►
From YouTube: DASH Test Workgroup Community Meeting 20211117
Description
First DASH Community meeting November 17, 2021
A
You
know
some
of
the
side,
headers
and
attributes
and
put
them
in
here,
mainly
because,
as
you
guys
said,
that
right,
so
if
we,
if
we
need
the
folks
to
implement
all
the
side,
attributes
that's
too
much
and
most
of
that
not
really
useful
right
and
did
not
really
be
used
in
this
project.
So
therefore
we
took
a
subset.
We
analyzed,
you
know
what
other
needed,
attributes
and
objects
we
need
for.
The
underlay
stops
that
we
can
reuse
them.
A
So
therefore
you
know,
obviously
you
know
the
underlay
header
is
much
smaller
than
the
than
what's
in
the
side
side,
repo
right.
So
those
are
the
subset
of
this
attribute
and
object.
We
think
we're
going
to
use
in
this
in
this
stmpli.
B
A
C
A
B
A
We
don't
we
don't.
We
don't
need
sonic
at
all
for
for
this
testing
purpose
right.
So
the
reason
is
that
I
think,
if
you
look
at
the
new
type
ci
test
framework
right,
so
there
are
two
parts.
One
part
is
that
you
need
to
have
a
automatic
generating
to
test
the
framework.
A
You
know
the
size
through
the
server
which
you
know.
I
think
you're
gonna
be
discussing
on
that
to
how
to
adapt
that
to
the
dash
headers,
but
that's
that
that
server
also
is
automatic
generator
for
the
existing
site.
Headers
and
then
the
other
parts
of
the
of
that
new
framework
is
very
summer-based.
A
Setup
allowed
you
to
initialize
the
size
and
initialize
the
ports.
So
basically
you
get
all
those
ports
available
and
you
can
just
use
the
size
swift
apis
to
to
initialize
all
those
ports
and
and
and
set
up
some
basic
stuff,
and
then
you
know
basically
build
that
test
environment.
A
B
A
I
think
that
is
so,
I
think
they're.
So
in
some
you
know
very
specific
case.
There
may
be
some
interaction
between
the
underlay
and
overlay,
but
for
most
of
the
cases
right.
So
you
know
if
we
focus
in
some
specific
scenarios,
I
think
once
you
set
up
those
underlay,
then
you
don't
really
need
cheat.
You
don't
really
need
to
change
the
underlay
at
all
right
so,
but
I
do
believe
that
you
know
you
know
in
this.
You
know
in
order
you
know
to
test
the
full
esteem
scenario.
A
You
not
only
this
site
test
framework
is
not
only
test
the
underlay
but
also,
oh
sorry,
not
testing
the
overlay,
but
also
test
the
underlay.
For
example.
You
know
we
need
to
make
sure
the
bgp
packet
is
being
trapped
right.
So
we
set
up
the
trap.
We
know
so
that's
part
of
the
sdn
appliance
functionalities
right.
So
you
know
as
part
of
the
test.
A
You
also
need
to
test
the
underlay
as
well,
but
I
think
maybe
some
of
the
underlay
test
has
been
covered
in
the
existing,
the
the
size
with
the
test
framework,
so
that
you
know
without
the
framework
they're
about.
If
I
remember
correctly
about
maybe
a
thousand
test
cases
already,
you
know
a
subset
of
those
tests
can
be.
A
C
D
Hey:
hey
gohan.
This
is
a
question
for
you
yeah.
I
completely
understand
what
you're
saying
I
think
it
makes
sense
to
have
you
know
the
changes
that
are
less
or
the
aspects
that
are
less
likely
to
change,
be
in
the
underlay
and
you
know
probably
more
likely
to
change
in
the
overlay.
But
my
question
is:
do
you
anticipate
the
overlay
to
change?
Often
I
mean
I
mean
the
p4
specification
that
we
are
using
here
for
the
pipeline.
Is
that
going
to
change
often.
