►
From YouTube: DASH Behavioral Model WG Dec 1 2022
Description
BMV2 implementation ignores and sends packet out on same interface. Test will still work with default route that is pointing to both interfaces. Gap is underlay MAC address that was pushed needs to be updated. If we can plumb in MAC address using existing SAI API.
We need P4 table updated just enough
Marian – agree MAC address portion is open
Created Issue 292 and assigned to James Lo (Accton)
Basic underlay routing bmv2 to handle MAC address updates/resolution #292
IPv6 & IPv4 scale again
A
Was
from
the
very
last
meeting
before
we
went
off
for
a
break
so
I'm
also
trying
to
figure
out
if
we
added
anything
into
the
backlog
for
the
behavioral
model
project,
so
I
believe
this
is
the
conversation
from
last
time
or
we
can
maybe
go
through
it.
A
little
bit
and
I'll
go
ahead
and
grab
the
behavioral
model
project
didn't
while
I'm
looking
for
it
did
anyone
have
a
PR
they
wanted
to
bring
up
and
cover.
A
B
Wanted
to
ask
if
we
added
an
item
to
move
to
structured
annotations
and
the
behavioral
model.
C
A
C
Have
it
right
here
this
should
be
a
linked
issue
and
I
I
added
all
that
text.
Like
a
week
ago,
I
noticed
that
Youssef
was
going
through
and
and
kind
of
refreshing.
This
and
pinging
people,
so
I
went
in
and
added,
but
do
you
want
help
on
this
Marianne?
Is
there
sort
of
a
formula
that
can
be
followed
or
is
it
pretty
straightforward.
B
B
Yeah
because
it's
as
you
mentioned
before,
it's
very
straightforward,
just
look
up
the
key
and
The
annotation
and
pass
it
to
the
dictionary
that
then
is
passed
to
to
the
templates
like
it.
Then
it's
pretty
much
zero
logic.
C
A
A
A
A
So
that's
great
okay,
and
so
these
are
so
James
Lowe
had
offered
to.
You,
know,
try
and
help
out
with
behavioral
model,
and
he
had
requested
that
we
put
an
extra
here
in
the
in
the
status
field.
He
had
asked
for.
You
know,
needs
test
under
test.
A
You
know
Fields
so
that
we
could
tell
what
was
going
on
and
then
you
know
pending
resource
allocations.
So
it
did
that
and
he
want
I
don't
know
if
James
is
on
the
call.
He
wants
to
be
helpful,
so
just
FYI
to
everyone
he's
from
Acton.
A
So,
okay
was
there
a
specific
PR
this
week.
Anyone
wanted
to
cover
or
talk
about
or
anything
you
need
help
with
Marian.
D
I
wanted
to
ask
if
the
line
item
that
we
wanted
to
add
in
the
cast
list.
Last
time,
which
is
you
know,
Mac
address
handling
under
underlay
routing
resolution
in
bmv2.
D
I
haven't
added
it,
but
guys
I
can
add
it.
If
you
can
please
yeah.
A
B
D
What
priority
would
it
be?
One
zero
is
the
highest
I'm
guessing
so
yeah.
A
B
It's
not
dependent
on
anything,
however.
I
cannot
say
that
I
will
be
able
to
do
that
this
month.
I
would
want
to,
but
probably
I
won't
be
able
to.
B
Yeah
yeah
I
will
likely
be
away
most
of
the
summer.
E
A
D
A
And
anything
else
we
needed
to
cover
from
the
last
meeting
and
default
routing,
GPU,
okay,.
A
Okay
and
then
deployment
well
I
saw
this
line.
Deployment
topology
from
Microsoft
would
be
helpful,
so
we
did
have
the
drawing
where
we
were
saying.
You
know,
for
example,
you
could
place
this.
You
know
pretty
much
at
any
tier
in
the
data
center.
Is
that
what
we
were
talking
about
here?
I
wasn't
in
this
particular
meeting
right.
D
I
think
the
the
group
wanted
to
get
some
more
detail
on
whether
the
packets
that
come
in
on
one
port
need
to
go
into
a
different
port
or
same
port
and
how
the
Gateway
is
connected.
Something
like
that
to
figure
out
the
simplest
underlay
routing.
