►
From YouTube: DASH Behavioral Model Weekly 20220317 (March 17, 2022)
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
That
we
didn't
have
any
software
model
that
could
actually
do
the
packet
processing
and
this
time
we
want
to
start
with
something
that
can
actually.
A
Run
the
same
traffic,
although
at
the
lower
lows
and
provide
provide
the
standard
of
the
behavior
for
this
kind
of
pipeline,
which
is
dash
right
right
now
we
want
to
do
the
communication
between
vms
in
the
future.
It
will,
of
course,
as
dash
will
grow
and
mature.
It
will
also
include
other
pipelines
like
load,
balancing,
maybe
more
stuff
like
net
and
other
things,
but
we
need
to
start
somewhere.
A
So
we
are
starting
with
the
vm
to
vm
or
v-net
to
v-net
communication,
which
includes
things
like
mainly
v-net,
routing
and
connection
tracking
and
yeah,
and
also
like.
We
kicked
off
aha,
as
you
may
already
know,
but
it's
far
from
being
in
the
behavioral
model.
A
It
will
take
some
time
to
get
there
so
right
now
the
focus
is
on
unit
to
unit
and
we
have
a
very,
very
basic
scenario
for
doing
the
v-net
routing,
which
includes
for
now
ipv4
with
tcp.
So
we
have
and
also
some
functionality
missing.
So
we
have
kind
of
directions
to
to
grow
the
behavioral
model,
and,
let's
say
vertically,
by
improving
the
stuff
that
you
already.
A
By
adding
write
down
more
protocols
and
in
the
future
more
functions,
so
here
we
have
a
list
of
different
things
that
are
still
missing
from
the
behavioral
model.
A
By
the
way,
there
is
a
readme
file
that
I
will
post
a
link
to.
That
explains
how
to
use
a
behavioral
model.
It
is
spending
on
merging
my
pull
request.
I
will
merge
it
today
and
I
will
send
you
a
link
on
the
how
to
how
you
can
set
up
the
behavioral
model
and
how
you
can
run
it.
So
if
you
have
any
problems
with
that,
you
can
reach
out
to
me
and
we
can
do
more
changes
more
fixes
to
it.
A
But
with
regards
to
two
things:
the
implementation
of
dash
pipeline,
which
is
the
p4
code
and
the
improvements
to
the
compiler
and
the
simulator
simulator.
So
he
can
actually
do
all
of
the
expressions
that
we
have
like,
for
example,
connection
tracking,
which
is
only
available
in
pna,
for
instance,
which
is
not
really
supported
by
the
simulator.
A
So
all
these
areas
are
up
for
improvement,
and
this
is
what
this
list
of
tasks
consists
of.
So
we
have
right
now
how
many
13
of
them,
so
probably
let's
go
one
by
one.
I
will
explain
what
they
are
about
and
you
can
probably
sit
on
that
for
some
time
and
decide
what
are
the
areas
that
you
would
like
to
participate
in
and
to
provide
further
contributions
to
to
this
project?
C
A
Yes,
yes,
so
you
can
find
it
in
the
behavioral
model
itself.
The
code
for
behavioral
model
is
give
me
one.
Second,
I
will
post
a
link
to
it.
No.
C
Just
for
clarification
right
couple
of
clarification
questions,
then
we
can
go
through
the
list
right
and
then
that's
one
right.
So
there
is
a
p4,
and
if
you
have
your
own
p4
code,
p4
tool
chain
gives
you
a
way
to
create
a
behavioral
model,
configure
it
send
some
packets
and
test
it
right.
So
that's
exactly
there
a
while
back
and
then
on
top.
The
very
first
thing
would
be
a
psi
dot
p4
right.
C
That
would
make
a
psi
underlay
and
then
you
have
another
set
of
p4
code
that
will
give
you
the
we
need
to.
We
need
communication,
ignore
the
connection
tracking
and
all
the
details
later,
but
there
is
that
com
comes
and
gives
you
an
overlay
on
top
right.
So
can
you
talk
a
little
bit
about
the
psi
p4
bm
before
we
look
at
hey?
