
make
the
same
call
on
the
list,
but
this
is
a
fairly
short
turn.
B
Okay,
the
first
one
is
exactly
on
this
of
checks,
and
it's
one
that
having
talked
to
three
different
implementation,
people
are
actually
more
like
five
or
six.
There
are
several
ways
to
do
it.
The
one
is
minimal
data
reception
checks
are
the
tlv
valid
on
the
wire,
and
you
assume
the
client
has
done
something
useful.
Now
what
can
happen?
You
can
shoot
yourself
in
the
foot
by
having
bad
data
and
as
ND
as
Andy
will
tell
me
if
you
get
bad
data
in
the
data
store.
B
Life
is
just
really
bad,
it
stays
bad
likely,
but
there
are
some
implementations
which
do
enough
checking
and
they
want
enough
speak
on
routes
that
they
would
like
this.
The
next
one
is
non
referential
cheats,
don't
do
leaf
referee
once
don't
do
instance,
identifier
and
that's
medium
ground
was
sort
of
suggested
as
a
way
to
speed
things
up
and
the
next
one
is
full
checks.
Now.
E
Andy
Berman:
this
is
very
server
centric
approach.
What
does
this
give?
The
client
developer
is
trying
to
write
code.
It
works
on
multiple
boxes.
It
gives
them
no
confidence
that
the
gang
is
actually
implemented,
so
it
really
diminishes
the
value
of
yang,
as
as
describing
the
constraints.
Also,
we
should
note
that
these
machine-readable
constraints
are
optional.
You
could
put
all
your
constraints
in
the
description
clause,
so
you're
going
to
say
well
we're
not
ignoring
the
description
clause.
Two
I
don't
need
to
read.
I.
Don't
need
to
enforce
that.
E
You
know,
based
on
arbitrary
things
like
leaf
ref
statement,
because
I
could
put
the
requirement
for
leaf
ref
in
the
description
claws
instead.
So
this
is
is
really
a
simplistic
view
of
how
things-
and
you
know-
and
it
also
goes
to
well
I
trust.
The
client
and
no
client
is
ever
going
to
try
to
attack
me
in
an
exploit
information,
or
you
know,
set
up
a
path
to
to
send
a
you
know:
privacy
leak
privacy
data,
so
the
in
the
ITF
it
it's
going
to
raise
security,
guys
some
interest.
B
Am
well
aware
we'll
be
raising
the
security
guys
interests
so
I
have
had
people
who
are
doing
implementations.
Do
all
three
I'm
glad
and
he's
here
to
represent
the
full
validation.
But
this
is
one
in
your
chairs
hat
off
personal
opinion.
We
really
have
people
doing
it
three
ways
and
I
don't
want
to
break
it,
but
the
past
of
least
resistance
is
the
full
checks,
and
we
can
do
that
and
say
it's
too
busy.
Anyone
else
want
to
make
a
comment
on
this:
hey
you're
cheers
thank
you.
B
B
There
is
a
requirement
right
now
that
come
and
please
correct
me
if
I'm
wrong
can
do
both
xml
and
Jason
as
far
as
a
form.
Oh,
but
but
there's
a
press
conf
into
XML
and
Jason.
Thank
you
very
much.
I'll
correct
the
slides,
I.
Don't
know
that
anybody
really
needs
anything
more
than
this
at
this
time,
but
that
was
one
of
the
requirements
is
to
be
able
to
do
data
fat
formats
or
beyond
that
and
a
ski
excuse
me
patrol.
B
G
B
B
D
Transport
thread
yeah,
Eric
voice.
This
is
more
of
a
question
that
I
asked
at
net
mod,
and
that
is
how
would
we
integrate
with
the
protobuf
and
other
things
that
are
not
necessarily
ietf
standards
specs,
so
we
have
a
lot
of
desire
to
it,
but
I
don't
know
exactly
how
to
do
that
binding.
Is
there
any
thought
people
have
on
that.
