►
From YouTube: SONiC DASH Workgroup Community Meeting May 31 2023
Description
Thank you to Chris Sommers for leading the call!
PR373 - Update Vnet PTF Tests (Yuriy) got merged during meeting.
PR371 Update to DASH_ROUTE_TABLE entries with action direct (Oleksandr) - requesting review. Update - was approved, merged.
PR372 - ixia-c security updates (Chris) - requesting review by Ze Gan (pterosaur) - Ze said he would look at it week of 6/8/2023
PR365 - Support for metering (Vijay) - extensive updates were made per several past meetings and online reviews. Requesting final review from commenters and anyone else so we can commit this.
This PR had collateral test-case issues, these tests are temporarily disabled pending newly filed issues 374,375; issue 376 was filed to ensure we re-enable them.
Vijay and Reshma requested we push the auto-generated headers to opencompute/SAI repo (under experimental/).
Hanif - any versioning for SAI repo changes? Ans: SAI 1.12 has "stable" DASH APIs under experimental. DASH experimental APIs get included in SAI "releases" through normal SAI community process.
Eventually DASH APIs should get promoted to first-class status under SAI/inc instead of experimental/, e.g. when DASH is at "1.0" release level, including stable APIs and all associated test artifacts.
PR366 - needs review, request Prince to walk us through this- with the community
Chris mentioned upcoming SmartNIC Summit June 13-15 including DASH all-day workshop on June 13.
A
Recording
right
thanks:
okay,
we're
recording
welcome
everybody.
This
is
the
Dash
Sonic
Dash
working
group
community
meeting
and
today
is
Wednesday
May,
31st
and
I'm,
covering
for
Christine
I'm,
all
just
on
holiday.
So
thanks
for
all
coming
got
another
18
people
we'll
get
rolling,
so
we'll
just
cover
some
of
the
review.
What
happened
last
week
just
look
at
the
minutes
that
Christina
sent
out
briefly.
A
So
last
week
we
covered
several
PR's.
There
are
several
outstanding
and
we'll
we'll
take
a
review
of
those
in
a
moment.
They're
listed
here
and
also
there's
some
commits
to
update
some
documents.
The
dash
route
table
entry
was
updated
and
also
Sonic
hld
was
was
updated
to
elaborate
some
return
values
for
the
ACL
tagging.
A
B
I
thought
to
cover
it.
Lastly,
but
basically
the
number
of
flows
tables
from
64
million
was
reduced
to
32
million.
That
was
one
of
the
big
changes
to
account
for
32
gig
of
RAM
for
the
200
gig
Asic
and
then
400
gigans
on
a
double
of
that.
So
all
the
numbers
or
flows
have
been
reduced
by
half
and
also
for
the
hero
test
is
the
background
traffic.
The
packet
size
has
been
reduced
from
1500
to
500
size.
B
This
will
increase
basically
by
three
times
the
number
of
packets
on
the
wire
for
background
traffic,
to
have
more
stress
and
also
during
the
Hero
Test.
The
background
traffic
from
1
million
TCP
flows
and
1
million
UDP
flows
will
have
almost
to
the
max
so
15
million
and
15
million
for
a
total
of
30
million
flows
active
during
background
traffic
and
then
for
the
CPS
test.
There
are
like
two
millions
left
available.
A
C
B
A
B
A
A
So
another
PR
which
got
closed
emerged
was
368.
and
Vincent
from
excite
was
made
some
changes.
He
was
using
a
newer
Ubuntu
to
get
some
compiler
optimizations
for
his
work
and
it
caused
issues
trying
to
link
is
lib
side
with
the
psythrip
generated
code,
which
had
a
Ubuntu
2004
per
the
build
environment,
doctor
files,
and
so
he
made
some
changes
to
allow
you
to
pass
in
a
different,
basically
a
different
tag
for
building
a
new
Ubuntu
based,
Dockers,
and
so
he's
got
an
environment
variable.