A
That's
a
you
know,
that's
a
very
good
question.
I
think
the
very
reason
we're
using
the
p4
is
mainly
because
I
think
for
us
right.
So
we
we
really,
you
know
want
to.
A
You
know,
use
a
side
header
for
this
project,
but
traditionally
you
know
in
when
we
developed
the
switch
aside
for
the
switch
we
didn't
really
develop.
You
know
a
a
rigorous
is
people
program
and
I
think
part
of
the
reason
is
because
I
think
the
switch
behavior
is
more
well
understood
and
the
people
have
a
kind
of
a
common
agreement
on
what
the
switch
pipeline
looks
like.
A
But,
however,
in
p4
we
feel,
like
you
know
this,
this
customer
pipeline
is
is,
is
not
universally
known
and
we
do
need
the
you
know:
pri
precise
language,
to
describe
how
what
is
the
packet
behavior.
Therefore,
we
we
use
p4,
I
think
you
know,
as
as
the
project
grows
and
as
we
have
more
and
more
example,
sorry,
more
and
more
use
cases
as
azure
grows.
A
I'm
sure
you
know
we'll
add
more
and
more
functionalities
into
this
into
this
p4
program
and
therefore
there
may
be
a
question
you
know
if
we
start
changing,
adding
new
functionality
into
this
p4,
then
how
do
we
kind
of
keep
the
side
headers
evolve,
while
not
breaking
into
the
opera
applications
might
be?
Maybe
you
know
some?
How
do
we,
you
know,
make
sure
when
we're
adding
those
applications?
Sorry
adding
those
new
new
functions
while
still
keeping
backer
compatible?
You
know
making
sure
the
the
the
application
is
happy
right.
A
So
there
may
be
some
discussion
around
that,
but
I
think
the
the
main
purpose
for
introducing
the
p4
is
to
describe
the
behavior
precisely
and
so
that
everybody
know
what
exactly
the
expectation
of
this
pipeline.
D
Okay,
okay,
I
understood
and
just
to
add
on
to
that
you
know
I
think
you
know
in
the
switch
world.
You
know.
P4
has
a
lot
of
significance
because
you
know
it's
mostly
stateless,
but
when
it
comes
to
the
end
points
where
there
are
more
stateful
processing
involved,
maybe
there
are
some
new
options
in
the
before
that.
You
know
I
haven't
looked
at
the
latest
parts
available
in
there,
but
is
that
look
through?
If
do
we
have
all
the
ways
to
specify
stateful
looking.
A
Yes,
yes,
indeed,
I
think,
as
part
of
the
I
think
one
of
the
key
use
cases
here
is
to
do
the
tcp
connection
tracking
and
I
think,
in
the
current
model
we
do
introduce
some
stateful
concept
in
the
in
the
p4
language
to
to
do
that,
tcp
connection
tracking,
so
it
is
defined
in
the
spec.
I
don't
think
it's
being.
A
How
do
I
say
that,
like
you
know
it's
it's
it's
our
version
of
that
how
to
describe
that
state
for
concept
in
the
p4,
so.
C
It's
not
a
standard
p4
model.
Can
I
comment
on
that.
I
noticed
reading
the
code
that
in
the
contracting.p4
file
there's
some
non-p4
constructs
in
there
they're
new
inventions-
and
I
happened
to
run
into
marion
at
the
ocp,
show
and
asked
about
that,
and
he
said
yes,
but
that's
true,
but
we
didn't
get
a
chance
to
finish
the
conversation,
so
I
would
take
it
a
conceptual
proposal,
type
contracts.
E
Quick
sentence
there
just
a
quick
line:
there
is
a
portable
nick
architecture.
You
know
that
is
also
a
sort
of
consortium,
part
of
the
p4
line,
and
there
the
host
side
of
things
are
being
defined,
much
more
in
detail
and
being
added
to
before
the
externs
that
define
the
directionality
and
those
kind
of
things
that
are
not
switch,
specific
and
mostly
host
specific
are
being
defined
there.