A
D
Say
it
again,
yeah
yeah
sure
we
needed
to
add
some
basic
underlay
routing
in
bmv2,
and
the
question
came
up
that
whether
the
packets
that
come
in
on
one
port
in
a
dpu
needs
to
go
out
in
the
other
port
or
same
port
or
how
the
Gateway
is
connected.
Do
we
is
default
routing
sufficient,
so
or
you
know,
routing
to
specific
address
how
that
will
be
handled?
How
the
connection
is
overall,
topology
wise?
What
we
expect
I
think
that's
what
the.
F
I
think
James
could
answer
that
question
Christina.
We
can
take
that
down.
I'm,
not
sure
that
there's
a
you
know.
Sometimes
they'll
go
back
to
same
port
and
they've
got
different
designs.
Sometimes
they
go
back
the
same
port,
sometimes
to
go
the
other
part,
but
yeah.
D
G
B
F
Or
something
else
Well,
it
can't
randomly
go
out
for
sure
whatever
it
does
has
to
go
out
with
someone
I,
don't
know,
I
think
this
question
can
be
brought
to
James.
They
did
a
lot
of
work
on
this
and
that's
why
I
don't
want
a
concert
like
some
documents
on
this.
F
In
fact,
there's
a
document
that
many
of
you
have,
which
is
the
you
know,
the
architectural
document
that
shows
how
we
build
our
rack
and
that
actually
shows
you
power
comes
in
and
where
it
goes,
but
I
think
we
need
to
have
the
configurability
and
where
that
configurability
is
that's
a
good
question
so
we'd
have
to.
F
We
can
take
it
offline
with
James
and
get
the
right
people
to
get
there.
I,
don't
know
how
they
direct
the
traffic.
F
I
know
where
it
goes,
but
I
don't
know,
I,
don't
know
what
the
mechanism
to
force
it
that
way,
but
they
don't
do
it
randomly.
They
don't
like,
allow
it
to
come
in
one
point
in
some
profit
growth,
a
few
different
parts
like
it
all
goes
either
back
to
the
same
port
or
it
goes
out
the
other
point.
So
we
just
need
to
figure
out.
You
know
what
we
do
today
and
then
how
did
they
accomplish
directly
new
topic?
So
good
question.
F
F
F
A
A
A
Any
other
PRS
or
issues
to
to
give
me
an
update
on
here,
but
on
the
underlay
SCI
API
I
was
supposed
to
check
back
on
November
10th.
B
C
Yeah
I
just
did
some
sort
of
dummy
switch
Port
attributes
so
that
we
could
initialize
the
python
test
framework.
A
So
the
update
to
this
one,
then
Chris
is
still
missing
a
few
pieces.
C
A
And
then
Nixon
tracking
in
the
simulator
how's
that
gone.
B
Yeah,
so
I
actually
didn't
touch
that,
because
the
way
we
build
Dockers
was
changed.
But
now,
since
it's
done,
I
can
update
my
pull
request
to
have
this
Docker
build
based
on
checks
and
do
do
the
same
thing.
But
so
the
pull
request
is,
as
of
now
out
of
date,
no
need
to
update
it.
A
And
and
we're
still
this
one's
still
pending
is
this
still
pending
the
no
I'll
look
up
that
one
and
then
do.
We
still
need
this
translation,
the
P4
runtime
side,
simulator.
A
A
Well,
I
will
take
care
of
that,
we'll
close
that
one
okay,
we
did
all
that.
Okay,
great
anything,
you
guys
want
to
cover
or
talk
about.
Specifically
we,
you
know:
we've
got
Gerald,
we've
got
John,
we've
got
Everybody
Marine
reshma.
G
A
G
G
A
F
You
know
also,
one
thing
you
can
assume
is
the
underlay.
It's
V6,
okay
and
the
overlay
typically
is
V4.
But,
as
we've
said
in
the
past,
we
have
some
big
customers
that
actually
have
V6
networks,
big
retailers,
yeah
well,
no
well-known
retail,
they're
V6,
so
99
of
our
customers
are
be
poor,
but
we
have
some
really
big
customers
who
are
V6
and
that's
where
it
is
today.
What's
going
to
happen
tomorrow,
it's
hard
to
say,
but
that's
that's.