H
So
it's
just
a
matter
of
you
know
how
you
were
serialize
your
data.
So
if
you're
using
the
google
proto
buffs,
you
will
get
also
the
RPCs
and
in
this
case
so
sorry
to
correct
myself.
So
you
know
with
the
proto
buffs
you
can
just
utilize
the
data
and
then
you
can.
You
know,
map
them
into,
for
example,
if
it's
allometric
they
they
can
map
them
into
your
RPC
that
you
have
in
yang.
So
you
know:
go
ahead.
I'm.
F
Sorry,
yes,
you
can
make
proto
BOTS
work,
that's
not
a
question.
The
question
is:
what
are
we
going
to
standardize
and
for
now
what
we
have
agreed
to
standardize,
this
XML
and
JSON
and
let's
not
go
into
what
are
all
the
other
things
and
how
would
we
make
them
work?
Because
that's
not
the
question
on
the
table
for
us
well,.
A
I
do
have
to
support
Joel.
No,
we
have
to
start
with
the
mandatory
diplomats,
basically
just
because
we
need
to
make
progress.
That
said,
no
you
give
an
outside
of
I
hrs.
There
is
going
be
demand
for
this
type
of
telemetry.
The
right
places
for
that.
No
standardization
work
for
translating
yang
models
into
other
formats
and
other
transports
is
work
that
probably
belongs
net
Matt
and.
H
B
D
E
Under
idea,
actually
there
I
did
a
draft
a
while
back
and
had
a
proposal
for
negotiating
the
encoding.
There's
a
natural
point
in
the
net
conf
in
between
the
hollow
and
the
first
RPC,
where
they
have
to
change
the
framing
anyways
right
there.
You
can
negotiate
anything
you
want.
As
far
as
the
you
know.
A
B
D
A
H
A
B
I
think
we
took
care
of
two
and
three
all
at
once,
so
I'll
just
I've
got
it.
Sorry,
oh
no!
That's
fine!
Well,
there
must
be
a
four
okay.
This
is
where
we
get
to
the
IP
fix.
Should
we
in
the
first
version
and
I
will
call
it
in
the
first
version?
Allow
IP
fixes
one
of
them?
Is
anybody
want
IP
fix
in
the
first
version
Joel,
you
know
my
personal
opinion
on
this
is
no
so
and
your
opinion
is
know
if
you
wanted
to
sit
down.
I
B
B
Okay,
so
till
I
get
more
input
from
them,
no,
but
know
that
it
could
come
so
I
guess
I
gotta
frame
it
that
way,
I'm
good
to
hear
that
they're
trying
it.
That
was
in
the
specification
and
that's
the
two
of
four
and
five
is:
can
you
send
it,
and
can
you
choose
it
okay,
should
we
have,
should
we
be
able
to
support
a
data
traffic
that
scales
up
and
scales
down?
B
In
some
sense,
the
push
applications
has
the
ability
to
scale
subscriptions
up
and
down,
but
should
we
be
able
to
put
in
the
protocol
something
that
says?
Oh
yes,
if
you're
doing
a
memory
processing
time
or
upon
flow,
we
have
this
rate
now.
This
could
even
be
a
gang
model
for
the
the
IRS,
but
that's
one
of
the
things
anybody
feel
like
it
should
be
in
the
first
version.
H
B
A
F
Wait,
love
me
and
try
phrasing,
the
question
a
little
differently
and
Jeff.
If
this
doesn't
address
your
question,
you
can
bring
yours
back
up.
There's
two
very
different
notions
for
interact
with
OEM
one
is
being
able
to
observe
the
results
of
connectivity,
checks
and
opinions
and
other
things
that
are
started
by
whatever
that's
an
interaction
with
oam
and
one
that
makes
perfect
sense
to
me.
F
B
B
To
to
see
information
coming
back
for
the
resource
management
and
for
other
things,
but
not
in
the
first
version,
there
was
mpls
use
cases
which
acts
to
trigger
some
of
the
OEM
functions
and
that's
a
little
advanced
for
my
first
one.