How
do
we
take
the.
A
A
Api
standard,
which
is
called
p4
runtime,
and
it
is
used
for
configuring,
your
pipeline
after
you
compile
it
to
some
backend.
So
in
our
case,
the
back
in
this
bmv2
and
when
you
compile
it,
you
can
use
before
runtime
to
configure
the
pipeline
to
populate
the
table,
so
the
transformations
of
the
packet
will
be
done
according
to
those
rules.
A
What
we
need
to
do
is
to
align
it
with
the
sonic
approach,
so
sonic
is
using
psi
apis
and
we
need
to
define
a
new
set
of
apis
based
on
the
p4
code
for
how
we
can
configure
our
our
dash
yeah,
our
dbu.
A
So
in
order
to
do
that,
we
decided
not
not
to
have
like
two
different
things
to
support.
We
want
to
only
focus
on
the
p4
code
and
everything
else
will
be
generated
based
on
that.
So
today
you
can
find
that
the
p4
code,
as
I
mentioned,
keeps
the
vnet
routing
tables
it
has
the
acl
tables
and
so
on,
and
for
every
of
those
tables
there
is
a
site
api
generated.
C
Sorry
to
interpret
it
so,
but
before
I
get
to
overlay,
how
do
how
do
you
see
the
underlay
right?
Do
you
see
it
as
psi,
dot,
p4
and
then
compile
and
generate
the
behavioral
model
that
is
supporting,
let's
say,
as
my
first
step
that
and
then
add
additional
well
underlay.
A
Is
very
minimal.
Underlay
is
very
minimal
in
our
case,
because
there
is
really
no
underlay
routing.
Nothing
like
that.
So
we
have
a
few
objects
that
will
we
will
need
to
support,
to
define
maybe
some
fields
in
the
package
like
the
neighbors
will
in
the
future.
A
We
will
also
need
mirroring,
policers
and
stuff
like
that,
but
right
now,
underlay
is
just
a
set
of
global
variable
values
that
you
provide
to
your
dpu,
which
is
the
ip
address
of
the
appliance
itself,
the
mac
address
of
the
appliance
and
the
mac
address
of
your
neighborhood
or
switch.
So
this
is
currently
all
of
the
underlay
that
we
have,
but
it's
not
really.
D
I
think
you
know
one
thing
is:
I
think
what
one
card
is
correct
me
if
I'm
wrong,
but
I
think
what
you're
trying
to
basically
ask
the
question
in
terms
of
how
are
you
going
to
generate
the
psy
apis
for
the
underlay,
much
like
how
we
are
basically
trying
to
generate
the
psi
apis
for
the
overlay
using
the
p4
models?
D
We
essentially,
you
know,
define
the
apis,
but
but
we
have
taken
a
rather
different
approach
in
dash,
which
is
to
say
that
okay
start
with
the
p4
model
instead
of
visio,
diagrams
and
topology
diagrams
and
so
forth,
and
then
go
ahead
and
use
the
peter
model
as
a
source
of
truth
to
define
this
sky
apis.
C
C
As
for
the
model
that
you
are
talking
about
the
basic
dot
p4
so
that
one
it's
all
combined
in
in
the
in
the
data
plane,
dot,
p4
right,
it's
underlay
and
overlay-
those
are
all
combined.
So
your
pipeline
will
have
all
those
things
when
you
define.
I
don't
think
so.
They
took
something
from
the
side
up
before,
so
they
are
rewriting
that
from
the
scratch
the
pipeline
itself,
it
will
cover
both
underlay
as
well
as
overlay
from
the
underlay
overlay.
So
what
yeah?
But
what.
C
Was
right
now,
based
on
this
conversation
just
now
so
right,
so
this
psi
api
alone
was
under
whatever
tables
already
exist.
We
are
not
reinventing
the
wheel.
We
are
just
taking
those
apis
from
the
psi
community
already
and
whatever
is
not
available.
We
have
to
make
sure
that
it
gets
compiled
and
automatics
iaps
are
getting
generated
for
the
tables
or
new
new
things.
