►
From YouTube: Behavioral model WG Weekly 20220314 (March 14, 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
Probably
time
them
out
up
to
some
point,
but
the
point
is
you
can't
even
set
it
on
the
first
pin
because
the
you
don't
know
when
the
second
pin
is
going
to
come
along.
So
that's
one
thing
or
you
can
close
the
one
direction.
That's
an
option.
You
could
close
that,
but
you
would
still
have
the
other
one
open
and
you'd
have
to
close
it
completely,
and
you
know
the
only
reason
you
do
that
is
to
save
space.
A
A
Tcp
to
force
something
to
close
period
is
to
send
a
reset,
and
then
there,
if
you
do,
then
the
both
directions
have
to
be
shut
down
immediately.
But
but,
if
you're
sending
a
fin,
you
have
to
wait
for
the
other,
the
application
to
end
its
it's.
What
it's
sending
and
then
send
its
own
pin
and
then
you
could
set
a
timer
waiting
for
the
ack
to
come
back.
A
So
so
it's
more
complicated
than
just
saying
set
a
timer.
You
can't
set
a
timer
on
the
first
in
unless
it
was
a
really
long
timer,
because
you
just
don't
know.
D
D
A
But
the
problem
with
that
is
the
first
connection
could
be
closed
completely
if
you
did
it
based
on
the
fin
and
the
neck
on
the
first
in
the
one
direction,
the
the
direction
that
initiates
the
pin,
you
can
close
that
half
down
and
you
save
space
and
memory,
and
then
you
can.
A
When,
when
it's
appropriate,
but
the
only
way
to
do
that
is
to
do
that,
based
on
you
could
do
it
based
on
tracking
the
the
finance
sequence,
you
could
set
a
timer
there,
but
it
would
have
to
be
separate
from
any
timer
that
is
based
on
closing
the
opposite
direction.
D
Right
so
gerald
you
know
to
be,
to
be
honest,
I
haven't
really
read
through
the
whole
thread.
I
will
read
carefully
and
go
through
it.
I
just
had
a
snippet
from
richmond.
I
looked
at
it
one
one
more.
A
If
we
write
the
behavior
model
and
it
closes
it
based
on
the
fin
and
storing
the
sequence
and
it
closes
it
based
on
that,
that
is
the
correct
behavior,
whether
you
implement
your
closing,
based
on
how
your
hardware
is
optimized
based
on
a
timer.
I
think
it's
good
discussions,
but
remember
the
behavioral
model
is
not
an
implementation.
It
is
what
drives
the
test
cases
right,
but
it's
not
the
implementation.
You
might
say
these.
A
My
hardware
just
doesn't
support
that,
I'm
going
to
use
timers
and
when
we
test
it,
if
it
tests
properly-
and
you
know-
closes
connections
properly,
we
don't
care,
but
when
we
model
it,
I
don't
think
we
should
model
it
as
as
us.
You
know
necessarily
as
a
set
of
timers
the
way
it's
being
talked
about,
because
it's
too
close
to
your
implementation,
it
should
be.
We
know
how
to
close
the
one
direction
down.
We
know
how
to
do
it
with
sequence
number.
We
know
it
will
test
properly.
That
way
we
add
timers.
A
A
D
Right
right,
general,
one
part
that
I
didn't
understand
so
as
far
as
what
we
were
discussing
there's
only
one
flow
rule
that
is
used
in
both
directions,
so
we
didn't
really
have
a
concept
of
closed
from
one
direction
and
open
from
another
direction.
D
Even
when
you
do
asymmetric,
you
know
firewalls,
it's
the
same.
A
D
The
handshake
is
two
direction,
but
the
flow
rule
and
and
and
you
know,
the
timers
and
whatever
that
we
are
doing
they're
all
a
single
per
flow.
A
So
not
as
yes
and
no
yes
for
the
most
part,
but
think
about
it,
though
a
rule
is
setting
also
is
followed
by
what
you
need
to
encap
with,
if
I'm
or
decap,
with
two
different
actions,
two
different
things
so
like
there
are.
There
are
variances
between
the
the
receiving
the
transmit
the
rules
on
timers.