Once
we
get
some
example
code,
we
might
be
doing
fine,
hey
I,
hope
someone.
I
hope
this
recording
is
really
working,
because
I
will
listen
to
it
carefully
as
well.
Okay,.
B
B
Hey
8,
ok,
so
one
of
the
things
that
I
came
about
when
I
was
doing
two
pieces
of
work,
14,
r,
2
s
and
14
I
to
NSF,
which
is
a
security
management
interface,
is
I,
suddenly
realize
that
this
could
be
running
in
the
midst
of
an
outage
like
we
could
we
oftentimes
do
want
to
manage
our
boxes
for
short-term
during
some
sort
of
attack
during
a
denial
of
service
attack.
You
may
have
a
blast.
Should
you
be
able
to
change
your
protocol
or
change?
How
your
protocol
interacts
during
these
times?
B
B
J
J
Context
where
the
messaging
is
decoupled
from
the
transport
so
that
you
can
change
transports
as
well
as
needed
or
characteristics
of
transports,
as
well
as
as
needed,
and
that
attacks
on
transport
does
not
attack
your
ability
to
do
your
messaging.
So
this
is
a
really
some
actually
work
that
soon
I've
been
working
on
for
over
a
year
and
now
coming
to
a
I
think
some
solid
basis
on
it.
So
you'll
be
seeing
presentations,
lawton,
listen,
I
to
NSF
in
Dotson
mile
arm
wear
these
wear.
These
attacks
are
being
dealt
with,
but
I
would
think.
J
J
H
B
H
B
Next
to
the
protocol,
ok,
we
have
thought
we
have
will
be
doing
some
hackathon
work
and
some
experimentation
on
this.
We
will
report
back
so
I
will
go
on,
but
this
is
another
piece.
Ok,
so
for
the
people
who
want
to
be
able
to
choose
the
transport,
should
we
be
able
to
you
specify
it
now
and
yang
in
the
push
the
transport?
B
D
B
A
So,
to
throw
a
comment
from
prior
discussions,
we
had
discussed
that
rest
comp
is
probably
a
better
fit
for
a
lot
of
the
things
that
are
tending
to
do
just
simply
because
it
is
the
atomicity
that
we
were
expecting
a
lot
for
our
applications
and
less
ambiguity
as
to
what
the
data
stores
actually
mean.
Although
it
sounded
like
the
datastore
question
itself
might
be
getting
some
level
of
refinement.
That
net
cough
may
be
less
of
an
issue,
so
I
think
we're
still
open
to
either
of
them.
D
B
B
K
B
This
work
started
out
with
a
really
good
idea
from
panton
Andy,
when
we
were
describing
a
whole
bunch
of
complex
things,
and
they
gave
us
a
very
simple
solution.
After
that
they
have
been
hard
at
work,
making
that
come
from
rest
comp
and
all
the
net
mod
working
right,
which
has
made
our
life
easier.
So
I
will
try
to
go
through
this.
They
both
agree
to
be
here
and
throw
bricks
if
I
get
it
wrong.
B
In
the
past,
there's
been
candidate
running
and
start
up
an
operational
data,
but
lately,
thanks
to
ken
I,
believe
it
is
your
graph.
Yes,
yep,
there's
a
net
lot
off
state,
which
has
a
different
view:
point
where
you
have
candidate
running
startup
configuration
intended
configuration
and
applied
configuration
and
drive
state.
Our
protocol
strong
did
I
do
something
wrong.
Please
do.
G
B
Everyone
assume
in
the
future
slides
it.
The
line
is
down
and
I'll
send
out
I'll
post
a
corrected
slide
after
the
meeting
looks
like
I
still
have
it,
but
what
we
have
in
the
ephemeral
state
ephemeral
does
not
affect
a
candidate
or
running
or
startup.
It
is
simply
something
that
directly
makes
a
change
to
the
intended
configuration.