We
are
defining
right
now.
D
A
A
Is
that
yeah
we
have
some
p4
code
included
in
dash
pipeline
right
now
for
underlay.
The
difference
between
p4
code
for
underlaying
p4
code
for
overlay
is
that
for
overlay
we
generate
apis
based
on
p4
for
underlay.
Probably
some
manual
stitching
will
be
done
to
standard
size
apis
to
be
able
to
program
those
tables
and
yeah,
and
it
is
just
a
subset
of
of
today's
site.
C
Okay,
so
I
can
summarize
and
see
you
correct
me
if
my
understanding
is
right
right,
there
are
two
ways
of
doing
it:
one
could
have
been
a
full
blown
or
even
a
partial
psi,
dot,
p4
and
a
behavioral
model,
and
to
that
that
behavioral
model
could
have
worked
with
another
behavioral
model
for
overlay,
which
comes
from
the
series
pipeline
dot.
P4
right,
that's
not
the
approach
we
are
taking.
We
have
a
a
single
you
know,
p4
from
which,
which
we
generate
partial.
C
C
E
E
E
E
Based
on
the
you
know,
various
stages
we
are
on
the
p4
code.
We
will
be
defining
ingress,
all
three
aqua
stages,
egress
and
then
everything
and
then
based
on
that,
whatever
the
psy
api
generator
python
script
that
you
have,
it
will
basically
look
up
all
the
tables
and
actions
and
everything,
and
then
it
will
give
us
the
the
dot
h
file,
the
psi
api
dot,
h5
correct.
D
A
D
So
so
so
you
know
hopefully
quickly
you
know
considering
this.
All,
basically
is
you
know
it's
it's
all
targeting
for
the
winner
to
v
net
communication,
as
you
mentioned,
considering
that
you
know,
do
you
not
think
that
we
should
also
track
the
for
h.a
as
well
in
this
or
do
you
think
that
that's
actually
a
separate
item?
F
F
This
is
the
first
and
then,
when
we
get
to
aha,
it
will
result
in
this
being
updated
and
we
have,
you
know,
excites
been
working
on
that
in
the
background
as
a
separate,
parallel
exercise
and
then
that
will
come
in
you
know
when
the
community
gets
more
done
there
and
that
and
yes,
this
model
will
have
to
be
updated
once
once
we
have
the
aha
all
documented
and
and
the
community
agreeing
on
the
approach.
So
this
is
just
the
first
of
many.
This.
D
One
sounds
good,
so
the
next
thing
is
basically
on
the
fragment
handling
right.
Why
are
we
basically
talking
about
fragment
handling
on
the
connection
tracking?
Only
should
it
not
be
just
you
know
on
its
own
to
say
that
okay,
you
know
we
do
need
to
handle
fragment
handling,
whether
we
are
doing
connection
tracking
or
not
connection
right.
F
In
practice,
that's
not
really
true,
so
when
you,
you
can't
accept,
you
know
the
fragments
without
having
the
connection.
So
we
have
talked
about
this
and
you
have
a
connection
and
then
you
have
temporal
connections,
which
means
they're
temporary.
F
Every
time
you
receive
a
fragment,
you
have
to
do
things
in
a
slightly
different
way
and
they
get
timed
out
very
quickly
because
we
don't
it's
not
like
timing.
Other
connections,
you
can't
just
send
some
fragments
and
and
leave
it
dangling.
So
we
we
have
proposals
for
that
we've.
Given
those
proposals
in
writing
to
melanox
and
they
are
part
of
the
connection
you
have
to
have
a
connection
push
you
can't
have
fragments
without
a
connection
and
connection
always
perceives
fragments.
D
Okay,
okay,
so
the
last
thing
that
I
just
want
to
ask
about.
You
know
these
all
these
seven
use
cases
when
we
talk
about
when
it
when
you
basically
zoom
into
that.
What
we
really
see
is
that
essentially,
these
are
really
talking
about
how
we
handle
slow
path.