You
have
to
have
to
be
the
same,
but
there
there
are
differences
in
the
rules,
at
least
in
the
in
caps
and
and
how
you
do
that.
D
But
that
is
the
that
is
separated
out
right,
general.
We
are
talking
about
complete
table
of
load
and
not
something
which
is
a
flattened
cash.
So
if
it
is
a
flattened
cache,
yes,
you
would
have
different
rules,
but
if
you're
not
talking
about
a
flattened
cache
here,
the
flow
cache
is
not
does
not
have
details
on
your
end
cap
dcap,
because
that's
in
separate
tables.
C
So
definitely
agree
that
you
know
from
the
application
point
of
view,
it
is
required
for
the
application
to
close
the
connection
and
the
direction
comes
into
picture
for
sure,
and
for
that
reason
we
have
to
keep
it
as
generic
as
possible
from
the
behavioral
point
model
point
of
view,
and
as
long
as
the
timers
are
settable,
the
application
will
be
able
to
set
it
according
to
what
it
may
need.
No.
D
This
is
not
the
application
setting
it
reshma,
like
you
know,
that's
the
thing
right.
This
flow
cache
like
even
in
our
device.
You
know
how
long
you
set
this
absolute
timer
or
replenishing
time.
It
is
a
policy
on
the
infrastructure
side
as
to
like
how
long
they
want
to
allow
any
connection
to
be
alive
and.
C
Yeah
yeah,
these
timers
are
settable
right,
anjali
per
eni.
A
But
we're
we're
saying
in
the
behavior
model
they
are
acceptable
for
eni
because
we
do
have
different
vms
from
different
mbas
that
will
have
different
tcp
requirements,
so
they
have
to
be
settled
so
again.
That's
an
implementation
on
your
side.
D
D
The
requirement
for
whatever
we
tell
for
the
next
requirements.
A
A
A
Typically
we're
going
to
try
to
force
them
to
one
because
it
consumes
our
resources,
but
it
is
definitely
associated
with
the
nick
interfaces
of
the
vm,
okay,
so
yeah.
So
it
is
a
sub
component.
Officially.
E
Sounds
good
marian
and
had
you
had
a
chance
to
read
through
this.
B
Yeah,
but
I
have
a
question
to
the
conversation
in
general:
I
don't.
I
still
don't
understand
why
we
are
mixing
timers
with
this
handling.
We
assume
that
we
receive
every
packet
of
the
flow.
How
timers
are
related
to
this.
F
I
mean,
I
think
the
timers
are
there
because
you
know
like
endpoints
can
go
away,
the
power
can
go
out,
they
can
crash.
Things
can
happen
right.
Oh.
B
B
A
A
B
A
Well,
that's
that's
what
we're
talking
about
we're
talking
about
how
to
close
a
connection
so
john
sent
an
email
itemizing,
the
different
things
that
can
happen
and
why
you
need
to
close
in
a
particular
way,
and
this
stuff
is
documented
out
there
I
mean
I
was
reading
the
same
book
as
he
was
reading.
It's
all
old
stuff
right.
So
what
we're
trying
not
to
do
is
to
build
a
behavioral
model
that
doesn't
reflect
the
air
estates
or
not
even
error.
You
can
get
out
of
order
packets
and
not
only
that
you
can
get.
A
You
know
it's
not
a
matter
of
you
know
some
of
the
some
of
the
notions
of
even
how
to
close
are
wrong
because,
for
example,
in
the
coming
book
it
talks
about
you
get
the
fin.
You
act
it
right
away.
It's
not
a
thing!
You
you
act
that
fin
right
away.
A
It
comes
back
as
an
act
then,
when
the
application
closes,
it
will
send
the
pin
hack
after
the
fact,
and
it
could
receive
multiple
acts
for
previous
frames
in
the
meantime,
even
before
it
sends
out
the
pin
act
so
and
it's
even
possible
if
it
was
fast
enough
that
get
an
act
after
the
thing
comes
up,
so
all
of
those
cases
actually
exist
in
real
life.
So
what
john
was
saying?
