►
From YouTube: Node.js Benchmarking WG Meeting - April 19 2019
Description
A
A
B
You
know
Gabriel,
he
is
definitely
part
of
the
committee
and
he's
going
to
help
us
attending
this
meeting
going
forward
along
with
octave
and
soul.
Here,
they've
been
working
on
node
for
some
time
with
me
and
since
I'm,
getting
a
little
pushed
out
to
do
something
else.
Octavian
is
going
to
be
one
of
the
person
who's.
Actually,
maintaining
no
DC
is
going
forward
and
he'll
be
kind
of
contributing
to
working
group
progress.
Also,
okay,.
B
In
a
few
different
directions,
yeah
so
I'm
still
going
to
work
on
node.js,
optimization,
but
more
an
optimization
side
so
and
the
workload
Octavian
is
kind
of
taking
a
world
okay
and
we,
he
has
already
fixed
some
issues
raised
by
Jamie,
like
to
support
multi
micro
services,
mode
of
no
DC
I
can't
actively
what
you'll
know.
Dc
is
right:
okay,.
A
We'll
definitely
welcome
you
to,
and
hopefully
we'll
see
you
at
getting
a
lot
of
the
stuff
done
as
we
move
forward.
Okay.
So,
let's
move
on
to
the
two
issues
on
the
agenda
there.
The
first
one
is
investigate
the
time
taking
a
number
of
benchmarks
in
core
I
haven't
managed
to
catch
up
with
Gareth,
so
I
don't
have
an
update
on
that
one.
B
But
that's
something
we
can
do
discuss
with
another
Gabrielle.
Is
there
yeah?
We,
because
we
have
been
looking
at
some
of
the
important
micro
benchmarks.
We
have
in
no
DC
node.js,
for
example
the
crypto
workload,
which
are
not
too
many
the
compression
workload.
So
we
can
maybe
start
there
somewhere
instead
of
now
looking
at
since
the
Garrett
is
not
here,
yeah
we'll
start
somewhere
and
I
thought
that
could
be
a
good
starting
point.
Sure.
B
A
B
At
I
mean,
though,
for
the
Gabriel
the
background
is,
we
were
thinking
of
adding
some
of
these
internal
micro
benchmarks
for
page
as
a
negative
thing
and
we
were
not
able
because
there's
just
so
many
of
them
any
takes
I
think
multiple
days
to
cover
all
of
them.
So
we
want
to
start
with
some
small
subsets
which
have
really
really
important
from
the
performance
perspective
that
or
nightly
results
right.
C
A
That
that's
the
thing
is
that,
like,
if
you
know
the
micro
benchmarks
in
core
there's
a
huge
number
of
them,
yeah
yeah
yeah,
you
can't
run
them
all
like
running
them
all.
Take
something
like
six
days
or
something
as
I
understand
right.
You
know
a
very
long
time.
You'd
have
way
too
many
numbers
for
anybody
to
look
at.
So
our
challenge
over
the
years
really
has
been
to
say
how
do
we?
B
C
C
C
Okay,
well,
there's
I,
don't
know,
there's
a
difference
between
between
catching
a
regression
and
and
and
and
just
charting
performance
right,
so,
like
yeah
it'll
show
up
in
the
chart
as
a
spike,
either
positive
or
negative.
But
but
you
know
as
assuming
that
there
are
no
such
spikes,
then
you
know
what
can
we
meaningfully
convey?
You
know
like,
for
example,
let's
say
let's
say
we're
comparing
ourselves
to
like
nginx
and
what
are
what
are
some,
what
are
some
some
benchmarks
that
nginx
uses
you
know
to
to
tout?
It's
it's
right.
A
C
C
Don't
necessarily
mean
that
either,
but
what
I'm
saying
is
that
we
should
you
know
we
should
report
the
results
in
a
language
that
people
are
familiar
with,
you
know
and
then,
and
then
comparing
to
competitors
is,
is
just
a
side
effect
of
that
right,
because
now
we're
talking
out
two
apples,
kind
of
thing
you
know,
and
if
people
are
familiar
with
looking
at
the
performance
of
nginx,
then
what
does
nginx
show
them?
You
know
so
that
we
show
them
the
same
thing
cuz.
