►
From YouTube: DASH HA Working Group June 21, 2022
Description
Please review/leave Comments for SAI HA API proposal by Marian - thank you!
https://github.com/opencomputeproject/SAI/pull/1500
A
Thanks
for
coming
to
this
week's
high
availability
meeting
and
this
week,
I
do
have
a
pr
submitted
by
marion
from
nvidia
to
the
open,
compute
psi
branch
experimental
here.
He
wanted
me
to
share
it
with
you
all,
I'm
not
sure
if
y'all
have
seen
it,
but
he
asked
me
to
share
it
here
with
the
community
and
and
maybe
look
it
over
before
we
meet
next
week.
So
it
looks
like
he's
defined
some
code
here
and
I
can
put
the
the
link
in
the
chat
window.
B
A
Okay,
so
I
guess
there's
there's
about
206
lines
to
look
over
if
this
is
something
you
guys
want
to
take
a
look
at.
C
So
one
quick
question,
I
believe
you
know.
Eventually,
our
goal
is
to
auto
generate
this
one
through
some
p4
behavior
model
right.
B
Well,
this
experimental,
even
currently
the
behavioral
model,
v2
project
that
marion
put
under
dash,
is
using
experimental
for
dash,
so
dash
right
now
is
still
under
the
experimental
area
of
psy.
B
C
B
Okay,
and
if
you,
if
you
actually,
if
you
go
look
at
the
commit
history
of
this
repo
you'll
see
that
marion
already
added
dash
to
sci
experimental
some
months
ago,
you
might
even
might
be
able
to
see
it.
You
actually
need
to
go
to
the
sci
repo
you're
in
you're
in
marion's.
B
B
D
You
had
talked
earlier
talked
about
using
psy
as
a
sub
module
within
dash.
Yes,
this
is
not
the
same
right.
B
No,
this
is
a
different
subject.
This
is,
I
mean
well
they're,
they're
they're,
orthogonal
issues,
so
marion
is
created
by
experimental
entry.
I
started
in
march
he's
just
adding
to
it
as
we
go
along,
but
including
I
as
a
sub
module
in
the
dash
is
a
separate
topic.
That's
just
a
way
of
configuration
management
and
I've
already
got
that
working
and
I'll
show
it.
Tomorrow
morning.
D
Okay
sure
in
that
case,
because
even
like
from
intel
the
work
that
we
are
doing
for
the
you
know
the
test
harness
and
all
that
the
code
has
to
be
upstreamed
at
some
point
and
the
question
is
within
psy:
do
we
want
to
have
a
separate
dash
project
for
all
these
changes,
or
will
it
go
directly
under
psy?
That's
yeah!
Those
are
the
you
know,
discussions.
I
think
we
have
not
yet
had.
B
D
E
This
is
a
job
we
need
to
do
that
because
having
where
you
have
the
repo
makes
a
difference
legally
on.
What's
what's
going
on
there
and
we've
had
these
issues
in
the
past,
where
it
sits
and
whether
companies
can
contribute
and
all
that
kind
of
stuff.
So
we
need
to
have
that
discussion
not
in
this
meeting.
But
let's
take
note,
we
should
make
sure
that
the
whole
community
is
in
alignment
on
where
this
is
sitting
and
what
it
actually
means
legally.
B
Yeah,
I
don't
really
know
the
best
practices
of
the
whole
psy.
You
know.
Community
development
so
probably
be
more
of
an
audience
in
that
discussion,
but
I
am
incorporating
what's
currently
in
place
into
our
build
structure,
and
you
know
we
can
always.
E
B
E
A
Yeah
gerald
what
I
was
presenting
was
marion,
submitted
a
pull
request
here
to
the
open,
compute
sai
folder,
and
asked
me
to
present
it.
He
has
a
qbr.
So
this
is
his
side
proposal
for
h,
api
and
share
with
the
group
and
ask
to
leave
comments.
So
that's
what
I
was
doing
right
here.
I've
put
this
link
in
the
chat
window.
It's
about
200
lines
of
definition.
E
And
is
it
and
okay?
So
we
need
michael
and
other
engineers
to
review
this,
I'm
sure
they
did
it
based
on
what
we
talked
about,
but
we'll
just
need
to
go
through
this
in
detail.
I
don't
think
we
can
do
that.
I
think
that
if
people
know
us
here,
everybody
can
have
a
chance
over
the
next
week
to
go,
read
it
and
come
with
comments,
as
opposed
to
trying
to
read
it
real.
A
Yeah
correct
yeah,
I
was
just
presenting
it
for
people
to
see
and
put
it
in
the
chat
window
for
people
to
review.
D
A
Okay,
great
great
and
other
things
to
address
today
so
so
last
week
we
were
talking
about
whether
you
know
the
technology
provider
chooses
a
strict
or
loosely
supported
option
and
the
trade-offs
in
the
algorithm
etc,
and
I
haven't
haven't
seen
more
of
a
write-up
on
that
it
was.
A
Was
there
a
paper
to
be
written?
