►
From YouTube: WebPerfWG call - February 27th 2020
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
Otherwise,
regarding
the
face-to-face
I
sent
out
invites
for
industry
a
bunch
of
industry
folks
today
we'll
see
what
what
the
turnout
will
be
and
what
will
be
the
responses.
I
think
I
have
sent
invites
as
well
to
everyone
who
the
regular
here,
but
if
you
haven't
received
one
please
let
me
know
and
I'll
rectify
that.
C
C
A
The
next
meeting
is
marked
well
two
weeks
from
now
same
time:
okay
and
then
I
sent
out
invites
for
the
basically
for
the
first
day
of
the
face-to-face
earlier
today
for
April,
2nd
in
the
Bay
Area,
so
basically
sent
out
RSVPs
for
folks
for
industry.
Folks,
everyone
who's,
a
regular
here
could
be,
you
know,
should
also
be
an
N
by
list.
A
Okay
and
I
think
that,
with
that
we
can
move
on
to
proposal
than
we
have
on
the
agenda
today.
One
proposal
from
the
deck
about
the
browser
ablation
though
oh
just
just
before
that
in
terms
of
thriving
I,
can
probably
describe
for
the
first
part,
but
I
would
love
someone
described
on
the
second
part
of
the
call
where
I'll
be
talking
about
timing
law,
origin
and
why
things
are
that
so
I
can
inscribe.
A
C
D
A
Stations
yeah.
Obviously,
if
there
is
some
sort
of
you
know,
restriction
on
travel
that
will
have
an
impact
I
think
that,
right
now
we
can
plan
on
it
not
being
there
and
then
see
how
things
evolve.
But
obviously,
if
there's
a
health
risk,
we
will
have
to
replace
it
with
some
online
conference
thing.
E
A
F
C
C
A
H
G
A
D
One
recommendation
here
is
to
invest
and
investigate
what
we
can
do
online,
even
just
as
a
backup
option.
Right.
Let's
say
we
cancel
the
actual.
If,
if
we
have
to
cancel
the
event,
one
option
is
to
defer
an
entirely
and
other
options
to
investigate
if
we
can
actually
just
host
an
online
version
of
it,
maybe
in
a
slightly
different
format,
so
we
should
run
through
those,
and
also
another
option
is
maybe
it
is
a
case
that
some
people
are
unable
to
travel
due
to
the
ban,
in
which
case
we
have
reduced
in
person
participation.
D
C
I
think
the
situation
is
not
just
to
possibly
the
people
may
not
be
unable
to
travel
right
like
it.
There's
also
the
very
serious
like
realistic
scenario,
which
maybe
one
of
them
who
having
been
in
contact
with
somebody
who's
contracted
virus,
in
which
case
my
spread
the
virus,
and
that
would
be
very
irresponsible
thing
to
do
so.
I
think
I
mean
obviously
that
probability
is
very,
very
low.
The
current
estate
I
think
we
have
to
monitor
and
evaluate
the
situation
as
the
meaning
approaches.
G
H
A
Impression,
theoretically,
I
can
imagine
us,
instead
of
doing
two
days
of
you
know,
presents
physical
presence.
We
can
have
four
half
days
in
ways
that
can
be
inclusive
to
various
time
zones.
So,
even
if
some
people
have
to
stay
route,
they're
like
evenings
or
their
early
mornings,
it's
just
four
hours-
it's
not
eight
hours.
I
think
this
is.
C
Yeah
I
think
another
constellation
is
to
my
god.
Situation
like
a
lunch
right
I
could
eating
things
in
a
group
with
everyone
involved.
This
also
raises
the
risk
of
contraction,
because
the
way
most
viruses
contract
or
at
least
of
this
device
contract
is
through
the
droplets.
So
in
your
field,
eating
and
talking
at
the
same
time,
that's
very
good
way
to
contract
the
virus
or.