Then
people
will
understand.
B
Know
yeah
but
Gabriel.
This
was
moving
up,
for
example
our
crypto
or
even
before,
API
right,
so
we
had
a
buffer
analog
before
unsay.
Before
now
we
have
before
safe
right
by
default
and
people
do
see
the
performance
going
down
and
some
of
the
change
is
happening
in
some
appear,
so
we
don't
need
to
compete
directly
with
anything
else,
but
just
the
node
is
progressing.
That's
all
right!
Right!
Okay,
after
some
time,
maybe
after
we
have
enough
data
all
this
a
6-month,
then
we
can
start
comparing
if
needed.
B
C
A
C
A
C
How
about
how
about
we
we
take
like
the
there's,
the
zero
zero
knowledge
approach
like
we
I
can
I
can
put
together
like
just
a
list
of
all
the
benchmarks
and
and
then
just
have
a
poll
people.
Whole
people
can
take
off
what
they
want
to
see
out
of
them.
You
know
that
can
probably
be
done
fairly
efficiently,
but.
A
B
The
thing
I'm
basically
kind
of
thinking
is,
let's
say
we
added
HTTP
to
last
year,
so
what
it
does,
especially
if
it
compresses
the
response,
so
you
see
a
huge
reduction
in
the
data
transferred
using
the
HTTP
protocol.
So
if
we
can't
just
start
as
small
as
that
one.
So
since
we
added
that
as
a
new
API
and
then
we
kind
of
see
the
result
of
how
the
HTTP
to
so
anything
new
feature,
we
had
anything.
New
changes
like
before
allocation
was
one
thing,
because
that
was
a
big
change
right.
B
B
A
Benchmarks
right
yeah,
so
I
I'd
be
like
you
know.
What
I'd
find
would
be
really
interesting
is
like
an
HTTP
one,
which
would
somehow
take
the
best
or
the
most
important
from
our
our
micro
benchmarks,
or
even
something
else
and
say:
okay,
yeah
we've
got
something
which
is
watching
the
performance
of
HTTP,
and
hopefully
we
could
then
like
HTTP
to
could
somehow
be
part
of
that
because
that's
clearly
an
important
area
because,
like
you
know
having
something
that
says
well,
we
know
HTTP
is
a
functional
area.
Let's
watch
the
performance.
Thought
is
good.
B
C
I
mean
you
know
we
can
we
can
sort
of
trim
it
after
a
while
I,
don't
know
I,
you
know
it's
yeah,
I,
guess
I
guess
if
some
things
do
not
receive
a
lot
of
attention.
You
know
like
like
an
overhaul
or
something
is
usually
those
are
the
things
that
that
tend
to
affect
performance
right.
But
if,
if
the
code
hasn't
been
touched
in
years,
then
I
think
we
can
safely
drop
the
benchmark
and
and
in
in
favor
of
the
ones
that
that
are
that
are
more
in
flux.
C
Right,
so
that
that
may
be
one
aspect
and
and
the
other
one
is
that
you
know
we
could
make
it
sort
of
I.
Don't
know
if
are
a
requirement,
but
about
a
very
strong
suggestion
or
for
for,
like
any
you
like
a
strategic
initiative.
You
know
a
major
new
feature
to
also
include
benchmarks
and
then
and
then
they
will
be
sort
of
featured
for
Oh
for
a
while
until
this
feature
stabilizes.
No.
C
D
D
B
B
So,
for
example,
so
these
are
it's
like
a
giving
you
trend
right
whole
time
we,
so
we
are
not
getting
at
the
exact
committed
that
one,
but
we
are
getting
some
set
one
two
month.
Okay,
this
is
how
whatever
the
changes
happened
during
the
month.
This
is
the
performance
for
any
particular
one.
If
someone
makes
a
change,
for
example,
I
go
and
change
something
zle
and
I
can
go
on
at
the
same
with
number
between
these
two.
My
check-in
is
up
accepted
it
integrated,
and
this
is
the
performances.
It's
gone
bad
or
it
is
better.
B
So
people
do
people,
they
choose
what
they
are
interested,
so
we
provide
all
of
them.
They
said
what
would
be
important
for
them.