Intel
is
also
involved
there,
together
with
nvidia
and
marion,
as
you
mentioned
chris
great.
B
G
I
I
have
a.
I
have
an
example
of
something
that
that
I'm,
you
know,
I'm
just
curious
about
the
the
the
20
did
to
make
the
savannah's
point,
the
20
million
connections
per
second,
you
know
10
in
10
out.
D
Yeah
yeah
you're
right
absolutely
right.
I
think
our
point
was
it's
a
dash
from
the
definition
of
dash.
It
applies
to
both
the
clients
and
other
endpoints
right.
Obviously,
the
number
that
you're
talking
about
is
more
applicable
for
or
the
current
definition
is
more
applicable
applicable
for
appliance
science,
as
I
understand
it,
but
once
if
there
is
more
definitions
coming
up
for
other
endpoints,
would
it
be
the
same
forum
to
discuss
and
adopt
it
or
something
else.
G
No,
I
I'm
just
I'm
just
suggesting
that
it
may
be
the
same
form,
but
we
need
a
focus
on
the
smart
on
a
smart
neck
or
a
nick
or
whatever.
We
want
to
call
a
dp
whatever
we
need
to
focus
on
that,
so
we
need.
You
know
a
version
of
that.
You
know
packet
transforms
document
that
focuses
on
that,
for
example,.
C
Yeah,
I
want
to
pause
for
a
minute
and
just
christina,
maybe
take
if
you're
taking
any
notes.
There's
three
topics
here:
we've
covered
in
rapid
succession
that
indicate,
let's
say
a
gap
in
the
documentation
and
before
we
get
further
with
five
more
things
that
we
glossed
over,
maybe
we
ought
to
document
better
the
the
relationship
of
the
psi
header
files
and
the
underlying
overlay,
because,
if
we're
asking
these
questions,
this
media
means
we
need
to
somehow
improve
our
documentation.
C
So
that's
like
a
to
do
for
somebody
who's,
a
good
writer,
and
then
we
should
probably
talk
about
the
p4
program
and
how
there's
some,
let's
say
few
forward.
Looking
things
in
the
current
p4
model,
you
can't
actually
compile
connection
tracking.p4
as
it
is
right
if
it
says
v1
model,
but
there's
statements
in
the
rp4
v1
model,
so
it's
kind
of
an
aspirational
program.
C
I
think
there
was
one
other
thing
that
we
went
through
real
quickly
probably
ought
to
backfill
some
of
the
documentation
as
time
allows
because.
H
C
Maybe
have
a
a
different
picture
that
sort
of
zooms
in
on
that,
as
I
learn
more
from
intel
this
afternoon
and
other
I'll
try
to
document
that
better.
I
want
to
show
you
one
more
diagram.
C
F
That's
my
yeah,
but
yeah.
That's
that's!
Fine!
It's
the
machine
generated,
but
but
the
question
I'm
asking
is
today:
the
underlay
apis
go
through
a
site
community
to
you
know
to
to
to
to
become
basically
as
a
standard
psy
apis
that
everybody
basically
has
to
write
their.
You
know
psi
code
for,
and
in
this
particular
case,
is
there
going
to
be
an
effort
to
standardize
the
psi
header
apis
or
psi
apis
that
are
generated
right
out
of
the
p4
program.
A
Maybe
maybe
let
me
tell
me,
I
think
you
know
there
seems
a
you
know,
lots
of
interest
on
this.
You
know
the
people
program.
I
think
what
what
I
think
you
know
the
we
like
the
community,
you
know
interest
and
discuss.
Is
you
know
we
looted
this
people
pro?
Maybe
we
should.
You
know,
work
as
a
community
to
to
standardize
this
people
program
right.
So
you
know,
I
think,
intel.
A
All
the
time
for
for
all
of
the
people
program
right,
so
no
no
for
the
dash
right.
So
the
the
you
know
the
the
on
the
overlay
stuff
right.