A
I
What
the
right
magnitude
of
that
investment
should
be
relative
to
where
you
are
in
the
performance
spectrum.
There's
also
a
lot
of
variation
that
we've
observed
between
sites.
Some
sites
are
very
sensitive
to
kind
of
a
first
content
will
paint
experience.
They
want
to
get
sort
of
photos
or
media
rich
content
shopping
sites
when
I
get.
You
know
beautiful
imagery
of
their
products
in
front
of
the
user
to
entice
them
to
engage
more,
whereas
other
sites,
where
there's
a
little
bit
more
interaction,
are
more
sensitive
to
things.
I
Like
first
input,
delay,
NLCS
and
so
I
think
understanding
both
what
the
motivation
is
for
an
individual
property
to
invest
in
speed
and
also
customizing,
that
understanding
for
individual
site
characteristics
is
kind
of
the
focus
here.
The
specific
mechanism
with
there's
details
in
the
proposal
is
essentially
a
very,
relatively
simple
mechanism
in
the
browser,
but
intended
to
be
exploited
in
a
more
sophisticated
way
by
analytics
providers
hosting
platforms
and
also
the
more
sophisticated
sites
mechanism.
I
Focused
on
getting
ablation,
which
is
to
say
artificially
inserted
delays
as
close
as
possible
to
the
actual
metrics
so
that
the
correlation
can
be
can
be
more
closely
established.
So,
rather
than
inserting
you
know
arbitrary
sleeps
in
your
JavaScript
code,
we
would
actually
just
go
ahead
and
have
the
browser
intentionally
delay
the
first
paint
or
intentionally
delay,
first
input
and
and
and
my
kind
of
that
experimental
basis
to
say
if
your.
I
I
Yeah,
it
would
actually,
when
you're,
doing
an
ablation
study
you,
you
apply
a
change,
you
kind
of
worsen,
something
by
a
small
measure
for
a
small
group
of
users
to
see
how
sensitive
the
metrics
you
care
about.
Like
conversion
rates
are
to
the
to
to
delays
in
performance.
We
can't
artificially
speed
something
up,
so
we're
constrained
in
the
direction
that
we
can
move
when
doing
these
experiments.
But
it's
a
tended
to
be
a
small
change
for
a
relatively
smaller,
but
that
will
be
up
to
the
analytics
provider
or
the
individual
site.
I
So
some
of
them
can
be
done.
It
can
be
simulated
in
in
in
script,
but
some
of
them
are
there's
two
main
reasons.
One
is
for
example.
If
you
want
to
delay,
you
know
loading
of
the
first
JavaScript
payload
and
that's
something
that
can't
be
specified
or
simulated
in
JavaScript
itself
and
the
other
one
is
that
some
of
these
metrics
are
difficult
to
affect
directly
in
measurable
ways.
Through
script,
there
is
able
to
actually
just
two
legs.
I
We
put
the
browser
already
owns
and
is
responsible
for
then
we
can
say
when
you
specify
15
milliseconds
away
the
first
paint.
The
browser
will
actually
delay
first
main
life.
If
and
that
kind
of
when,
when
mapping
helps
with
the
statistical
validity
of
the
experiment
and
reduces
the
amount
of
ablation,
they
need
to
apply
to
get.
I
B
B
I
I
think
the
main
points
of
feedback-
and
this
is
helpful-
is
to
understand
you
know
what
what
aspects
of
performance
makes
sense
to
make
possible
by
this
experimental
API.
If
folks
are
interested,
like
you
said,
in
memory
footprint
in
loading
of
JavaScript
other
things
that
might
be
relevant
to
CD
ends.
If
there's
certain
areas
that
we
want
to
explore,
the
Browns
are
assisting
mining
really
solid.
They
should
values
marking,
that's
something
that
we
can
add
to
them.
Yeah.
B
B
A
The
industry
folks
have
any
thoughts,
Stephen
yeah,
yeah.
J
I
was
thinking
so
for
us
for
Salesforce
I.