James
might
be
interesting
how
the
http/2
is
doing
right
onto
month,
for
example
or
I
might
be
interesting
how
the
huge
payers
are
doing
for
from
1
to
right.
So
I
guess
you.
We.
B
C
We
can,
we
can
also,
maybe
maybe
have
a
look
at
well.
I,
don't
know,
do
we
I,
don't
know
if
we
can
establish
a
clear
mapping
between
between
a
source,
source
directory
and
benchmark
directory,
because
we
could
look
at
what
files
were
touched
during
that
month
and
that
just
no
update
some
of
the
benchmarks
that
could
be.
That
could
be
an
AI
complete
exercise.
So
so
I'm
not
sure
yeah.
A
But
I
mean
I,
think
I
think
these
are
all
good
things
to
explore
and
like
if
we
can
sort
of
write
down
what
we
think
would
make
sense.
You
know
we'll
work
through
the
challenges
and
like
the
idea
of
running
at
once
a
month.
If
we
think
that,
like
having
the
data
available,
you
know
in
a
text
form
once
a
month
is
useful,
then
you
know
that
could
that's
probably
doable
right,
yeah.
C
C
A
C
But
but
if
if,
if
people
in
the
process
of
committing
some
some
change,
they
are
aware
that
they
need
to
they
need
to
check
the
performance
and
then
maybe
we
can
advertise
that,
please,
you
know
fire
off
this
job
in
addition
to
the
sky
job
you
know,
then,
then
these
additional
points
will
show
up
on
some
of
the
charts.
But
no
the
month,
the.
A
C
C
A
B
B
Too
much
data
right:
what
did
I
yeah?
So
what
that's?
Why
I
thought
do
you
fish
do
once
a
month
and
we
so
with.
Let's
divide
this
2
into
2
Pro.
No
one
is
actually
running
something
when
someone,
so
we
have
some
data,
yeah
again
would
be.
How
do
we
present
yeah?
Do
we
then,
let's
divide
into
two
problems
and
then
first
one
it
easy.
We
can
always
write
this
one
script
and
which
runs
all
the
workloads
every
first
of
a
month,
whatever
the
committee's
and
the
second
problem
will
let's
work
on
that?
C
Yeah,
it's
actually
like
a
visualization
problem
too,
because
you
know
if
we,
if
we
look
at,
if
we
look
at
even
a
single
benchmark,
like
let's
say
we
look
at
hash
string
creation,
it
has
been.
Creation
has
has
like
about
200
parameters,
that
it
checks
and
for
each
one.
It
produces
a
data
point
you
know,
like
you
know,
you
run
with
512
bytes
1024,
bytes,
2,
megabytes
right.
You
know
new
API,
the
old
API,
it's
a
combinatorial
explosion.
At
that
point,.
C
A
A
So
like
that's,
why
you
know
I've
always
thought
if
we
have
a
if
we
could
boil
those
down
to
a
few.
You
know
a
few
numbers
that
you
could
see
a
big
change
in
then
you
might
know
to
go.
Look
for
something,
but
if
it's
like
500
numbers,
you
know,
500
charts,
you
know
and
they
go
up
and
down
is
the
other
challenge.
It's
like
they're,
not,
you
know
super
consistent,
so
you
really
want
to
see
something
that
says
the
trend
is
not
going
down
or
whatever
can.
C
We
can
we
I,
like
looking
at
the
charts
that
we
have
I,
see
that
most
often
they
tend
to
sort
of
they
took
before
form
this
sort
of
not
EE
sort
of
line
moving
forward.
C
So
so
they
points
are
clustered
and
and
and
they
are
moving
in
X
or
before
we
were
as
the
time
progresses
right
and
so
can
we
can
we
not
a
write,
a
tool
that
recognizes
when
there's
a
significant
dip
or
so
I
mean
you
know,
like
you
know,
if
there's
a
if
there's
a
you
know
if
it
exceeds
I,
don't
know
how
many
Sigma
of
deviation,
then,
then,
then,
you
know
issue
an
alert
or
it's
a
chart
or
something
like
that.
That.
A
Would
definitely
be
useful,
it's
it's
something
that
I
always
have
in
mind.
We
just
doesn't
really
had
anybody
did
to
do
it
it
because
I'm
good.