B
Stop
there
and
say
that
again
it
does
not
change
any
of
the
candidate
running
or
startup
configuration.
So
if
you
thought
back
to
our
hackathon
example-
and
you
know
anything
about
routing
demons-
the
routing
demon
for
quad,
the
the
approach
we
took
just
simply
sent
routes
into
the
route
table
never
entered
in
any
any
configurations.
That's
why
we
thought
it
was
a
good
example.
B
B
B
Now
and
II
thought
of
a
very
simple
thermostat,
where
some
of
us
have
been
sitting
in
these
rooms
would
like
to
have
a
thermostat
IP
controlled,
where
you
have
a
temperature,
and
then
you
have
operational
state
and
that
derived
temper
desired.
Temperature
might
go
to
in
a
general
configuration
the
intended
config
and
go
all
the
way
through
to
apply
config
and
derive
state.
B
Now,
if
you
had
an
hour
to
asst
scheduler
for
some
reason
who
wanted
to
temporarily
either
schedule
a
change,
it's
hot
and
we're
all
coming
to
the
meeting,
and
so
I
would
like
this
room,
a
little
cooler,
and
then
you
might
have
something
that
would
hold
it
in
case
we
had
a
badly
performing
heating
system
and
it
got
too
cold
to
hold.
These
would
be
to
Idris
clients
that
might
change
the
state
in
the
ephemeral
candidate
which
well
it's
in
femoral,.
B
There's
a
femoral
intended
config
and
then
they
would
go
into
the
applied
and
the
derived
ok
and
that's
the
basic
way
it
fits
inside
of
the
ephemeral
store.
Now
this
is
where
we
would
have
some
rang
challenges.
The
proposed
one
long
ago
was
just
to
say,
FML,
true
and
I.
Think
we've
had
fairly
good
agreement
on
a
femoral.
B
True,
if
we
were
going
to
do
any
change
on
validation,
opposite
of
what
Andy
proposed,
you
would
have
to
do
something
that
says
this
particular
thing
would
have
either
no
checks
or
minimal
checks,
rough,
no
reference
or
full
checks,
that's
where
it
would
come
now.
How
would
it
look
if
you
had
a
press
conf?
You
would
have
breast
comp
day
that
thermostat
desire
temperature,
a
value
of
18
centigrade.
You
might
have
breasts
comp
with
the
context
of
ephemeral.
B
B
B
I'm
glad
you
asked,
because
it
gives
me
a
chance
to
restate
this
and
I'm
the
idea
that
I
had
with
some
implementers
and
I
in
a
design
team,
asked
three
different
types
of
implementers
and
one
said:
gee
I,
do
all
my
checking
and
upper-level
odl
work
that
I'm
doing
for
my
client
and
I.
Don't
want
anything.
I
want
and
availability
to
have
an
RPC
that
just
dumps
in
the
traffic
and
if
I've
made
a
mistake,
it's
my
problem
that
datastore
messes
up
I
torres
said
that
it's
your
chance.
B
If
you
mess
up
your
data
store,
as
andy
says,
it
might
be
really
unuseful
for
their
favorite
clients,
but
you
have
a
specialized
relationship
with
that
one
second,
one
might
be.
You
might
not
want
to
do.
Referential
checks,
no,
must
snow
leaf,
reps,
that's
an
option.
The
third
one
and
the
default
one
is
the
normal
net.
Comfo
validation
for
validation.
Right
now
we
haven't.
Had
anyone
stand
up
and
say
just
reporting.
One
chair
to
another:
I
must
have
the
no
validation.
B
E
B
Is
correct
and
before
I
let
this
come
to
this
point.
I
had
a
little
chat
with
my
ad
who
said
we
have
a
remote
question.
Oh
please.
Let
me
finish
this
and
I'll
pick
up
the
road
who
said
this
routing
interface
has
in
the
architecture.
