►
From YouTube: SONiC DASH Workgroup Community Meeting Mar 22 2023
Description
PR289 - Still needs volunteer - Dockerfile.saithrift-client has a static tag in the FROM instruction
PR336 - [ACL]: Optimize DASH ACL by introducing cluster tag
DASH Community will have a few comments, Marian, Prince, Reshma, et al will take a look and come back
PR246 - SAI bulk add/remove implementation - Status Issue
Possibly resubmit this as another PR - error popped out of the blue
Git Command amend and force?
PR296 - Add DASH outbound metering to Sonic HLD
https://github.com/sonic-net/DASH/pull/352 - Kristina merged 3/25/2023
PR315 - Added more Vnet-to-Vnet PTF test cases check-in to see if this is finished
ISSUE351- direction decision using eni mac/ip not the vni alone (Mircea)
VXLan per RFC or per DASH use cases? Discuss use of multiple VNIs…
B
Chris
I
was
going
to
show
you.
We
have
on
campus
on
Microsoft
campus,
a
farm
scenario
set
up
where
they
have
sensors
and
it's
as
if
we
had
Farm
links
to
monitor
pieces
of
a
farm.
It's
an
agriculture
thing
over
by
building
43.
I
took
pictures
of
it
for
you
and
I
just
have
it
sent
into
you
yet,
because
your
cell
phone
doesn't
get
reception
and
I
I
haven't
done
it
through
teams,
but
anyway
hi
everyone
thanks
for
coming
to
the
March
22nd
Dash
meeting.
B
What
I'm
showing
on
my
screen
is
what
we
talked
about
last
week
and
we
talked
about
the
second
half
of
alibaba's,
submission
and
they're,
going
to
go
back
and
think
about
that.
Some
more
and
talking
about
you
know
becoming
eni
aligned
and
maybe
unifying
the
float
table,
and
we
also
had
a
behavioral
model
last
behavioral
model
meeting
last
week,
which
was
good
and
Prince
covered
the
inbound
part
of
service
tunnel.
Private
link,
not
the
BMV
part
bmv2
part,
but
he
covered
the
documentation
that
he
had
submitted.
B
That
was
awesome,
and
so
that's
what
we
did
last
week
and
this
week
I
saw
a
few
PRS
coming
through
just
some
automatic
notifications
about
this
morning.
Petitsar
had
made
updates
to
this
cluster
tag
one,
and
then
we
had
an
issue
with
246,
where
we
were
having
some
sort
of
status
issue
and
296.
We
could
take
a
look
at
and
then
here
on
the
side
of
the
screen.
These
are
all
the
PRS,
so
I
was
wondering.
Did
anyone
have
one?
In
particular?
We
wanted
to
start
with
today,
Prince
or
anyone.
A
A
A
C
A
A
A
A
B
B
D
D
B
B
B
A
E
E
B
B
F
F
G
F
So
we
can
have
a
meeting
like
outside
of
this
meeting
if
it
need
be
just
to
go
through
it
with
maybe
Prince
and
Gohan
or
somebody,
but
it's
it's
come
up
multiple
times
now.
So
obviously
it's
not
clear
and-
and
we
do
have
one
implementation-
that's
kind
of
blocked
by
a
little
bit
right
now.
So
it
would
be
good
if
we
just.
B
F
G
G
So
what
has
necessary
GRE
to
do
with
the
dni
from
dxlan
and
then,
if
we
have
ipsec
as
a
tunneling
or
any
other
tunneling
protocol,
making
a
decision
based
on
dni
may
not
be
surprising.
Now,
Mac
and
IP
by
themselves
are
not
enough,
because
customer
one
can
have
190
to
168
0
1
with
a
Mac
and
customer
too,
if
he's
lucky
can
have
also
192
1601
for
their
VM.
With
the
same
map,
you
are
allowed
to
have
two
VMS
with
mipc
map
under
different
customers
yeah.
G
So,
since
that's
a
possibility,
we
need,
on
top
of
it
and
Mac
to
add
something,
but
that
something
student
visa
via
nice
to
be
either
a
tunnel
ID
or
a
customer
ID
or
something
more
generic
and
then
the
funnel.
If
we
choose
tunnel
ID
tunnel,
ID
can
have
a
protocol
type
where
it
can
be
vxlan
as
one
protocol,
the
other
IP.
Second.
So
on
I'm,
not
sure
if
there
is
a
case
where
there
is
no
tunnel
from
the
VM
to
the
dpu
on.
H
C
H
C
Is
unique
to
the
VM
right
within
that
v-net,
so
once
it's
outside
of
the
v-net,
then
the
BNI
could
be
different
pertaining
to
that
winner,
but
within
the
v9,
where
the
VM
is
originating
the
traffic,
the
vni
should
uniquely
identify
that
VM
and
the
appliances
in
that
v-net
itself.
So
it
should
be
completely
unique.
I
G
G
I
H
But
you
it
has,
you
know,
there's
a
use
case
for
that
right,
so
I
think
the
dash
is
based
on
the
you
know,
real
use
cases
here
right.