Itemizing
is
there's
a
way
to
cover
these
cases
and
where
all
of
these
things
will
happen.
A
And
then
just
give
one
second,
and
then
he
did
bring
up
an
email.
Just
like
I
read
too
the
fin
act
comes
later.
You
know
it
could
come
very
quickly,
but
it
may
come
after
they
finish
sending
their
packets,
which
means
later
so.
In
the
meantime,
the
reverse
direction
is
still
valid
and
flowing
until
it's
till
that
time
and
which
means
that
those
things
are
dislocated
for
a
time.
You
can
close
the
one
direction,
but
you
can't
close
the
other
direction.
A
We
may
use
a
timeout
for
that
in
case
it
doesn't
close
properly,
or
we
also
know
that
we
can
at
any
time
receive
a
reset
when
either
side
thinks
when
either
side
thinks
anything
has
gone
wrong.
They
can
send
a
reset
and
then
then
that's
a
different
occasion.
Now
you
have
to
you,
have
to
shut
both
directions
down
because
you're
not
allowed
to
say
anything
anymore,
reset
buttons,
I'm
no
longer
in
sync.
I
can
no
longer
guarantee
that
I've
received
everything,
so
you
send
a
reset.
You
shut
it
down
immediately.
A
So
all
of
those
things
he
talked
about
those
things
in
his
email.
What
we're
saying
is
we
need
the
behavioral
model
to
do
to
have
the
code
to
deal
with
it
like
he's
talking
about.
It's
not
like
a
behavioral
model.
Isn't
a
high
level
good
path.
It's
the
it's
all
of
this
and
that's
what
we're
trying
to
get
to
and
that's
why
we've
been
recruiting
people
are
designers
lately
to
to
help
and
go
and
write
this
code.
The
way
it
you
know
accurately
down
to
the
level
that
we
need
to
test
at.
A
So
when
we
finish,
we
should
have
a
set
of
test
cases
that
test
every
single
one
of
these
scenarios,
that
is
in
his
email
and
more
and
they
should
pass
and
if
they
don't
pass,
then
we
don't
have
a
golden
image
for
a
behavioral
model,
because
it's
it's
not
really
that
useful
to
us.
It
leaked
the
past
because
other
people
want
to
use
that
behavioral
model
to
test
their
hard
work.
F
F
Like
all
the
different
states
you
can
be
in
and
all
the
different
ways
you
can
terminate
a
connection
and
like
there
are
even
other
ones
like,
for
example,
if
you
try
to
sin
to
it
to
an
end
point
that
doesn't
exist,
but
well
like
that
sort
of
partially
open
connection
will
sit
in
your
flow
table
until
you
time
it
out
right
and
so
there's
all
of
these
cases-
and
I
think,
like
one
of
the
things
that
I
was
suggesting
in
this
email,
was
that
the
open,
contrail
connection
tracking,
like
expresses
the
logic
for
all
of
these
different
cases.
F
It
doesn't
have
timers
in
it.
It's
just
the
states
that
you
can
go
through
and
like.
I
was
suggesting
that
we
should,
like
borrow
that
logic
as
the
as
the
logic
that's
in
the
behavioral
model
for
dash,
because
I
I
believe
that
that
it
covers
all
the
corner
cases.
A
F
A
We
have
a
starting
point,
that's
well
thought
out
and
then
timers
on
top
of
it.
We
can
debate
for
a
while
and
that's
okay
and
then
we
can.
We
can
add.
F
But
right-
and
I
suggested
what
I
suggested
my
thoughts
on
like
the
timers
like-
I
think
that
there
can
be
like
three
different
timers
one
for
like.
If
you
got
a
sin
with
no
ack
like
you
could
time
that
out
quickly
right
then,
when
you're
in
the
established
state.
That's
when
you
want
to
have
like
a
lengthy
timer
and
then
when
you
get
the
second
fin,
you
can
go
back
to
a
short
timer
because
you're
just
waiting
for
that
final
act.
F
So
like
I
I
I
think
that
like,
but
all
of
the
states
are
represented
in
that
open
contract
code
yeah
mm-hmm
here
yeah,
because
I
I
my
fear
is
that,
like
we
can,
we
can
talk
about
this
and
talk
about
this
and,
like
think
that
we
understand
what
all
the
corner
cases
are.