Don't
think
it
would
be
a
useful
API
for
the
use
case,
unaware
of
because
we
we
don't
have
the
congressional
Eric's.
Like
you
know,
a
retail
website
would
have
you
know
people
use
our
software
to
do
their
tasks.
So
you
know
if
we
delay
their
yeah
their
user
interaction.
It
would
be
hard
to
know
what's
going
to
change
other
than
this
game.
They're
going
to
be
more
unhappy.
J
I
I
think
the
intent
is
to
treat
kind
of
a
speed
metrics
as
the
dependent
variable
they're,
usually
not
that
you're
the
output
right.
But
if
we
change
them
as
an
input,
we
you
need
some
new
output
variable
to
to
then
observe
as
results
and
so
I
think
is.
This
is
intended
for
analytics
providers
that
that
have
that
other
variable
in
terms
of
conversion
or
add
to
cart
actions.
E
H
That
sounds
interesting,
I
think
for
us.
It
could
also
be
useful
for
even
synthetic
testing
figuring
out
the
effect
of
certain
things
on
the
ladder
matrix
yeah.
This
is
definitely
something
we
would
play
with,
because
we've
been
talking
about
many
years
about
trying
that
out
in
production.
You
know,
if
anything's
on
the
purpose,
to
reproduce
the
these
over
studies
that
some
folks
have
done
and
just
to
certainly
make
it
a
lot
easier
to
do
with
the
stock
that
so
thumbs
up.
For
me,.
J
H
J
H
If
we
can
apply
it
to
a
lot
of
different
metrics,
what
would
be
interesting
that
what
might
come
out
of
it
is
that
some
might
be
more
important
than
others
like,
for
example,
first
of
the
delays
fairly
new.
We
don't
know
how
important
that
is.
You
know
we're
going
to
start
measuring
it.
It
doesn't
apply
to
every
page,
but
you
know
if
you
start
playing
with
these
and
being
able
to
make
them
worse,
I
realize
okay.
H
Instead
of
spending
all
our
efforts
on
making
faster
paint,
maybe
he's
actually
a
lot
more
reward
to
focus
on
on
events
and
then
first
upon
delay.
So
for
us
we'd
be
a
matter
of
directing
our
focus
and
resources
better.
We
can't
make
like
increasingly,
we
have
more
and
more
engineers
pushing
features,
and
we
have
to
make
tough
choices
about
what
we
care
about
and
at
the
moment,
I
couldn't
really
say
which
performance
metric
matters
most,
especially
since
a
lot
of
them
are
codependent.
H
H
H
H
It's
like,
should
we
focus
on
improving
a
medium
or
who
just
look
at
the
extreme
scenarios,
like
maybe
that
0.01%
of
easier
experience,
then
maybe
that
makes
them
leave
and
and
something
like
this
would
allow
us
to
have
more
data,
because
by
definition
the
extreme
doesn't
happen
very
often,
and
so
maybe
you
don't
have
enough
data
to
draw
those
conclusions.
But
if
you
can
trigger
the
extreme
for
more
people,
then
you
can
figure
out.
Ok
like
this
is
really
a
situation
we
need
to
take
care
of.
H
This
is
where
this
is
the
threshold
beyond
which
people
are
disengaged
and
the
development
is
very
hard
for
us
to
figure.
It
out.
I
think
something
like
this
might
shift
the
focus
from
looking
at
the
higher
percentiles,
the
more
unusual
cases,
instead
of
always
trying
to
make
the
median
faster,
which
two-degree
happens
itself
due
to
the
incentive
and
faster.
So
maybe
it's
again
like
this
is
what
most
people
tend
to
focus
on,
but
it
might
not
be
where
most
of
the
rewards
could
be
so
yeah.
A
Thanks
anything
else,
any
other
comments,
or
can
we
move
on
ok,
so
thanks
the
deck
I
hope
this
was
useful.