If
you
shoot
yourself
in
the
foot,
you
shoot
yourself
in
the
foot,
but
please
remote
question:
go.
C
M
Hi,
can
you
hear
me
you
can
hear
you
we
can
hear
you
awesome,
so
I'm
sort
of
I'm
coming
to
this
pretty
new,
although
I've
been
working
on
with
this
OP
state
stuff
for
a
while,
so
I've
got
a
question
when
we
saw,
I
see
that
we
have
like
the
Intendant
config
in
the
ephemeral
can
intended
so
I
I
think
obviously
like
the
ephemeral
would
would
be
overriding.
The
intended
is
that
right.
M
Basically,
you
know,
subscribed
or
whatever
and
checking
intended
and
applied,
and
then
noting
any
differences
you
know
so,
for
example,
if
if
something
doesn't
go
to
applied
after
show
amount
of
time,
raise
an
alarm
so
I
guess
is
there
some
kind
of
indication
in
the
applied
whether
that
applied
config
has
been
over
ruled
or
overridden
by
the
ephemeral,
so
that
you
so
that
you
know
things
that
are
checking
know
that
the
the
can
be
applied,
that
it's
looking
at
shouldn't
match
the
intended
config.
So.
B
Let
me
answer
the
question
in
two
pieces:
in
the
normal
net
comp
and
again
that
conf
folks
and
Ian
the
rest
correct,
there's
both
the
synchronous
and
the
synchronous
right
and
I'm.
Assuming
in
your
question,
you
are
not
asking
about
the
difference
between
the
asynchronous
and
the
synchronous
right.
The
asynchronous
may
have
the
may
have
changed
the
sense
stuff
to
the
applied
config,
but
it's
not
actually
I've
been
applied
yet
where
the
synchronous,
my
understanding,
is
the
acknowledgement
coming
back,
actually
says
that
that
it's
already
been
applied.
I'm
assuming
that's,
not
the
question.
B
What
you're
asking
is
what
happens
if
the
ire
to
us
over
rights,
the
intended
config
and
if
it
overwrites
the
intended
config
the
intended
config
is
changed
and
it
doesn't
and
whenever
it
does,
that
a
synchronous
right
or
synchronous
right.
Those
changes
that
were
in
the
attendant
config
actually
go
down
in
one
suit,
so
the
place
that
the
online
code
has
to
actually
detect
that
the
ephemeral
has
overwritten.
The
configuration
data
is
in
the
Intendant
config
in
the
implementation.
D
B
We
tried
to
do
in
the
routing
demon
Chris.
We
were
just
going
to
put
for
the
the
route
to
the
routing
demon
we
were
going
to
treat.
Our
tres
is
a
different
protocol
and
like
I,
ospf
Riss
would
have
been
a
protocol
and
that
ways
the
way
we
would
detect
inside
essentially
the
values
that
would
be
written
from
the
rib.
That
might
be
a
little
hacking,
but
it
was
yeah.
M
That
one's
a
little
complex,
let
me
can
I
jump
back
to
the
easy.
The
the
easy
example
he
gave
earlier
was
like
a
thermostat,
so
let's
say
I've
net
can't
config.
You
know,
through
the
normal
method,
I've
set
a
temperature
in
that's
an
intended
config
and
then
I
have
a
client
come
in
and
change
that
temperature
into
an
ephemeral.
M
You
know
raise
it
five
degrees
or
whatever
now
I
have
this
manager
out
there
running
around
and
when
it
queries
it's
just
basically
looking
at
you
know
it's
going
to
look
at
the
path
to
the
intended
config
for
the
temperature
and
then
it's
going
to
look
at
the
path
to
the
applied
and
it's
going
to
say
are
these
matching.
Are
you
saying
that
when
the
ephemeral
client
changes
the
the
thing
when
I
look
at
the
in
the
normal
intended
config
path
that
will
change
to
the
ephemeral
value?
That's.
B
How
many
people
would
like
me
to
walk
through
the
changes
or
how
many
people
would
like
me
to
walk?