But
I
don't
think
you
really
know
what
all
the
corner
cases
are
until
you've
been
like
burned
by
them.
Sure
yeah.
F
A
D
I
have
a
couple
more
questions,
so
one
sorry
go
ahead.
Yeah
go
ahead.
Yeah
you
said
you
would
like
to
configure
these
timer
values
independently
per
eni.
D
How
many
different
domains
do
you
see
they
fall
into
because,
of
course,
you
know,
keep
having
time
values
per
eni
and
eni
numbers
increasing
day
by
day.
I
want
to
just
be
sure:
it's
like
you
know
a
dozen
or
something
that
you
would
look
at,
or
it
is
something
you're
looking
at
like.
A
I
think
we
need
to
look
at,
we
do
it
today.
We
do
have
vms
and
we
have
to
set
these
things
differently,
but
we
have
default
values
which
we
we
could
agree
on
with.
Is
the
right
approach
globally
and
if
you
don't
set
it
for
eni,
then
it
will
inherit
the
default
values.
And
then,
if
you
do
it's
because
we
know
that
there's
a
particular
implementation
out
there.
D
Right
that
that's
a
fine
approach,
I
just
want
to
you
know
kind
of
get
an
upper
bound
of
like
you
know.
If
there
are
1000,
you
know,
interfaces
and
vms
or
whatever.
A
He
can
go
there,
but
realistically
what
we've
been
trying
to
tell
people
is
we're
trying
to
tell
you
like
practical.
You
know
we
only
have
typically
32
e9's,
typically
on
a
cpu
today
moving
to,
but
that's
for
200
gigs.
So
let's
say
we
go
to
800
gig.
Well,
then
that
would
be
four
times
that
so
would
that
would
end
up
being
like
128..
A
Kind
of
urban
a
thousand
is
arbitrary.
We
don't
want
to
like
use
arbitrary
numbers,
because
the
32
isn't
arbitrary.
It
actually
comes
from
us
adding
up
the
band
like
every
vm
gets
a
bandwidth
and
it
connections
per
second,
and
when
we
mix
our
typical
mix
together-
and
it
so
happens
by
coincidence
that
the
typical
mix
of
bandwidth
and
connections
intersect
at
around
30.
It
was
around
30,
so
we're
saying,
32.
A
and
and
and
we
can't
get
more
enis
on
it
because
of
the
bandwidth.
They
expect
bandwidth
and
I
expect
connections
per
second,
so
so
32
for
a
200,
gig
128
for
an
800
gig.
If
you
just
double
it
and
say
well,
you
know
things
might
change
you're
gonna
double
over
subscribe.
A
A
D
D
Yeah-
and
I
I
just
want
to
very
quickly
double
confirm
right,
so
we
said
this
is
not
a
flattened
flow
cache
model,
so
in
that
case
your
tables
for
all
the
encap
decap.
Everything
else
is
separate
and
what
we're
looking
at
is
a
you
know
the
flow
rule
associated
with
the
innermost
packet.
You
know
the
connection,
in
which
case
it's
a
single
flow
rule,
for
you
know
a
connection.
A
A
It
goes
back
to
all
the
numbers
we
talked
about.
Where
you
have.
You
know
a
thousand
rules
per
lookup
up
to
five
lookups.
Oh,
you
know
that's
where
you
get
the
dimensions
of
that.
When
you
talk
about
the
forwarding
table,
we
have
a
document.
Is
that
published,
christina.
D
Yeah
general,
maybe
yeah,
I'm
not
talking
about
the
dimensions
for
the
forwarding
table.
I
just
want
to
make
sure
that,
because
it
came
up
whether
it's
two
rules
per
flow
one
in
each
direction
or
not
that's
what
I'm
double
checking
that
it
is
one
rule,
because
smaller
tables
are
independent.
A
We'd
need
michael
on,
I
wish
michael
would
attend
these
calls
christina.
I
don't
know
that's
to
be
true
in
one.