That
was
the
other
challenge
I
was
gonna.
Throw
is
that
if
we
look
at
the
current
numbers,
you
know
it
almost
looks
to
me
like
we're
lower
on
master
for
a
number
right
yeah,
and
we
haven't
really
had
anybody
watching
it
like
DCIS
pops.
A
C
A
C
C
All
right
thanks,
okay,
yeah
so
I
mean
yet
doing
that
to
time
doing
that
once
a
month
doesn't
sound
like
such
a
bad
idea
and
at
least
at
least
recording
the
results.
You
know
for
future
analysis,
analysis
tools.
Now
we
can
write
them
as
we
collect
data,
because
you
know,
there's
gonna
be
a
chicken-and-egg
problem.
You
write
the
analysis
tool
and
it
has
no
data
to
work
with.
Oh
so.
B
I
just
want
to
I
just
want
to
shout
that
give
a
shout
that
okay,
because
he
has
some
deep
and
good
understanding
of
some
visualization
tools.
So
that's
why
I
said
list.
If
you
divide
this
into
two
problems
is
just
running
it
at
least
having
the
data,
and
then
we
what
how
it
can
present
best
yeah
and
I'm
hoping
Octavian's
knowledge
of
different
tools
he
has
used.
We
can
probably
take
advantage
of
that.
What
do
you
think
Octavian
thank.
D
B
D
D
I
think
I
found
an
image,
so
I
don't
know
how
to
share
here.
Share
the
screen
the
screen
share.
So
do
you
see
my
my
so?
What
I'm
thinking
is?
Maybe
this
type
of
coordinates?
There
are
more
more
advanced
visualizations
than
this,
but
actually
a
n
dimensional
point.
It's
it's
actually
a
line.
For
example,
if
you
follow
this
line,
do
you
see
my
mom's
mm-hmm.
A
D
This
line,
actually
it's
an
N
dimensional
point,
so
there
can
be
different
features.
I,
don't
know,
say
pal
with
sepal
length
that
I'll
being
petal
length
or
I.
Don't
know.
I
think
this
is
something
for
floss
or
something
like
this,
not
in
any
way
it's
possible
to
characterize
different
behaviors,
there's
also
tool
that
is
called
I.
D
Don't
remember
no,
but
it's
a
tool
that,
for
example,
we
can
we
can
mark.
We
can
mark
a
red
zone,
for
example
the
blue
region,
and
then
everything
gets
colored
in
blue
think
about
everything
is
in
one
color
and
then
with
blue.
We
separate
some
things,
so
some
stuff
like
this
scan,
can
bring
a
lot
of
understanding.
Maybe
even
if
you
put
a
lot
of
data,
we
can
understand
fast
yeah.
D
Another
feature
that
can
be
done
interactively:
we
can
change
axis.
So
basically
we
can
put
dogs
that
are
less
relevant.
Let's
say
all
the
lines
are
parallel.
Maybe
it's
less
relevant
there,
but
if,
but
if
there
are
some
axes
which
are
important,
then
we
can
zoom
in
so
this
axis
can
be
moved
in
some.
Some
of
these
types
of
tours.
B
A
B
A
D
A
C
B
A
B
C
Still
think
I
still
think
that
it
might
be.
It
might
be
useful
to
to
parameterize
the
job
by
by
a
benchmark
like,
for
example,
by
directory,
because
we
have
33
directories
right
and
so
so
that
so
that
the
collaborator
who
who
lands
a
PR,
that
is,
that
has
the
potential
of
influencing
the
performance
of
certain
benchmarks
or
the
numbers
on
certain
benchmarks.
That
collaborator
should
have
the
option
of
firing
off
a
job
that
normally
only
runs
once
a
month
for
every
at
the
moment.
C
A
A
A
I
know
there
was
something
that
Anna
wanted
to
update
to
it,
which
didn't
quite
get
done
so
I'm
sure
if
you
found
that
issue
and
help
with
that
she'd
be
happy.
I
thought
last
time,
I
was
waiting
for
her,
but
we
kind
of
go
back
and
forth
where
she's
waiting
on
us
to
do
something
and
then
I
ask
a
question
or
whatever
so
anyway,
but
that
lets
you
basically
say:
okay,
I'm
gonna
run
a
subset
of
the
benchmarks
and
get
the
results.