I
was
going
to
go,
walk
through
the
route
case,
given
the
audience
which
is
not
as
many
net
comp
people
as
routing
people
and
then
I'll
come
back
through
all
the
changes
that
are
listed
in
the
protocol.
Straumann.
A
B
Turned
off
this
one,
so
then
we
did
it
laughter
villain,
okay,
I,
dad,
okay!
This
is
a
very
simple
routing
case:
1
128
tattoo,
with
next
up
one
added
by
netcom,
config,
woohoo,
191
28
that
two
plus
16
added
with
next
hop
to
buy
ephemeral
rights
over
it.
That's
our
little
client
writing
over
it,
and
it's
being
it
that's
happening
for
the
very
example
of
a
das
client.
You
want
to
send
this
traffic
a
different
way.
Maybe
you
want
to
send
it
to
the
sink
soup.
A
B
B
Okay,
so
here's
our
little
higher
to
s
agent,
it's
writing
to
the
ephemeral
intended
that
running
rights
to
the
intended
config
ephemeral
rights
over
it.
Let's
see
how
that
look,
oh
by
the
way.
This
is
the
route
table
in
case
you
missed.
It
has
operational
data
and
route
data,
but
we're
only
going
to
focus
on
route
data
simply
first
time
that
the
running
config
writes
128
dot.
B
B
B
B
B
B
B
I'll
ask:
is
there
any
other
questions
about
the
simple
next
top
case,
as
crisp
and
Jeff
might
say,
you'll
see
is
overwritten
if
this
reboots
it's
gone.
If
another
ephemeral,
client
overwrites
it
it's
gone,
say
the
scheduler
client
over
wrote
it
with
the
next
top
id3.
It's
gone,
there's
no
saving!
It's
like
a
right
that
might
happen
from
a
routing
demon
to
a
central
routing
table
once
you're
right
over.
It's
gone
now,
if
this
seems
really
simple,
Thank,
kenton
and
Andy
for
thinking
of
a
single
solution.
B
Ok,
so
our
basic
tools
for
our
ra
chua
sproat
acall,
are
traceability.
Our
log
and
this
very
simple
use
of
ephemeral
state
the
protocol.
One
should
go
out
through
net
Gump,
hopefully
with
fairly
minimal
changes.
We
hope
that
would
be
good
right.
I
see
you
smile
at
last
I
see
two
people
smiling
in
the
death,
then
cultures
so.
B
1420
are
there
any
questions
about
the
protocol.
Stuff
is
important
to
get
feedback.
The
data
models
are
all
progressing
I'll
go
from
there.
Any
questions
about
the
protocol
in
general,
since
most
of
you
are
thinking
about
the
functional
aspects,
the
next
we're
about
to
go
into
the
deep
dive
on
that
come
please
Joe.
N
Clark
it,
it
may
not
be
what
you
would
consider
basic,
but
it
goes
to
my
comment:
I
made
on
the
straw,
man,
the
whole
pane
of
glass.
Thinking
to
me
at
least
brakes
literally
shatters.
If
you
do
the
next
right
to
intended,
config,
because
I
think
what
you're
saying
is,
if
I
then
right
to
intended
config
with
like
netcom
to
the
actual,
what
you
call
normal
data
store
that
pane
of
glass
shatters
on
top
of
it
and
I
win
by
default,
because
I'm
allow.
N
B
F
Joel
halperin
that
I
think
that
what
people
get
confused
is
your
phrasing
which
says
it
does
this
by
default.
But
you
can
do
that.
We
say
you
can
configure
it
to
preserve
to
pay
attention
to
it.
If
you
we
do
have
a
what
is
it
setting
if
you
haven't
changed
this
setting
up
any
particular
way
and
the
setting
is
then
that
normal
changes
to
config
take
precedence
when
they
get
applied,
but
that's
that's
the
death
simply
a
behavior
among
the
allowed
range,
because
different
operators
have
different
requirements
so.