Behavior
right,
however,
when
it
comes
to
fast
path,
it's
essentially
is
common
across
all
of
them,
because
once
the
flow
is
set
up,
essentially
it
is,
it
is
the
common
part.
So.
F
It's
sort
of,
except
for
in
the
common
part,
if
you
look
at
the
forwarding
table
documentation
that
there's
all
kinds
of
things
in
the
boarding
table
that
actually
yeah
do
need
to
be
supported
for
these
seven
services.
So
there's
so
yeah.
If
you
were
thinking
of
the
comment
thing
as
being
at
the
venus,
no,
it's
more
than
that.
The
forwarding
table
has
to
be
able
to
handle
all
seven
services,
which
means
all
different
types
of
encapsulations
and
behaviors,
depending
on
which
service
so
in
a
simplistic
way.
Yes,
but
it's
not
quite
that
cool.
F
Yeah,
so
we
tried
to
account
for
the
use
cases
that
we
know
so
the
document
that
we
have
in
the
where
the
forwarding
table
is
defined.
It
tries
to
account
for
the
seven
services.
However,
just
remember
that
those
seven
services
that
weapon
is
needed
from
microsoft's
cloud
and
nothing
more,
but
for
the
enterprise
there
may
be
more.
We
can't
account
for
what
we
don't
know.
So
as
more
services
are
added,
you
know
over
time.
Of
course
the
forwarding
table
may
have
to
be
or
how
that
works
and
the
behaviors
may
have
to
change.
F
So
there's
there's
always
a
chance
in
the
future
that
we
have
to
update
it
with
the
new
services,
so
we're
trying
to
account
for
it.
But
you
know,
obviously
we
can't
account
for
the
services.
You
know
that
google
might
want
or
ibm
might
want
or
dell
might
want,
or
we
build
an
outpost
like
appliance,
and
we
need
some
new
thing
there
like
there's.
It's
definitely
gonna
change
over
time.
Okay,
so
the
probability.
D
F
D
G
F
B
F
Because
I
had
to
reset
my
computer,
but
did
you
guys
discuss?
I
just
got
off
the
phone
with
francis
mccow
and
he
told
me
that,
as
of
like
within
the
last
couple
days
that
melanox
agreed
to
use
the
p4
connection,
state
tracking
model
from
amsambo
is
that
is
that
your
understanding,
marion
or
like
do
we.
A
Yeah,
so
actually
we
reviewed
it
with
mario,
who
is
the
author
of
the
pull
request
seems
like,
with
the
few
more
things
that
are
currently
under
discussion
in
the
pna
community.
We
will
be
able
to
accommodate
for
the
future
use
cases
as
well.
So
what
we
decided
to
do
is
to
work
on
the
compiler
and
simulator
support
for
this
pna
model
proposed
by
fensandoya.
F
G
A
Think
I
think
we're
gonna.
Do
it
the
in
a
slightly
different
way.
We
want
to
merge
the
initial
version
and
then
add
improvements
according
to
to
all
the
use
cases,
but
we
want
to
have
at
least
something
working
in
simulator,
and
then
we
can
improve.
A
Yeah,
but
the
problem
is
that
not
everything
can
be
expressed
even
in
the
dna
right
now.
So
that's
the
bigger
problem.
A
It
just
requires
more
work
on
the
simulator
and
the
compiler.
It's
not
like.
We
just
don't
have
everything
from
day.
One.
F
A
It
will
take
like
a
couple
of
months.
I
think
I
think
we
only
started
working
on
the
simulator.
G
G
F
With
this
new
information,
christina's
like
they
didn't
cover
all
the
cases
for
connection
to
state
tracking,
so
that's
not
done
anymore.
F
G
A
G
F
F
In
the
last
meeting,
we
had
don
fingerhut,
who
came
on
and
and
he's
been
sending
emails
and
pull
requests
and
what
he
did
was
he
came
on
and
said
this
is.
This
is
connection
tracking
from
the
linux
open
source
community.
Oh.
H
F
Very
very
complete,
and
it
was
a
matter
of
okay.