A
A
What
was
a
static,
Ubuntu
tag
with
an
environment
variable
that
can
be
passed
in
that
way
you
can
customize
your
build.
If
you're,
you
know
vendor
implementation
say,
can
you
still
use
the
dash
repo,
but
they
can
customize
easily
to
use
a
different
Ubuntu.
So
that's
what
this
PR
was
any
any
questions
about
that.
A
A
A
A
A
Let's
see
if
there's
anything
else
here,
so
another
pull
request
that
is
pending
and
it's
looking
for
a
view
from
public
from
Z
Gunn
who
filed
an
issue
this
issue
about
security
risk
here
in
HCC
controller,
so
xiac
software
packet
generator.
It
consists
of
two
Docker
images:
one's
a
controller
with
a
rest
endpoint
and
consists
of
also
traffic
engines
that
are
managed
by
that
controller,
and
he
discovered
that
the
controller
had
some
older,
TLS,
1.0
and
1.1
were
still
enabled.
A
That
was
just
because
it
was
using
a
you
know:
standard
Go,
distribution
for
the
endpoint
server
and
no
one
had
disabled
those
modes.
So
we
have
I,
have
updated
to
use
newer
version
where
we
took
out
those
modes
and
just
needs
approval,
so
I'm
bumped,
bumped
Z,
to
ask
him
to
take
a
review
of
this
down
here
and
so
yeah.
A
Hopefully,
you'll
see
this
ping
here
and
review
it
and
we'll
sponsor
view.
It
they're
welcome
so
I,
updated,
xcc
controller
and
also,
as
we
upgraded
HCC
controller,
we're
using
the
default
Port
of
8443.
Now,
instead
of
443,
because
this
was
interfering
with
some
people's
environments,
they
were
already
using
https.
A
A
E
A
F
A
F
A
A
D
A
G
Yeah
Chris,
so
just
an
update
that
I've
updated
the
pr
with
all
the
comments.
I
think
there
were
a
few
comments
on
the
pr,
and
there
was
also
one
by
Andy
about
changing
the
P4
code
for
the
counter,
so
I've
addressed
pretty
much
all
of
them.
Yeah
I
think
there
are
no
other
outstanding
comments
and
it
should
be
ready
for
March
if
everyone
is
okay.
A
G
A
H
C
A
I
think
this
also
brought
up
the
discussion.
That's
that
we
have
some
attributes
in
inside
when
they're,
when
they're
not
mandatory,
if
they're,
if
they're
optional,
that
we're
still
kind
of
rejecting
and
we're
rejecting
entries
and
the
P4
runtime,
they
don't
have
every
attribute
I
think
that's
really
a
P4
runtime
P4
limitation.
If
I'm
not
mistaken-
and
you
recall
that
discussion-
if
we
have
a
table
and
we're
mapping,
let's
say
table
action
parameters
to
psi
attributes
P4
runtime
is
not
forgiving
with
respect
to.
A
I
A
I
D
A
D
A
A
G
A
D
A
D
A
D
H
F
H
H
F
H
You're,
if
you're
still
experimental,
then
perhaps
you
know
it
might
be
okay,
but
at
some
point
you
know
you
say:
okay,
I'm
I'm,
you
know
I'm
giving
you
these
things
and
I'm
I'm.
Basically,
you
know
solidifying
a
certain
release
version
of
PSI
API
and
then
you
can,
if
you
can
conform
to
that,
then
you
know.
Essentially,
this
is
what
it
is,
but
and.
D
H
Obviously,
we
know
that
you
know
the
apis
do
change
and
then,
but
they
are
in
some
ways:
Backward
Compatible
or
so
forth.
But
at
some
point
we
have
to
see
that
okay,
we
have
to
have
a
process
through
which
we
are
going
to
publish
these
Auto
generated
apis
for
people
to
consume
and
then
make
their
thigh
conforming
to
those
apis.
Libraries
I
think
we
have
to
have
some
process
around
it.