F
You
know
when
people
come
back
like
our
overlay
should
be
V6,
because
our
network
is
V6
capable,
and
you
know
everybody
has
problems
getting
ipv4
addresses
so
that
that
I
think
is
almost
a
given,
but
the
but
the
overlay.
Today,
yes,
99
of
customers
are
V4,
but
our
biggest
customers
are
some
of
our
biggest
customers
are
V6
so
that
that's
the
data
that
we
have
for
now
help
us
customers
move
to
V6.
So
it's
very.
D
Sorry,
you
mentioned
V6
in
overlay
right
or
underlay.
Please.
F
D
A
Internal
Azure
devops
and
you
know
we
we
do
know
V6
takes
more
memory
space,
but
are
there
going
to
be
as
many
entries
blah
blah
blah
and
then
before?
And
so
you
know,
of
course,
they
came
back
and
said:
well,
it'd
be
best
if
it
was
configurable,
so
I
am
still
working
with
them
to
get
you
guys.
The
best
answer,
yeah.
G
For
flow
scale,
yeah,
and
if
we
were
to
assume
like
that
all
64
million
flows
or
you
know.
F
G
To
be
able
to
be
V6,
then
that
is
like
a
very
different
amount
of
memory
than
if
they
all
have
to
be
before
I
mean
It.
Ultimately,
it
doesn't
matter
to
us
it's,
but
you
know
like
it's
it's
just
how
economical
can
we
be
on
the
memory
yeah.
G
You
know
there
are
some
I
mean
there
are
trade-offs
that
that
can
be
made,
but
yeah
so
like.
If
just
if
we
could
just
get
more
insight
into
like
that
and
like
I
mean
like
Gerald
made.
The
point
like
oh
99
of
the
customers
are
going
to
be
V4,
but
the
big
ones
might
be
B6.
Does
that
mean
that
the
entire
Dash
data
plane
could
potentially
be
consumed
by
like
a
big
customer
and
where
everything
has
to
be
V6
I.
F
G
F
G
Okay,
that
that's
that's
good
to
know,
and
it
would
be
good
to
know
like
more
specifically
like,
like
what
scale
requirement
for
V
like
separately.
What's
the
scale
requirement
for
V6
and
the
scale
requirement
for
V4
yeah.
A
F
Know
the
engineering
answer
is
the
same.
It.
F
F
I
think
that
what
would
be
good
to
know
is
that
we
kind
of
put
a
limit
on
the
memory,
because
it
costs
32,
gig
or
64
gig
would
be
nice
to
know
is
how
many
IPv6
could
fit
into
32,
gig
or
64
gig,
then
that
would
be
give
them
something
to
work
with
right,
as
opposed
to
the
other
way.
Around
I
mean
right
and.
G
F
G
Think
that
that's
going
to
vary
by
by
supplier,
it.
C
C
F
Well,
I
think
in
the
past,
even
if
you
go
back
through
history,
we
put
out
some
numbers
that
didn't
seem
to
be
perfect.
Let's
say,
and
after
working
with
with
suppliers
who
work
on
this,
we
come
up
with
new
numbers
which
seem
to
more
match
reality
and
I.
Think
that's
where
we
could
go
so
I
think
we
could
put
out
for
profiles.
I
think
that's
a
good
idea
and
then
just
ask
people
like
what
what
would
you
know.
F
We
wouldn't
be
able
to
share
back
the
individual
answers,
but
we
would
probably
come
to
some
conclusion
across
the
industry.
What
would
be
reasonable
and
I
think
that's
what
we
we
always
have
to
be
reasonable
right.
If
they're
not
doable,
then
it's
then
it's
there's
no
sense
asking
for
it.
So
maybe
that's
how
we
approach
it
because,
right
now
the
standard
answer
from
engineering
is
yeah
the
same
and
and
they
know
the
issue
right
so
yeah.
F
The
reason
is
they
don't
know
what
the
industry
can
do.
They
don't
know,
you
know
what
numbers
you'd
get,
and
so
they
don't
want
to
come
back
with
a
number
because
they
don't
even
have
a
clue
what
you
guys
will
come
back
with.