If
we
have
code,
that's
tried
and
true,
we
know
worked
because
it's
in
the
it's
been
around
for
a
long
time,
then
we
can
convert
that
to
p4
and
that's
where
we
left
off
since
then,
there's
been
a
different
agreement
to
use
a
different
model
and
I'm
okay
with
that.
But
my
point
is:
I
want
the
community
to
be
aware
of
it,
because
there
was
a
fair
number
of
people
in
the
last
meeting
and
we've
changed
directions.
F
So
if
this
is
a
good
direction,
I'm
fine
with
it,
but
I
also
need
to
date
and
need
everybody
to
be
aware
that
we've
changed
direction.
F
F
After
that
and
says
it's
missing,
this
is
missing
that
and
that's
what
happened
and
as
we
address
those
of
course,
those
things
that
it's
missing
need
to
be
addressed
and,
and
some
of
them
are
like,
as
we
just
said,
they
take
a
couple
months.
So
it's
not
a
matter
of
this
is
what
we
adopted,
that
that
was
a
rough,
and
it
was
always
said
that
that
is
the
first
proposed
draft
of
the
first
keep
or
mo
and
now
we're.
F
We
still
have
a
lot
of
work
to
do,
but,
as
I
always
say
to
all
my
people,
please
submit
what
you
have.
You
got
to
start
somewhere
and
that's
what
happened
they
started
somewhere
now.
We
need
to
finish
the
work.
B
Right
right,
yeah,
but
what
I'm
trying
to
say
is
that
you're
right,
that's
not
adopted,
it
wasn't
the
right
term.
That's
the
the
model
that
we
used
as
a
starting
point.
Now
we
are
just
saying
we
use
a
slightly
different
starting
point,
but
that's
that's
the
announcement
we
have
to
do
to
the
community.
It
has
not
to
do
with
the
open
contrail.
F
I
think
what
we've
agreed
here
is:
if
we're
going
to
use
the
p4
model
that
sort
of
is
in
existence,
then
the
action
item
is
to
at
least
compare
it
to
the
contrail
model,
to
make
sure
that
it's
actually
not
another
skeleton
of
an
implementation
that
it's
actually
covering
all
the
cases
that
the
contrail
code
covers.
That's
the
action
out
of
this
right.
B
Right
perfect
and
that's
not
that's
not
diverging
from
what
was
said
yesterday.
We
just
use
a
different
starting
point,
but
the
the
goal
is
the
same.
So
we
haven't
really
changed
direction
as
far
as
making
sure
that
everything
that
it's
open,
it's
in
open
control,
it's
gonna
be
included.
F
B
E
F
Yeah
yeah,
I
don't
think
the
side.
I
don't
think
the
side
changes
it
changes
the
behavior
model
right.
So
you
know
the
psi
is
like
half
the
battle
right,
not
even
half
the
battle.
The
behavioral
model
is
the
battle
and
then
that's
why
we
agree
that
we'll
compile
size
out
of
the
behavioral
models.
We
need
to
hear
the
model
correct
right
and
the
side
yeah.
I
agree.
Maybe
it'll
fit
out
the
same
side,
but
the
object
itself
will
will
change.
C
Just
for
clarity,
so
the
the
comparison
between
going
with
the
contrail
model
versus
the
pinsander
model
is:
if
you
go
with
the
control
model,
we
still
have
to
write
new
p4
files
to
have
all
the
cases
covered
in
it,
whereas
the
pencender
model
already
has
the
p4
files
that
implements
this.
We
just
need
to
make
sure
it
covers
all.
The
cases
is.
F
That
that's
exactly,
I
don't
have
a
problem
with
it
going
to
the
the
the
prince
on
the
mall
at
all,
but
I
just
I
just
would
have
a
problem
if
we
don't
compare
and
make
sure
that
what
we
have
found
out
over
the
last
two
weeks
was
what
we
had
was
incomplete,
so
we
know
there's
open
source
and
we
know
it's
tried
and
true.
