►
From YouTube: 2017-09-20 Performance WG
Description
Agenda and Notes: https://docs.google.com/document/d/12hWGuzbqyNWc2l1ysbPcXwc0pzHEy4bodagrlNGCuQU/edit?usp=sharing
B
B
C
E
C
D
Do
you
know
if,
if.
A
A
Sure,
ok,
so
this
is
the
second
meeting.
It's
been
what
almost
two
months
I
guess
since
last
meeting
and
we
wanted
it
to
kind
of
go
over.
What's
happened
in
the
meantime
and
see
if
anyone's
you
know
done
any
work
like
benchmarks
or
anything
like
that.
So
I
don't
have
an
agenda
for
this
one,
but
we
can
kind
of
add
items
that
are
of
interest.
One
that
I
wanted
to
mention
to
people
is
the
je
Malik
reprofiling,
that's
being
worked
on
right
now,.
A
D
So
so
III
mean
I'll
talk,
I'll
talk,
one
thing
to
those,
so
I
think
the
Apple
folks
are
turning
and
turning
those
on
and
playing
with
them
and
I
know
that
at
least
Dario
is
working
on.
D
The
auto
is
coming
from
Judith
when
I'm
talking,
because
I
waved
your
people.
Maybe
it's
not
weird
I,
don't
know.
Okay,
good,
so
I
know
that
Dario
is
gonna,
be
sitting
and
chatting
with
me
this
week
about
adding
an
even
higher
performance
message,
queue
specifically
for
the
case
where
we
have
multi
riders
single
reader.
So
so
we
put
a
queue
in
which
is
multi
multi
reader
multi
writer,
which
is
important
for
the
process
queue.
But
it's
not
important
for
them.
A
message:
queue
for
the
event,
queue
and
so
I
from
a
performance
perspective.
D
F
D
A
A
A
Do
folks
want
to
talk
through
the
Jade,
Malek
memory,
profiling
or
I?
Think
right
now,
where
it
is
is
been
know,
is
working
on
a
kind
of
a
design.
Write-Up
of
it
and
he'll
be
publishing
that
pretty
soon
so
we
can
either
I
mean
folks,
can
either
wait
for
that
and
have
a
look
at
it
or
if
we
want
to
talk
about
it
now
we
can
do
so.
We
don't
have
been
over
here,
though.
Unfortunately,
who's
Rick.
D
F
E
To
bring
out
it's
not
simple,
playing
or
not,
but
I
think
I
discussed
with
a
mother
like
so
currently
on
our
ancient
missus
interest
agent
process
to
use
like
up
to
you
usually
use
like
one
or
more
of
memory,
but
when
I
try
to
use
the
email
out
to
profile
the
our
missus
agent
process,
it's
the
profile
shows
that
only
like
dozen
of
megabytes
are
used.
So
there
are
some
unknown
portion
of
memory
that
either
not
profiled
or
like
I,
don't
know
it
will
come
from.
E
So
I
was
wondering
that
if
there's
anyone
like
understand
more
about
why
why
jameelah
can
not
profile
those
Emily
like
they
are
I,
get
hundreds
of
megabytes
and
I.
Just
don't
know
where
this
come
from.
You
know,
I'm
the
claim
electrophile
one
possibility.
I
guess
is
that
it
probably
cannot
profile.
M
map
amount
memory
or
maybe
something
related
to
the
I'm.
Not
they
have
a
long
time
that
James
James
mention
the.
G
A
Might
like
I
or
when
you
started
started
to
look
at
this
I
thought
that
we
were
just
storing
a
lot
of
history,
but
that
would
show
up
in
the
profile
yeah.
So
it's
a
little
odd
that
there's
memory
consumption
that
we
can't
isn't
showing
up
in
the
profile
right
now
it
is
sampling
based,
though
right.
So
it
could
be
missing.
Some
a
good.
F
F
A
Okay,
for.
D
C
D
C
D
D
G
B
C
B
C
C
D
Mean
it's
it's,
it
seems
like
it
would
be
good
to
do
two
things.
One
is
to
capture
somewhere
and
in
in
a
dock,
just
trade-offs
and
proposals
in
the
proposals,
the
trade-offs
and
the
proposals
that
you
have
Dimitri.
It
sounds
like
you
got
a
couple
ideas.
You
know
in
terms
of
how
to
use
arenas.
What
the
API
should
look
like
be
great
to
review
that
and
then
the
other
one
is
in
terms
of
asking
whether
or
not
the
community
wants
to
use
arenas.
I
I
mean
I'm,
assuming
the
performance.
Winds
are
really
big.
D
But
if
we
don't
know
if
the
performance
winds
are
big,
then
maybe
that's
worthwhile
to
just
do
a
quick
assessment
like
just
hack,
a
quick
prototype
together
to
see
what
that
looks
like
because
it
seems
like
if
the
performance
winds
are
big
and
we
can
come
up
with
a
good
proposal.