B
N
Normal
or
the
the
running
config,
and
if
it
is
the
default,
is
it
mandatory
to
implement
something?
That's
not
that
it's
not
this
is
it
if
you're
saying
it's
the
default,
but
is
an
operator
can
I
assume,
then
that
my
vendor,
who
is
implementing
the
agent
side
of
I
to
RS,
would
allow
me
to
do
something
where
the
pane
of
glass
takes
precedence,
no
matter
what
one
would
hope
you
have
an
intelligent.
A
B
B
Ok,
all
of
you
came
to
learn
a
little
bit
about
our
2's.
Are
there
any
questions
on
the
basics?
Because
then,
if
we
deep
dive
into
the
net
comp
and
restaurant
details,
I
want
to
make
sure
that
you
don't
get
lost.
Anybody
have
any
concerns
about
this
think.
Does
this
seem
reasonable?
How
many
people
think
this
is
a
reasonable
first
pass?
B
B
M
So
the
scheduler
client,
so
that,
from
what
I
heard
from
you
earlier,
the
you
know
the
I
have
an
overlay
in
of
the
ephemeral
on
this
slide
here
in
the
yellow
top
yellow
box
of
the
route
128
to
16
ID
next
type
ID
to
so
that's
overlaying
that
would
the
running
data
store.
I
think
that
this
diagram
is
wrong.
M
You
can
tell
me
if
I'm
wrong
in
my
viewing
of
it,
but
the
black
arrow
that
you
have
going
to
running
data
store,
I,
don't
think
is-
is
possible
now
right
because
when
you're
pointing
at
that
in
it
the
intended
config
with
the
overlay,
that
is
the
running
data
store.
So
the
way
that
this
I
think
should
be
drawn
is
that
that
black
line
goes
down
into
the
applied
config.
So
you
know
what
I'm
saying
like
if
you
overlay
the
intended
config
that's
what
the
running
datastore
is.
B
B
B
M
M
E
B
M
E
E
E
B
E
B
E
M
A
This
is
jeff,
so
I
think
we're
having
our
usual
hang
up
a
nitrous
that
certain
of
these
words
have
very
distinct
meanings
at
netcom,
so
running
in
the
context
of
net
conf
is
from
the
configuration
data
store,
which
we
would
call
static
and
fake
out
of
hrs
the
intended
config.
If
I'm
remembering
my
verbiage
correctly,
is
the
merged
view
of
the
running
config
from
static
merged
with
the
hrs
ephemeral
configuration
do
I
have
that
right,
Andy
yeah.
E
D
A
L
M
M
G
G
I,
like
the
new
arrow,
have
this
account
so
just
thinking
these
slides,
maybe
I
are
confusing
because
the
running
datastore
you
have
a
line
going
down
to
intended.
Config,
but
really
the
intended
configuration
is
the
composite
of
the
running
days
for
the
persisted
configuration
plus
the
ephemeral.
Yes,.
B
O
O
O
B
Okay,
the
net
conferen
rest
count,
folks
have
been
very
good
and
they
have
a
form
that
you
essentially
walk
through
add
to
fill
in
new
capabilities.
Is
that
a
good
way
to
describe
it
that
I
use
the
right?
You
usually
have
a
very
nice
form
when
I
find
it
to
describe
what
you
want
to
change
in
that
conference
comp
so,
hopefully
I
filled
in
the
form
brightly.
So
again,
what
is
in
a
femoral
data
store
in
a
more
precise
way?
It
is
not
intended
to
decipher
a
survivor
reboot.
B
A
N
Joe
Clark
again
I
make
sense
I,
don't
necessarily
like
it.
I
think
that
it
should
play
more
with
the
whole
priority.
The
config,
the
persistent
configure,
the
non
ephemeral
config
should
play
more
with
the
ephemeral
config
in
terms
of
priority,
because
to
me
it
makes
sense
until
you
say:
oh
well,
it's
the
last
right,
so
you
have
to
take
into
account
that
temporal
aspect
of
it
too
you're
not
just
looking
at
ephemeral.