So
what
we
don't
want
to
do
is
lose
sight
of
that,
and
not
do
this
comparison,
because
then
we'll
just
go
through
this
whole
exercise
of.
Why
did
this
not
work?
F
Why
did
that
not
work
like
so?
At
least
we
know,
so
the
only
action
out
of
this
is
not
to
change
anything
you're
doing
it
sounds
good.
I'm
glad
sensato
is
actually
contributing
that,
let's
just
do
the
comparison
and
make
sure
that
we're
covering
all
the
bases.
If
we
are
no
problem,
we
just
keep
moving
on.
If
we
aren't,
then
we're
going
to
have
to
update
them
all
is.
C
Can
I
suggest
something
here,
so
I
I
think
we
just
became
aware
of
this
contrail
only
like
a
day
or
two
ago
right.
It's
see,
I
think
that
gave
additional
details
about
how
efficiently
we
can
track
connections
age
them
so
that
the
the
flow
state
doesn't
blow
right.
That
is
good,
but
it's
a
software
implementation.
C
What
I
suggest
is
if
we
could
give
another
day
or
two
for
few
more
of
us
who
have
done
hardware,
implementations
in
the
past,
to
review
and
then
and-
and
then
we
can.
I
don't
know
if
adopt
is
again
the
right
word,
but
to
kind
of
say
that
hey
this
is
the
model
for
flow
tracking
flow
state
reduction,
and
then
we
can
go
definitely
and
go.
Look
at
you
know
either
compare
or
enhance
the
existing
pna
code
to
the
next
level.
I
see
it
as
there
are.
C
The
way
I
see
it
is
is
a
continuous
improvement,
which
is
good,
but
I
I
think
point
is
taken
in
terms
of
what
you
are
saying
right.
If
we
have
two
three
of
them,
they
will
conflict,
but
but
at
least
if,
if
you
give
us
couple
of
more
days,
I
think
it's
it's.
My
suggestion
is
to
look
at
the
open
control
model
from
a
hardware
implementation
point
of
view
and
multiple
implementations
that
will
evolve
over
a
period
of
time.
C
D
Achievable
at
the
same
time,
but
at
the
same
time
guys
you
know:
do
we
not
need
to
first
really
come
up
with
the
model
with
all
this?
Basically,
you
know
requirements
that
we
have
for
for
for
dash
requirement
purposes
for
the
connection
tracking
making
sure
that
it
covers
all
the
bases
before
we
take
it
to
the
p4
community.
D
Verification
can
be
discussed
within
you
know,
with
respect
to
the
pna
architecture,
to
say
that,
okay,
you
know
what
are
the
the
modification
or
what
are
the
any
enhancement
that
you're
suggesting
to
them
right.
So
I
think
there
is
some
dependency
that
I
see
that
first,
we
have
to
really
come
up
with
something
and
then
then
you
know
start
engaging
with
the
people
coming
in.
That's.
C
The
second
part
of
it
right
right
now,
you
have
to
first
identify
the
pipelines.
Then
if
you
see
that
anything
new,
then
you
can
go
back
to
the
p4
architectural
group
and
then
ask
them
to
implement
it
right.
The
first
step
is
to
identify
the
pipeline
and
is
there
anything
new
required
right
and
then,
as
I
agree
with
then
get
on
that
this
control
is
after
implementation,
and
then
we
need
to
look
at
from
the
hardware
perspective
in
the
pipeline
perspective.
C
G
Which
pr
was
it's
in
the.
D
Early
pr's,
you
go
if
you
go
down
scroll
down
yeah
here
it
is
oh
yeah.
C
Right
on
the
p4
and
the
versioning
right,
I
think
we
had
one
p4
specification
from
marian.
That
was
good
and
then,
when
I
think
I
remember
everyone
talking
about,
why
not
align
it
with
pna
because
there's
a
lot
of
work
and
lot
of
expertise
there.
I
think
that
was
I
I
will
say
it
as
we
agreed
on
and
said,
and
then
mario
took
and
enhanced
it.
But
since
pna
itself
is
evolving,
it's
not
a
bad
idea
again.