So
this
people
program,
we
we
as
a
community
need
to
you,
know,
discuss
and
agree.
So
we
put
this
draft
version
out
right,
so
we're
seeking
feedback
from
the
communities
and
if
people
all
agree
with
this,
you
know
the
this
p4
program.
A
I
think,
then
the
header
can
be
can
be
automatically
generated
so
the
either
you
know
conform
to
the
site,
spec
in
terms
of
those
metadata
and
all
those
things,
people
don't
have
to
worry
about
it
right.
So
we
can
focus
on
what
is
this
pipeline
and
how
it
should
be
look
like
and
then
let's
the
program
to
generate
the
header.
B
A
Yeah
yeah,
I
think
the
control
will
be
uploaded
and
you
know
people
can
also
look
at
the
tools
to
see
how
these
headers
are
generated
right.
So
I
think
my
point
is
that
you
know
previously
we're
focusing
on
the
header
files
and
explain
what
the
header
file
means
now.
I
think,
because
of
this
complexity
of
this
overlay
pipeline,
our
our
intention
is
that
you
know
let's
focus
on
the
p4
program
and
standardize
the
the
tools,
so
we
can
automatic
generate
headers
that
can
be
used
by
the
by
the
by
the
application.
B
J
E
Yeah,
I
think,
as
christina
as
you
mentioned,
it
was
going
to
be
a
preliminary
conversation
to
see
what
it
is
that
our
expectation
is
with
sciptf,
and
you
know
to
extend
it
with
our
engineering
team
within
intel
as
well
and
the
expectations,
and
you
know
to
jot
down
all
the
requirements
as
well
as
what
it
takes
for
us
to
get
there
if
needed.
E
I
was
actually
thinking
that
we
can
have
this
short
conversation
first
and
then
then
invite
everybody
with
the
more
with
the
information
that
we
have
and
and
then
discuss
more
deeper
conversations.
That's.
J
B
C
I
I
actually
have
a
doctor's
of
my
mouth
to
take
off
by
10
anyway.
I
wanted
to
show
one
more
diagram,
because
if
this
isn't
enough
complexity,
I
wanted
to
show
this
as
a
different
view
of
showing.
C
This
is
a
concept
I
wanted
to
propose
to
people
and
think
about
is
how
to
test
the
dash
size
apis
progressively
up
through
the
stack
layers.
So
this
is
using
cythrift
where
you're
just
testing
the
pure
data
plane
and
down
here.
This
is
the
long
game.
The
long
term
goal
is
tested
through
the
sonic
stack
with
gnmi
right,
and
presumably
you
should
be
able
to
test
these
with
one
abstract
set
of
test
cases.
It's
just
a
different
interface
right
to
the
data
plane.
C
C
It's
still
side-based
data
model
for
the
for
the
configuration,
but
it's
going
through
into
the
asic
db,
and
this
is
a
this-
is
a
way
for
developers
to
test
it
slightly
more
complicated
than
sci
thrift,
but
not
as
complicated
as
all
of
sonic.
C
This
is
pl
visions
tool
that
they
presented
at
ocp,
both
at
the
global
summit,
and
they
presented
it
at
a
special
ad
hoc
network
group
meeting,
and
it
looks
pretty
cool
to
me
and
I
just
thought:
I'd
bring
it
to
people's
attention,
because
they've
created
a
tool
that
they're
proposing
to
donate
to
ocp,
and
the
idea
is,
if
you
had
one
set
of
test
cases,
you
should
be
able
to
transform
those,
and
these
are
abstract,
they're,
not
written
against
any
particular
northbound
or
or
southbound.
C
You
should
be
able
to
transform
those
into
different
configuration
apis
and
scripts,
so
you
can
actually
work
your
way
down
the
stack
and
it
should
be
even
theoretically
possible
to
run
right
test
set
test
it
through
people
run
time
in
the
behavior
model.
So
I
just
want
people
to
think
about
this.