So
you
have
to
take
into
account
that
that
ephemeral
pain
shatters.
N
B
The
good,
the
good
point
is
once
we've
sort
of
close
this
first
one.
My
suggestion
is
that
you
and
Russ
work
on
a
design
team
to
go
through
the
other
one
I
just
will
close
one
off
and
if
you
think
of
a
better
time
by
the
time
it
gets
standard,
you
can
always
overwrite
the
current
stuff.
So
that's
that's
my
strong
suit.
That
was
my
agreement
with
my
co-chair
and
AD
is,
but
please,
you
and
Russ
I
hope
we'll
work
on
that
over
the
next
couple
months.
Ok,.
B
Multiple
clients
can
write
to
the
one
pane
of
glass,
but
the
highest
party
wins.
You
can
signal
a
capability
for
note
via
hello
and
we
went
through
the
air
checking
the
dependencies
are
the
ephemeral
flag.
It
doesn't
look
like
we
have
too
many
people
looking
for
the
validation
flank,
but
I'll
have
to
check
that
my
hessian
and
so
and
yang
models
must
support
notifications
of
right,
complex.
B
New
operations
you'll
need
a
link,
config
link
ephemeral
to
a
target
config
at
least
that's
my
understanding
for
link
and
bulk
right,
but
I
think
the
that
bulk
right
was
for
the
for
the
routes
and
everything
in
our
data.
Centers
data
models
uses
an
RPC,
so
I
don't
think
we
need
to
do
bulk
right
in
the
first
one.
It
just
might
come.
I
just
didn't
see
any
you
aren't
doing
any
bulk
rights.
Are
you
you're
just
using
regular
rights
for
the
topology.
B
B
L
B
We're
having
the
discussion
on
what
you
need
to
dishonor
on
the
read
and
write
the
three
choices.
Were
you
trust
the
client?
You
don't
do
must
referential
checks
and
you
do
the
full
checking,
and
this
is
just
something
that
would
allow
you
and
yang
to
specify
whether
you
want
to
have
that
/
model
on
some
models.
You
may
never
want
to
allow
someone
to
do
a
fast
right
for
maybe
four
the
route
you
might
want
to
allow
the
bulk
right,
but
maybe
for
the
ospf
interface,
never
going
to
allow
that.
B
E
Your
previous
life
had
the
this
validation
specified
in
the
yang,
but
that's
not
really
correct
you.
You
want
it
so
that
the
server
is
like
saying
I'm
only
doing
one
two
or
three
and
not
the
client
picking
it
see
if
the
client
picks
it
client.
Once
it's
something
says:
Oh
set
this
route
over
here
with
no
validation.
Client
do
comes
along
and
says,
set
my
route
over
here
with
validation.
Well,
yank
validation
doesn't
work
on
one
edit
at
the
time.
E
Yang
Yang
doesn't
have
any
concept
of
just
validating
and
edit
it's
the
whole
entire
datastore.
The
I
had
a
big
argument
with
someone
at
Cisco
many
many
years
ago
about
why
we
do
checking.
He
says:
I'm
writing,
routing
code
and
and
I'm
I
can't
be
too
bothered
to
check
parameters
because
s
be
fast
and
I
said
well
with
the
box.
Sec
faults
how
fast
you're
running
now.
So
the
reason
we
check
stuff
is
to
keep
the
box
from
crashing.
Yes,.
B
B
Okay,
we
have
left
an
I
er
to
us
time
slot
after
the
net
comp.
In
case
we
get
embroiled
in
some
large
discussion.
We
don't
have
any
topic
for
it.
It's
just
left
in
case
you
guys
need
a
room
after
we
have
our
joint
discussions.
I,
don't
anticipate.
We
will
send
a
note
out
during
the
net
comp,
meaning
if
we
need
it.
Thank
you
for
your
time.