A
A
Thank
you.
So
we
have
three
issues
that
we
briefly
talked
about
last
time,
but
generally,
there's
I
think
some
fundamental
issues
here
with
a
timing,
allow
origin
that
we'll
need
to
explore
very
soon.
So
if
we,
so
one
of
them
is
related
to
the
next
hop
protocol
which
so,
if
we
look
at
issue
221
that
was
opened
by
I'm,
not
sure
if
it's
open
by
the
ping
or
open
by
folks
who
are
in
the
ping,
because
but
basically
it's
the
next
hop
protocol
exposure
is
like.
There
is
issue
221
and
then
beyond
that.
A
A
A
A
Modify
some
of
the
opt-ins
and
there
is
there's,
still
work
to
be
done
and
clarifying
exactly
which,
which
characteristics
of
each
of
the
entries
we
expose
fits,
which
opt-in
and
there
is
work
to
be
done
to
make
sure
that
we
already
started
work
to
make
sure
the
timing.
Allow
origin
is
basically
timing.
Allowed
checks
will
be
implied
by
course,
opt-ins
and
I
think
we
can
also
claim
that
they
should
be
implied
by
cork
opt-ins.
G
A
A
There
are
cases
where
next-hop
protocol
can
reveal
details
about
the
users
network.
Also,
that's
not
really
a
cross
ojung
issue
right
I
mean
it
would
be.
This
is
not
a
cross
origin
issue.
I
I
agree
that
the
concerns
that
they
raised
on
that
front
is
not
a
cross
origin
issue.
I've
seen
other
concerns
that
are
cross
origin
issue
around
service
workers
and.
C
A
So
the
concern
that
I
have
seen
is
around
telling
whether
Service
Worker
is
installed
or
not
on
a
cross
origin
by
creating
an
iframe
to
a
resource
on
that
origin
and
then
inspecting
the
next
hop
protocol.
We
had
an
issue
with
that
with
workers
start
in
chrome
a
few
months
back
and
the
same
seems
to
apply
to
next.
How
protocol.
C
A
C
A
E
A
C
A
H
A
A
C
C
So
I
think
the
only
thing
is
that
the
iframe
itself
that
could
expose
the
president
I
guess.
Another
scenario
is
if
there
are
some
sort
of
like
a
corporate
site
scenario
where
iframe
is
loading
the
corporate
site,
an
intranet
somehow
that
could
explore
I
mean
that
you
could
already
do
that
right.
Cuz
I
mean
II,
I.
Think
the.
C
A
C
A
A
A
Alternatively,
and
the
part
that
I
was
alluding
to
before
may
be
transfer
sizes,
it's
not
something
that
should
be
blocked
on
timing
in
law,
origin,
but
something
that
should
be
blocked
on
something
like
cork,
because
one
of
the
complaints
is
not
just
this
particular
issue.
But
the
fact
that
between
resource
timing,
l1
and
l2,
the
optin
remained
the
same,
but
we
added
a
bunch
of
attributes
which
expose
different
dimensions
of
the
resource.
A
A
E
C
A
A
C
C
That
I
actually
don't
know
I'll
have
to
check
that.
That
would
be.
A
A
A
C
C
A
We're
out
of
time,
but
could
you
potentially
take
an
action
item
to
look
into
the
proposed
mitigation
here,
which
is
to
eliminate
cookies
and
so
cookie
sizes
from
the
request
and
that
cookie
sizes
from
the
response
or
no,
the
request
is
not
exposed,
but
just
from
the
response,
all
I'm
gonna
set
cookie
headers
from
the
reported
sizes.
In
order
to
tackle
that
particular
issue.
C
A
C
C
Wash
ashore
at
some
point,
I
should
be
able
to
I
I.
Just
kind
of.
Let
me
see
if
I
can
do
that
by
next
to
say
next,
you
know
meaning,
but
okay,
yeah
I
will
try
to
gather
information.