You
know
it's
just
it's
kind
of
an
academic
proposal,
but
I
think
it
could
be
practical
too,
if
that's
interesting
to
people-
and
this
is
if
we
ought
to
look
at
this
as
as
nirvana
for
testing
a
data
plan.
A
I
think
what
kind
of
you
know
the
information
that
is
controlled
through
the
site
right,
so
I
you
know
we
did
some,
because
I
think
the
you
know.
One
thing
I
think
I
need
to
I
want
to
point
out
is
that
you
know
the
most
critical
state
changes
or
you
know,
performance
driven
state
changes
like
the
connection
tracking
those
things
are
not
explicitly
exposed
to
a
side.
Therefore,
those
information
is
not
going
to
be
the
radius,
the
the
one
that
in
the
radius.
A
I
think,
for
example,
in
the
v-net
to
win
it
traffic
pattern.
What
is
going
to
be
in
the
radius
is
going
to
be
all
those
mappings
right
so
say
to
pmp.
A
You
know
those
vin
and
mappings,
so
you
know
those
things
that
are
not-
or
maybe
I
I
need
to
you
know,
help
to
ask
you
to
clarify
when
you
say
you
know
the
scalability
issue
of
the
radish.
Do
you
mean
you
know
the
number
of
objects
stolen,
the
radius
or
the
you
know
the
rate
of
the
changes?
B
D
A
D
A
That
I
can
explain
right
so
you
know
no,
I
think,
the
the
site
header.
If
you
look
at
it's,
it's
a
it's
a
control.
He
says
that
you
know.
Basically,
the
site
itself
is
going
to
maintain
the
connection
tracking.
It's
not
the
above
the
side
application.
It
doesn't
need
to
maintain
that
connection
tracking.
Therefore,
you
know
those
connection.
Tracking
information
is
not
going
to
be
in
the
in
the
radius.
B
B
B
You
know
I
mean
I'm
not
saying
that
it's
impossible,
but
I
think
we
need
to
be
conscious
when
we
decide
what
goes
into
redis
and
what
doesn't
go
that
you
know
some
of
this
card
initially
at
least
may
have
limited
memory,
space
and
limited
processor
capability.
C
A
I
agree,
I
think
you
know
so
that
that's
why
I
think
we
explicitly
cut
out
that
you
know
those
connection
tracking
is
not
in
the
radius
right.
So
therefore,
watching
the
radius
is
those
static
mappings,
as
I
said
right,
so
you
know
like
say
to
pm
mappings
and
also
those
those
sales
right,
so
that
we,
that
user
has
been
comfortable.
Those
things.
C
Maybe
maybe
someone
ought
to
do
a
quick
proof
of
concept
on
you
know
loading
up
redis
with
this
many
items,
you
know
it's
a
string
based
key
value
store,
there's
a
lot
of
inefficient
stuff
going
on
in
such
a
you
know,
hashing
and
lookups.
It's
when
you
scale
to
certain
size.
I
don't
know
how
it
performs.
So
maybe
someone
ought
to
do
a
quick
sanity
check
on
that.
I
don't
know,
but
that's
kind
of
outside
of
the
test
working
group,
except
if
you
can't
test
it,
you
know
it
doesn't
work.
C
You
could
even
just
run
redis
on
a
pc
and
and
say:
okay.
How
long
does
it
take
to
populate
this
many
entries
on
a
fast
x86?
Well,
I
was
going
to
work
on
a
slower,
smaller
arm.
You
know
that
those
are
good
kind
of
reality.
Checks
maybe
ought
to
be
done.
That's
what
I'm
that's
what
I'm
hearing
anyone
have
any
opinions
on
that.
B
C
D
D
J
Probably
gonna
be
different
and
everything's
kind
of
under
development
as
we're
going
along.
So
you
know
it
definitely
a
underdevelopment
project
so,
like
you,
said,
take
what
we
have
and
take
in
the
feedback
and
move
forward.