C
A
A
Small
sequel
database
on
the
bench
on
the
original
benchmarking
machine,
nothing
very
sophisticated
where
we
basically
just
push
in
like
there's
a
chart
ID,
you
should
probably
read
through
like
under
in
the
benchmarking
repo,
there's
a
whole
description
of
what
we
have,
that
you
know
how
the
current
charts
and
stuff
are
generated
and
that
kind
of
yeah
I'm
not
familiar
with
that.
Yes,.
C
D
A
A
C
Yeah
yeah
I
I
can
I
can
I
can
definitely
read
up
on
how
the
infrastructure
is
done
and
yeah
we
can.
We
can.
We
can
start
with
the
same
infrastructure
and
then
and
then
and
see
if
it
scales
up
and
if
it
does,
we
can
reuse
it.
If
not,
then
we
need
something
completely
yeah,
but
this
okay
sounds
like
a
lot
of
fun.
Yeah.
A
There's
certainly
lots
of
different
I
mean
and
I
still
think
it's
worth
considering
it,
and
as
we
go
through
this,
you
know
running
it
once
a
day.
We
may
get
more
insight
into
well.
How
could
you
could?
Could
you
come
up
with
instead
of
a
couple
hundred
like
five
hundred
numbers
come
up
with
one
with
33
numbers,
which
are
the
33
categories
right,
yeah.
C
A
That's
immediately
more
consumable
or
you
know,
maybe
you
go
and
look
at
one
chart
if
you
change
something
in
HTTP
and
it's
like
okay,
well,
that
chart
isn't
off
totally
broken.
So
we're
okay,
like
that.
Maybe
wishing
for
something
you
can't
get
to,
but
that's
kind
of
always
what
we
sort
of
thought
would
be
really
nice.
C
Well,
if
we,
if
we
did
that
correct
me,
if
I'm
wrong
for
for
each
category,
we
would
basically
have
to
take
each
benchmark
in
that
category
and
sort
of
a
sign
of
the
weight.
Yep
see
how
much
it
can
pull
down
the
whole
thing
or
or
or
push
up
the
whole
thing
exactly
so
coming
up
with
those
weights
is
gonna,
be
quite
a
thought-provoking
exercise.
I
suspect,
yeah.
A
Yeah
and
it's
then
like
how
much
does
each
one
of
those
things
vary
right
like
if
they
don't
vary
at
all
and
you
waited
them
in
some
way
that
you
know
I,
don't
know
just
want
to
be
recognizable
when
there's
a
change
right,
yeah,
like
is
even
from
the
existing
charts,
you
can
see,
there's
a
fair
amount
of
variation
up
and
down
yeah.
C
A
A
B
A
A
C
A
Okay,
so
that
sounds
good.
So
that's
our
next
step
on
that.
One
look
forward
to
seeing
that
I
think
it
should
be
good.
The
next
one
is
updating
the
charts
for
node
12x
yeah.
That
is
something
anything
is
simple
to
do.
I
think
I
can
do
it.
Okay,
so
do
you
have
all
the
information
you
need
to
do
that
you.
B
A
A
A
B
B
So
maybe
we
should
just
revert
back
to
the
change
and
try
and
night
I
don't
know
I
mean
how
so
is
that
the
bane
spike
one
correct
and
that's?
My
next
question
was
because
what
it
does
is
you
do
a
require
and
basically
get
kind
of
cache,
the
name
of
the
module
and
and
what
we
do.
Is
we
undefined
that
entry
into
the
cache,
then
that
is
the
one
of
the
benchmarks,
so
I
don't
know.
Is
that
the
correct
way
to
remove
that
cache
entry.
B
A
A
B
A
B
B
A
B
A
A
A
A
B
A
Whether
we
want
to
do
it
on
the
live
job
or
you
know,
experiment
with
something
else,
we
could
figure
that
out,
but
even
as
part
of
like
creating
the
one
for
twelve,
maybe
that's
you
know.
If
you're
gonna
have
time,
you
could
set
up
twelve
to
use
the
new,
the
two
new
things,
and
then
we
could
back
port
that
to
the
other
ones.
Okay,.