C
F
I
kind
of
like
that
agreement
that
doesn't
change
what
we
need
to
do.
People
still
need
to
go
back
and
look
at
it,
re-look
at
it
in
parallel
re-examine
the
control
re-examine.
You
know
whether
this
actually
works
for
hardware
that
can
be
done
in
parallel,
and
you
could
still
agree
that
today
it
seems
like
using
the
pna
mode.
As
a
starting
point
is
a
good
starting
point.
F
I
don't
think
we
need
to
debate
that
it's
whether
it
needs
to
be
updated
and
enhanced
would
be
definitely
a
debate
for
the
future
after
people
continue
to
do
more
work
that
what
I
do
like
about
the
pne
model
that
they're
proposing
it
it
has
been
deployed,
so
it's
not
like,
and
it
has
been
deployed
against
services
like
this.
So
I
I
feel
like
it's
at
least
a
good
starting
point.
E
C
E
One
more
question
is
like
we
have
a
participation
from
panzando
right.
I
don't
know
if
they
can
be
because
they
know
what
they
are
doing
in
pna
and
we
can.
If
we
can
get
some,
you
know,
clarity
from
them,
then
we'll
see
how
complete
that
is
for
us
to
use
in
behavioral
model
right.
That
will
help
us
crew.
G
E
F
G
F
B
F
I
think
it's
okay
to
start
with
the
pna
model,
one
it's
kind
of
also
trying
to
model
and
two
you
know
it's
it's
it's
much
better
than
where
we
are
today,
but
today
we're
we
don't
cover
any
of
these
faces.
So
I
I
think
it's
a
good
place
to
start.
There's
some
tried
to
tune
us
to
it.
F
I
think
it
is
agreed
by
the
people
on
this
call
that
in
parallel
we
need
to
go
up
and
look
at
this
and
make
sure
in
parallel
that
it
doesn't
need
to
be
enhanced
and
or
you
know
it
doesn't-
you
know
worst
case
it
doesn't
even
satisfy
our
requirements.
I
highly
doubt
that
so
I
think
it's
more
about
enhancing
potentially
to
cover.
G
We
were
talking
in
december
about
whether,
if
we
were
wanting
to
bring
changes
forward
to
the
pna
working
group
and
how
long
it
took
to
get
changes
etc,
like
we
had
talked
about
doing
that
back
in
december.
G
Wonderful,
wonderful,
okay!
So
let's
move
forward
marion
if
we
can-
and
we've
got
basically
mario
marion
and
fincat
offering
to
look
at
this-
is
that
right.
F
A
Yes,
so
actually,
what
we
think
is
the
most
important
and
most
blocking
thing
we
are
already
working
on
that.
This
is
support
for
the
connection
tracking
and
the
simulator
and
then
compiler.
So
because
we
need
to
start
somewhere.
We
need
to
be
able
to
run
at
least
something
that's
what
we
are
doing.
A
So
they
don't
really,
they
don't
need
anything
else.
They
can
be
done
on
its
own.
They
have
support
from
from
software,
just
need
to
be
coded
in
the
in
our
p4
behavioral
model.
G
So
nvidia
is
assigned
here
and
here
and
here
and
here
and
so
udp
support.
We
would
need
a
volunteer
in
ipv6.
C
Can
I
take
udp
support
and
then
there
are
some
questions
we'll
have
marianne
in
terms
of
how
to
do
it.
I
know
it's
generic
software,
it's
on
c
plus
plus,
but
there
may
be
some
questions
in
terms
of
the
environment.
What
not
and
we
need
to
come
up
with
mechanism
on
how
that
information
is
disseminated
to
the
rest.
E
C
C
If
you
think
we
should
do
offline,
I'm
okay,
I'll
keep
quiet,
but
at
least
putting
three
to
five
bullets
under
each
of
them
will
give
us
more
clarity.
The
reason
I'm
saying
is,
I
have
experience
with
the
p4c
code
directly.
I
have
changed
it
coded
it.
I
have
directly
coded
the
behavioral
models,
also
that
c
plus
plus
so
I
see
so.