So
if
you're
talking
about
okay,
you
know
with
the
testing
you
want
to
do
this,
or
is
this
there's
some
customers
that
have
a
valid?
You
know
use
case
here.
I
mean
yes,
I.
Think
I
suggest
that
you
know
the
customer
to
bring
their
use
case
here
and
we
can
consider
to
enhance
the
you
know
the
dash
model
here
right
so
I.
F
H
G
G
G
H
Where
is
the
use
case?
I
mean?
Is
that
use
case
for
just
for
testing
or
is
the
use
case
for
some
real
scenario?
There
is
a
need
to
use
multiple
viewing
eyes,
I,
don't
know
why
we
are
using
multiple
bearing
because
you
know
it's
clear
that
you
know
we
are.
You
know
for
all
the
outbound
traffic
we're
just
using
one
VMI
right
so
to
to
know
whether
this
is
the
outer
bound
or
inbound.
G
H
G
H
I
But
I
think
this,
so
special
Vienna
is
only
for
setting
the
direction
for
the
outbound
other
than
that
you
have
a
unique
vni
per
V
in
it,
and
the
eni
can
belong
to
different
v-nets,
so
so
outbound
direction
is
set
by
something
like
you
know.
The
host
is
sending
to
this
Appliance
just
for
for
setting
that
outbound
path.
G
D
G
G
That
is
correct,
but
the
IP
and
the
Mac
could
be
under
the
control
of
the
customer.
If
I'm,
creating
a
VM
in
mine
that
I
want
to
be
part
of
my
network
and
I
started
in
the
data
center.
I
can
start
the
VM
with
192
16801,
in
my
account,
and
you
could
start
with
VM
with
cmip
in
your
account
192
16801,
because
that's
how
we
have
our
company
Network
and
that's
an
ideal
sign.
So
everything
goes
fine.
I
I
think
machine.
In
that
case
those
two
VMS
have
its
own
unique
or
same
IP
doesn't
matter,
but
when
the
host
is
sending
that
to
this
Appliance,
it
will
send
with
this
vni
in
the
outer
header,
so
that
the
appliance
knows
that
it
is
coming
from
the
VM
so
to
set
the
outbound
and
then
it
based
on
the
inner
Mac.
It
will
go
to
different
Enis.
That's
how
the
Enis
are
are
identified.
I
G
F
J
J
There
are
various
many
various
ways
to
do
this
in
the
data
plane
more
efficiently
than
linear.
You
know,
walk
through
the
list
of
tackle
rules,
and
but
some
of
them
have
different
scaling
properties
depending
upon
the
details
of
what
the
typical
configurations
look
like.
I
was
wondering
if
there
are
any.
J
F
You
can't
share
customer
data
for
one
but
two
I
think
Michael's
said
multiple
times.
You
cannot
assume
anything
about
them
because
they're
not
defined
by
us,
so
we
all
need
to
find
the
first
rules
table
and
that's
like
the
v-net
table,
but
all
the
subsequent
lookups,
all
of
the
four
following
tables
they're
actually
defined
by
the
user,
and
we
have
no,
we
don't
have
any
influence
all
over
them.
F
J
So
so
in
particular,
I
would
assume,
since
you
don't
have
control
over
them,
and
you
give
them
it's
probably
common
that
let's
say,
there's
n
rules
and
a
single
packet
can
match
any
number
of
those
rules
and
it's
the
highest
Priority
One
that
should
its
action
should
should
win
it's
it's
atypical
or
it
it's
very
common
for
a
single
packet
to
match.
Let's
say
to
match
a
prefix
in
multiple
rules
is.
J
F
F
J
F
J
Sure
no
I
this
was
this
is
just
like
I've
had
internal
discussions
Intel
and
the
first.
This
is
like
a
first
basic
thing
that
changes
the
complexity
of
the
problem
enormously.
Whether
there
could
be
multiple
materials
or
not
and
is
my
assumption-
was
multiple
rules
matching
as
normal?
Yes,
that's
true
thanks.
B
J
A
I
A
J
The
the
the
issue
I
know
the
issue
is
not
unique
to
dash
because
another
customer
has
had
a
similar
thing,
so
that
has
raised
the
priority
of
somebody
in
Intel
fixing
it
and
another
aspect.
Another
update
is
I,
have
successfully
taken
smaller
P4
programs
that
do
the
add-on
Miss
feature
and
successfully
run
them
on
the
dpk
software
switch
that
didn't
hit
this
bug
and
so
I've
gotten
further
than
that
with
other
non-dash
programs.
J
C
I
think
most
of
the
pipeline
requirements
for
Dash
is
some
more
generic
and
the
different
tables
that
we
have
can
be
accommodated
in
those
table
types
also
in
dpdk,
so
I'm
hoping
that
it
will
not
be
too
many.
Issues
that
will
run
into
this
construct
of
ironess
for
sure
is,
is
new
and
the
dash
I
don't
miss
is
much
more
complex
than
the
simple
simplest
version
that
we
had
Upstream
long
time
ago
in
typical
line
yeah.
Maybe
there
are
bugs
there,
but
I'm
hoping
there
are
not
too
many
issues
so.