E
D
H
E
D
D
A
H
There,
a
consent,
yeah
I,
think
we.
We
also
have
some
sort
of
a
milestone,
Chris
that
we
want
to
have
say:
okay,
it's
a
1.0,
Sonic,
Dash
or
whatever.
Basically,
you
want
to
come
out
with
right.
So
should
that
not
be
the
the
Milestone
that
we
can
say?
Okay,
this
is
my
1.04
0.1,
for
you
know
my
size
also,
and
then
we
at
that
time
they
graduate
from
experimental
to
the
next
level
right,
I,
guess.
A
A
A
A
A
A
I
A
A
A
Okay,
well,
we
can
keep
this
to
be
a
perfunctory
meeting,
just
good
to
have
refresh
of
everything
just
remind
people
that
smart
Nick
Summit
is
in
a
couple
of
weeks
in
San
Jose
in
case
anyone
had
any
thoughts
of
going.
It
would
be
there's
going
to
be
a
pretty
strong
Dash
component.
Microsoft
is
hosting
a
full
day
of
Dash
workshop
on
Tuesday
I
believe,
first
day
or
the
zeroth
day
of
the
show
they
call
it.
Pre-Conference
I
know
we'll
be
joining
in
some
presentations
there,
so
I
think
I.
A
A
Nick,
the
the
the
I
think
it's
it's
still
kind
of
in
flux.
You
know
it's,
it's
sort
of
a
I
think
some
items
have
been
locked
down.
There
might
be.
You
know
some
slots
that
are
still
being
debated,
but
I'll
ask
Christina.
If,
if
there's
kind
of
affirmative
agenda,
would
that
maybe
have
an
impact
on
whether
you
go
to
the
summit
or
not.
H
F
A
Yeah
I
know
they'll
definitely
be
like
intro
to
Dash,
and
you
know
for
people
who
will
be
some
people
come
to
these
Summits,
where
they
really
are
coming
to
learn
almost
from
you
know
very
little
knowledge
base
about
either
smart
Nicks
or
applications.
So
I
know
they'll,
be
kind
of
a
general
intro
part,
but
then
they'll
be
a
significant
slot
on
testing,
which
you
know
will
present
some
of
our
stuff,
which
will
be
very
familiar,
but
I
don't
know
about
all
the
other
topics.
Yet
so
I'll
ask
Christina
about
that.
That
would
be
good.
A
Thanks
for
asking
yeah,
and
just
just
a
slight
related
topic,
I
was
I
was
talking
to
Andy
here
about
people
runtime,
it's
related
in
that
you
know
we're
using
people
runtime
for
the
behavioral
model.
You
know,
we've
already
found
certain
limitations
in
the
P4
runtime
and
trying
to
map
it
to
Psy
and
having
optional
attributes.
Then
we've
done
that
mapping.
A
Intel
spearheading
a
look
into
using
something
called
generic
tables
and
adding
that
to
people
runtime,
and
so
it
does.
It
does
kind
of
intersect
with
the
interests
of
this
working
group.
So,
if
anyone's
interested,
you
know
they
can
drop
in
on
a
P4
working
group.
P4
API
working
group
anytime
and
I
co-chair
that
with
someone
from
Google
and
Andy
is
often
at
those
meetings
as
well
with
kind
of
partners
in
crime,
there
so
feel
free
to
join
up
just
to
listen
or
learn
or
participate.
A
The
the
normal
schedules
once
a
month,
every
I
should
say
every
every
fourth
Friday
to
be
more
accurate
at
9
30
Pacific
right
now
we're
having
weekly
meetings
to
work
on
this
generic
table
issue,
because
it's
pretty
involved-
and
you
know
we're
hammering
out
so
when
there's
sort
of
routine
business
every
four
weeks.
But
we
have
a
special,
let's
say
task
force
for
fewer
people
come
and
we're
meeting
once
a
week
to
try
to
make
progress
on
P4
runtime
generic
tables.