We
should
totally
use
them.
C
Yeah
I
hope
they
should
should
give
us
some
advantage
when
using
for,
like
for
processing
read-only
data.
But
if
we
need
to
the
problem
is
that
in
most
cases
we
defer
execution
to
to
some
other
method,
and
in
this
case
we
should
move
the
object,
but
we
cannot
move
it
from
arena
without
copying.
That's
the
main
problem.
It's
often
authenticate
something
then
defer
to
different
parameters
to
other
method,
and
that's
where
we
just
lose
all
advantages
of
arenas.
A
Yeah,
so
it's
almost
like
in
a
Const,
the
install
handler
is
taking
it
as
a
constant
thing,
and
you
know
that
not
gonna.
Well,
they
might
copy
it.
But
you
know
that
the
intention
is
not
to
mutate
that
and
then,
if
they're
taking
a
non-constant
on,
maybe
like
maybe
the
decision
about
whether
its
user
in
arena
is
based
on
the
constants
of
the
message.
I,
don't
know,
yeah,
that's.
C
D
Yeah,
hey
sounds
like
we
could
possibly
also
do.
A
I
could
possibly
also
do
I
mean
put
something
in
a
arena
pass
it
around
as
a
shared
pointer
as
a
concentrate
pointer.
D
A
D
Okay,
cool
I've
man
that
cuts
a
good
thread
to
keep
tugging
on,
but
it
seems
like
in
terms
of
prioritizing
which
ones
we
want
to
focus
on.
We
should
make
sure
we
have
benchmarks
and
or
some
prototypes
that
are
showing
the
performance
wins,
because
I
actually
have
no
idea
to
meet
you.
Maybe.
But
you
have
a
better
insight,
because
you've
tried
some
of
this
stuff
out
already
but
like
I,
have
no.
C
C
A
B
A
D
That's
the
biggest
thing
there
is
I'd
love
for
us
to
get
just
a
benchmark
cuz.
Once
we
have
benchmarks,
people
can
just
start
playing
like
it's
great.
That
Daario
can
now
be
playing
on
improving
the
improving
the
message,
press
and
performance.
You
can
just
keep
running
that
benchmark
like
HTTP
is
the
same
stuff.
That
could
be
really
great
if
we
can
get
a
benchmark
in
for
HCP
I.
G
F
Least
four
HTTP
I,
don't
know
I'm
personally,
not
aware
of
any
performance
issues
with
the
HTTP
layer
per
se,
like
the
things
that
I've
encountered
above
that,
like
dispatching
a
lot
as
a
side
effect
of
an
HTTP
call
or
just
generating
ridiculous
amounts
of
of
response
of
content
in
response
which
ties
up
a
process.
So
those
things
are
benchmark,
abort
but
they're,
not
part
of
and
and
clearly
fixable.
But
it's
not
like
really
sure.
A
Yeah
I
mean
I,
think
we
I
think
specifically
we're
thinking
not
just
the
process
HTTP,
but
the
masters
HTTP
like
if
you
look
at
some
discussions
that
happened
in
the
past,
where
someone
I
think
was
darĂo
benchmarked
the
old-style
API
versus
the
HTTP
API,
and
it
was
around
maybe
half
the
throughput
or
something
like
that,
possibly
just
because
of
what's
happening
above
the
HTTP
layer
of
the
process.
That's
that's
still
kind
of
band
and
performance.
Isn't
that's
good,
so
yeah
I
mean.
F
A
It
didn't
match
the
original
API
performance,
but
I
think
the
reason
that,
like
I'm
inclined
to
put
that
there
is
just
that
there
are
a
lot
of
known,
there's
a
lot
of
known
overhead
in
the
HTTP
layer
in
the
process.
Like
there's
a
lot
of
unnecessary
work
being
done
like
we
copied
the
request
multiple
times,
and
so
like
that's
stuff,
that
we
can
Fix,
that's
gonna,
definitely
improve
HP
layer
performance.
A
G
D
D
D
B
So
I
wanted
to
talk
about
a
recurring
schedule
for
this
meeting.
If
we
wanted
to
start
doing
it
on
a
more
regular
basis,
I
proposed
once
a
month
to
start
out
and
see
if
we
want
to
have
more
frequent
after
that.
Does
this
time
work
do
people
like
that
I
see
you
nodding,
head
yeah.
A
G
E
D
G
B
Awesome
cool
so
we'll
set
it
recurring
for
now
monthly
at
this
time,
and
if
we
we
can
float
it
in
the
slack
channel
and
if
other
people
want
to
join,
who
this
time
isn't
good
for
we
can
change
it.