A
F
So
I
think
here
we
go
the
the
the
the
the
action
might
just
be
for
us
to
come
up
with
four
profiles,
not
telling
them
the
numbers
just
four
profiles,
and
then
we
can
ask
the
you
know
the
suppliers,
and
maybe
we
can
work
out
a
number,
because
I
think
this
is
going
to
be
circular
I,
don't
think
we
can
give
them
number.
They
don't
want
to
give
us
the
number
so.
E
F
A
F
B
G
Like
we
would
have
to
size
each
of
those
memory
pools
for
the
maximum
of
V4
and
for
the
maximum
V6.
That's
like
one
option.
Another
option
might
be
that
there's
just
one
shared
memory
pool,
but
but
when
you
use
V6,
you
don't
get
as
many
as
when
you
use
V4
and
so
like
we.
We
like
this,
is
kind
of
holding
us
back
from
like
in
a
way
figuring
out.
G
What's
like
the
right
implementation
of
managing
the
memory
and
like
if
we,
if
we
understood
better
like
what
the
requirement
is
like,
we
might
like,
another
option
might
just
be
hey.
If
all
flows
have
to
be
V6,
then
there's
just
one
pool
and
it's
just
size
so
that
all
flows
can
be
V6.
So
it's
it's.
This
is
like
we're
working
on
this
and
we
and
like
deferring
like
the
information
is
just
like.
D
F
I
wonder
if
you
know
when
we
talk
about
connection
tables
and
it's
I,
don't
know
your
implementations.
Whether
the
key
inside
the
connection
table
is
a
connection.
Id,
not
you
know,
separately,
keeping
the
actual
true
key.
It
may
make
May
mean
that
you
can
hold
as
many
there's
many
connections,
whether
they're,
ipv4
or
V6,
some
trick
to
do
that.
A
All
right,
my
dog's
having
a
meltdown
okay,
so
yeah,
so
I'll
keep
that
for
this
guys.
Thank
you,
John
for
asking.
E
Hey
Gerald,
could
we
go
back
to
the
underlay
just
for
a
second
I
believe
you
said
a
few
minutes
ago
that
the
underlay
was
all
V6.
E
So
so
Marion
so
I
think
the
reference
P4
code
today
is
only
supports
of
V4
underlay.
Is
that
correct.
F
B
B
I,
don't
think
there
is
a
need
for
that.
I
can
just
reply
or
flying
later
today,
but
if
it's
missing
then
yeah
we
should
we
should
support
it
as
well.
Especially
I
will
tell
you
why,
because
there
are
other
scenarios
that
utilize
under
Levy
6.
B
E
A
B
Okay,
thank
you
yeah.
If
it's
not
there,
we
will
add
it
to
together
with
other
scenarios,
as
I
said,
it
will
be
needed
at
some
point
for
sure.
F
F
C
My
only
items
were
from
yesterday
we
have
a
couple
of
PRS
waiting
awaiting
approval.
Okay,
I'm
hoping
Gohan
can
take
a
look
at
soon.
Okay,.
A
And
Gohan's
not
on
this
call,
so
I'll
ping
him
today
with
you
on
the
team's
Channel
or
something
thank
you.
D
I
approved
the
285
PR
from
Gohan.
Overall,
it
looked
good,
you
know
so
I'm
hoping
that
that
is
merged
285,
and
we
will
incorporate
that
change
that
he
made
in
25
into
277..
Oh.
C
A
Yeah,
okay,
then
Chris
I'll
ping,
you
and
Gohan
so
early
today.
Thank
you
in
a
different
chat.
Okay,
anything
else
from
the
rest,
you
guys.
A
Nope,
okay,
then
it's
10,
36
I'll,
stop
the
recording
and
I
can
give
a
little
bit
of
time
back
but,
like
I
said,
Yusuf
went
through
the
whole
list
of
items
looking
for
updates
and
we're
going
to
close
the
ones
that
have
the
appropriate
update
and
try
and
keep
the
list.
You
know
a
little
bit
lower
and,
like
I
said
James
Lowe
wants
to
work
on
things
too.
I'll
try
and
reach
out
to
him.
So
thanks
everybody
and
reach
out
to
me,
if
you
need
anything,
it's
Chrisney
microsoft.com
or
on
teams.