You
know
with
the
smartest
guys
you
know
that
we
have
in
the
room
here.
I
think
it's
awesome.
B
J
B
J
J
Thank
you
great,
and
we
only
have
yes
11
minutes
left
and
you
know
I
do
feel
you're
right.
I
think
we
do
have
more.
J
B
The
way
I
have
a
final
question
on
this,
which
is
okay,
very
overlay,
very
big
question
and
go
on,
may
think
to
it
you
know,
or
whoever
will
think
to
it.
You
know
the
switch
in
p4
has
what
is
called
the
psi.
The
portable
switch
architecture
you
know,
and
as
intel
mentioned
in
the
nick
there
will
be
a
portable
nic
architecture.
The
pna
you
know
the
psn
the
pna
now,
what
is
not
clear
to
me
is
if
this
before
model
need
to
comply
with
the
psa
and
the
pna.
B
A
A
I
think
maybe
you
know
we
can
you
know,
I
think
the
intention
is,
you
know
to
to
seek
for
feedback
for
this
draft
right.
So
therefore,
I
think
we
should
probably
have
a
meeting
and
invite
you
know
the
the
author
of
this.
You
know
this
this
pipeline
to
and
to
make
some
introductions
and
we
have
community
discussions
right.
So
yeah.
I
wonder
you
know
christina.
If
you
can,
you
know
help
to
you
know:
contact
vedia,
maria
and
see
you
know.
A
I
don't
know
when
the
time
maybe
or
leverage
you
know
the
next
meeting
or
whatever
so
because
I
think
we
have
the
the
the
right
people
in
this
meeting
for
all
those
discussions.
Right
so
looks
like
we
have
a
lot
of.
J
A
No,
I
think
there
may
be
a
lot
of
various
questions.
I
think
the
the
the
the
first
step
is
to
invite
the
officer
to
make
an
introduction
and,
and
then
people
can
understand
a
little
bit
and
then
you
know,
discuss
all
those
questions
right.
I
think
those
are
just
some
examples.
I
think
I'm
sure
there
are
a
lot
of.
A
C
J
J
Yeah,
okay,
okey-dokey
I'll,
see
what
I
can
do
sure
sure
sure,
and
also
I
want
you
guys
to
know
I
I
did
hire
or
I
put
in
an
offer
for
a
technical
writer,
and
so
now
I
will
have
someone
who
can
help
with
the
documentation
and
start
converting
everything
I
have
to
mark
down
and
start
posting
it
to
the
repo
he
won't
start,
probably
for
four
weeks
or
two
weeks.
I'm
sorry
two
weeks,
but
you
know
there
will
be
more
clarity
coming
in
the
documentation.
You
know
these.
J
E
Definitely
it
would
be
good
to
invite
marianne
next
time.
Agree
that
you
know,
2
still
is,
you
know,
works
with
the
v1
model,
which
is
older
version,
and
we
need.
We
are
actually
there's
a
lot
of
thought
on
how
to
change
that,
using
psa
in
the
host
side
using
p40
pdk.
That's
just
sharing
some
internal
information.
E
J
Yeah
thanks
chris
thank
you,
restaurant
and,
like
I
said
I
think
I'm
talking
to
him
in
five
minutes
here
and
then
you
know
the
chris.
We
didn't
get
through
your
whole
deck
here
so
anyway,.
C
C
The
previous
one
is
I'm
going
to
upload
this.
Do
a
pull
request
to
put
this
in
the
slides,
folder.
C
I'll
do
that
this
afternoon,
when
I
get
back,
I
have
to
have
an
appointment,
but
this
should
be
up
there
later
today,
if
christina
approves
the
full
request
I'll.
J
J
J
Yeah
I'll
take
some
notes.
It's
gonna
take
me
a
couple
hours
to
send
them
out,
but
I'll
do
that
and
send
them
out,
and
I
appreciate
you
all
appreciate
your
time
and
I
will
try
and
get
marion
to
come
talk.