C
G
So
what
what
I'm
going
to
do
guys
is
when
we
get
off
the
phone
michael
miele
and
I
will
create
this
list
in
the
proper
spot
in
the
dash
repo
and
then
we
will
publish
it
out
to
you
guys
and
then
you
can.
Let
us
know.
F
A
So
probably,
I
would
need
to
provide
details
description
for
every
of
the
tasks.
Yes,
yeah.
E
C
C
F
And
I
should
just
assign
leads,
though,
like
we're
all
saying:
go
back
to
nvidia,
I
mean:
should
there
not
be
at
least
if
somebody
can
at
least
help
scope?
What
needs
to
get
done
to
get
the
lines
in
here
then?
Maybe
we
just
get
the
leads
for
that,
and
then
we
can
reassign
if
they're,
not
the
right
person
to
do
all
the
work.
We
can
add
more
people,
but
I
don't
want
it
to
end
up
we're
all
just
waiting
on
nvidia
like
we
should.
C
No,
definitely
that's
why
I
think
all
of
us
are
volunteering.
I
think
if
we
know
all
we
are
saying
is
we
need
at
least
three
to
five
bullets
under
each
one
of
them
either
we
could
do
it
right
now.
Okay
or
we
could,
you
know,
really,
you
know,
go
go
offline
either
way
is
okay.
I.
A
I
have
a
next
meeting
in
15
minutes
so,
but
I
can
provide
a
description
in
like
couple
of
hours
after
I'm
done
with
my
stuff.
Yeah
that'll
be
great
here.
G
Okay,
so
marion,
do
you
need
me
and
michael
to
help
write
the
descriptions
once
we
provide
the
framework
or.
F
And
I
think,
if
you
reply,
then
people
can
jump
in
the
real
goal
would
be
after
that's
done
and
you
reply
and
everybody
sees
it
being
assigned
to
some
area
right
here,
so
that
we
can
move
right.
G
And
what
I
really
do
need
is
everybody's
github
username,
because
that's
how
I
can
assign
things
so
because
we
are
in
github-
and
I
know
quite
a
few
of
you,
but
if
you
could
just
throw
it
in
the
chat
window,
I
could
copy
them
down.
C
And
one
more
request
from
in
right.
So
can
you
please
keep
us
in
loop
myself
in
vendetta
and
part
of
the
the
pna
model
discussion.
C
Because,
initially,
the
ask
was
to
take
a
look
at
connection
tracking
right
if
this
is
going
to
be
based
on
that's
what
we
are
looking
at
currently,
if
that
is
what
it
is,
so
we
want
to
be
because,
if
you
want
to,
if
we
are
going
to
pick
up
the
connection
tracking
from
the
land,
that's
what
I
just
want
to
be
in
the
discussion
for
this
one.
Oh
okay,.
G
C
H
G
C
C
G
C
Else
can
we
yeah
sorry
and
then
I
just
think
I'm
just
still
on
the
last
topic
right.
So
do
you
want
us
to
initiate,
because
mario
is
going
to
initiate
a
meeting
for
this
to
discuss
this
and
couple
of
days
once
we
review
this
open
conference.
G
I
can
let
you
guys
do
it
if
you
want
to
or
mario
do
you
want
to
initiate
the
meeting
or
how
do
you
want
to
do.
B
That
yeah,
it's
fine,
yeah
yeah.
I
would
I
just
don't
know
who
is
interested
in
participating
of
course
venkat
and
marion
and
myself.
But
if
anyone
else
is
interested
drop
me
an
email
and
I
will
include
in
the
invite,
does.
C
C
G
Yeah
all
right
I'll,
give
you
three
business
days:
okay,
we're
not
working
weekends,
okay,
anything
else
guys
for
today,
then,
and
so
michael
and
I
will
map
this
over
to
the
original
repo
and
work
with
you
know,
send
a
link
to
marion
marion
will
populate
what
he
needs
to
and
then
I'll
send
you
all
I'll
reply
to
this
meeting
thread
with
the
information
is
that
okay.