We
created
inside
the
we've
converted
john
carney's
email
to
us
into
a
markdown
file
for
everyone
to
read
in
the
dash
repo
I'll
show
you
guys.
So
we
can
take
a
look
at
that
and
that's
in
documentation
and
high
availability
here
and
then
design.
A
So
these
are
the
the
proposal,
the
new
ideas
from
john
and
the
original
requirements.
So
if
anyone
has
ideas
or
would
like
to
go
review,
these
please
feel
free
and
gerald.
Did
you
have
a
document?
We
were
maybe
gonna
look
at
her
right.
D
I
think
last
week
there
was
one
open,
which
is
that,
when
there
are
two
partners
that
are
equal
cost
from
the
host
from
the
vm
and
the
flows
would
be
you
know
distributed
in
the
sense,
it
would
go
to
one
of
the
partners
versus
the
other,
and
that
is
not
in
our
control
because
that's
based
on
the
algorithm,
that's
implemented
in
the
server.
D
So
the
question
was
that
how
will
we
program
from
sdn
the
status
of
one
of
the
partners
to
be
the
one
that
would
sink
the
flows?
To
the
other
side?
It
would
be
more
like
basically,
that's
not
possible,
because
the
sdn
may
not
be
able
to
control
that.
So
in
that
case,
that
variable
that
was
added
for
status
of
sorry
active
status
would
be
more
like
a
query
attribute,
not
something
that
sdn
would
write
down
to
the
device.
D
Role
of
the
partner
as
an
active,
I
forget
the
term,
but
you
know
if
it's
programmed
as
an
active
hr
partner
for
that
flow.
That
will
not
be
something
that
the
sdn
controller
will
be
able
to
determine.
That's
not
deterministic
from
the
standpoint
of
the
sdn
controller.
It
is
you
know
it
can
it.
D
It
will
be
based
on
how
the
you
know,
which
partner
the
flow
will
actually
go
to
and-
and
that
will
be
more
like
a
query,
attribute
suppose
right
as
querying
the
status
of
the
device
that,
whether
it
is
active
at
your
partner
or
not
for
that
particular
flow,
and
it
will
not
be
possible
for
us
to
assign
that
from
the
sdn
standpoint.
E
D
Correct
that
part
is
totally
agreeable.
I
think
if
we
go
back
to
the
slides
that
marian
had
presented
last
time-
or
maybe
it's
here
somewhere-
there
was
in
the
in
the
session
attribute
in
the
one
of
the
session
attributes
was
the
active
role
for
the
device
to
sync
the
flows
to
the
partner
hi
partner
device,
and
that's
something
that
we
cannot
assign
from
the
sdn
standpoint
in
case
of
the
topology,
where
the
two
partners
are
equidistant
from
the
from
the
server
where
the
flows
originated
from.
E
E
D
In
that
case,
one
of
the
paths
has
to
be
longer
and
one
path
has
to
be
shorter.
If.
E
It
is
not
physically
longer
just
bgp
allows
you
to
append
abs
artificially
to
make
it
look
longer
physically
they're
going
to
be,
they
could
be,
and
most
likely
will
be
equidistant,
but
logically
one
will
walk
longer.
E
E
All
we
need
to
guarantee
in
dash
is
that
it's
gonna
arrive
that
you
know
the
primary
will
be
the
preferred
and
that's
all
we
need
to
do
in
dash
and
bgp
and
how
people
want
to
configure
it,
whether
they
use
meds,
what
they
use.
You
know
asp,
pins
and
all
that
kind
of
stuff,
that's
kind
of
outside
the
scope.
As
long
as
it
happens,.
D
In
that
case,
we
need
to,
you
know,
clarify
that
the
paths
are
never
going
to
be
equidistant,
at
least
from
the
routing
standpoint,
and
then
it
will
be
totally
fine
and
it
can
always
be
written
from
the
sdn
yeah.
E
A
And
so
yeah,
that's
that's
what
I
have
for
this
week
and
unless
you
guys
have
anything
else.
E
And
that's
fine
they've
worked
done
a
lot
of
work,
obviously
to
get
it
to
that
state.
So
now
we
just
need
another
week
for
everybody
to
review
and
come
back
with
comments.
A
E
Yeah-
and
hopefully
we
can
get
through
this
relatively
quickly,
because
I
know
that
there's
a
lot
of
people
anxiously
awaiting
to
get
back
onto
the
how
aj
is
done.
But
I
think
this
isn't
the
right
order
and
importance
just
get
this
out
of
the
way
so
that
we
can.
We
can
move
on.
A
Okay,
all
right
thanks,
everybody
gerald
caroline
decided
to
join
today
so
wanted
to
let
you
know
she's
here:
yeah,
yeah,
okay!
A
Well,
that's
what
I
have
guys.
So
thank
you
very
much
for
coming
today
and
if
you
could
please
go
ahead
and
review
the
pr.
Let
mario
know
what
you
think
or
there's
any
changes
or
corrections
needed,
and
we
can
see
you
next
week.