►
From YouTube: NodeJS @ Intel: Innovations in Open Source by Suresh Srinivas, Principal Engineer, Intel Corporation
Description
NodeJS @ Intel: Innovations in Open Source - Suresh Srinivas, Principal Engineer, Intel Corporation
Node.JS usage is evolving rapidly in the Data Center and the IoT. Intel is actively working in the open source community on Workloads, Infrastructure and Runtime Optimization. In this session, you will learn about the latest Intel contributions to the NodeJS community, and how through openness and collaboration – we are helping advance and make NodeJs a high performance runtime. Our team has a long history of contributions to v8 JavaScript engine, part of Node.js as well as emerging contributions in Node.js core and libraries. At this Node.js Interactive conference, we are releasing a Node.js workload suite for the data-center in the open source for the community to evaluate NodeJS on different platforms. We have also developed the Language Performance Portal for the community to track NodeJS performance. Join us and learn more!
About Suresh Srinivas
Anything about Node performance, deployment, benchmarks or pain points
A
Thank
you.
Everybody
looks
like
they
give
me
a
really
big
room
here:
okay,
I'm
Suresh
famous,
you
can
follow
me
on
connect
with
me,
I,
Tara,
Sh,
and
today
what
I'm
gonna
talk
to
you
about
this,
our
work
at
Intel
on
on
nodejs
this
morning,
they've
shared
with
you
at
his
keynote
people
who,
when
they
came
to
the
booth,
they
were
asking
what
is
Intel
doing
it
with
Nord.
A
So
the
main
thing
that
we
are
focusing
on
us
is
about
performance
and
performance
is
really
central
to
Intel's
DNA
at
the
level
of
the
software
at
the
level
of
the
CPU
at
the
level
of
the
platform,
and
we've
been
essentially
doubling
our
performance
every
18
months
or
so
using
using
Moore's
law.
So
we'll
share
with
you
some
of
the
work
that
we
are
doing
there.
A
A
So
intel
has
been
active
in
managed
runtimes
for
the
last
10
years
and
they've
showed
you
this
in
the
morning.
So
we
have
been
working
on
java
right
from
the
early
days
of
java,
and
this
is
showing
you
on
the
spec
jbb
2005
benchmark
with
both
the
hardware
optimizations
and
the
software
optimizations.
A
So
these
are
the
three
parts
of
my
talk.
So
I'll
talk
about,
leading
which
they've
mentioned
about
servant
leadership
with
with
a
portal
that
we
have
and
then
I'll
talk
to.
You
also
workloads
and
find
it
close,
but
trying
to
link
the
all
of
it
together.
So
there's
language
performance
portal
that
we
announced
this
morning.
So
this
is
available
for
all
of
you,
and
this
is
a
service
that
Intel
is
providing
to
all
of
you.
A
A
So
this
is
the
this
is
the
portal
and
by
default
it
is
showing
you
data
about
the
Acme
air
workload
which
we,
which
we
run,
and
you
can
look
at
like
other
workloads,
also
and
see
how
performances
is
going
and
then,
if
you
want
details-
and
you
can
sign
up
on
the
mailing
list
for
this-
it's
also
sent
on
the
developer
mailing
lists.
So
every
night
we
publish
these
results
and
then
you
can
go
inside
and
look
so.
This
is
these.
A
Are
the
results
for
the
node.js
master
that
was
built
on
the
29th
like
this
morning
and
what
is
the
commit
and
then
what
CPU
did
we
run
this
on
right
and
what
is
this
compared
against
this
compared
against
the
V
4.2.2
with
that
commit,
and
then
it'll
tell
you
for
each
of
the
workload.
What
is
the
standard
deviation?
A
Was
we
had
a
buffer
index
off,
which
is
one
of
the
micro
workloads
that
showed
some
regression,
so
we
fixed
it,
and
we
got
some
performance
back
on
that
and
then
we
looked
at
like
the
improvement
that
Acme
air
got
and
we
isolated
it
down
to
v8
engine
that
improved
its
performance,
and
then
we
found
some
regression
and
Acme
here
and
we
contributed
to
the
community.
So
this
is
the
language
performance
portal,
so
please
check
it
out
right
and
we
are
also
very
open
to
workloads
to
add
to
this
portal.
A
So
if
your
use
case
is
not
being
represented
in
the
portal,
do
work
with
us
and
we
would
love
to
support
you
and
make
sure
that
your
workload
or
your
use
case
is
actually
handled.
We
have
some
open
source
workloads
like
let's
chat
and
the
ghost
blogging
platform
that
we've
been
working
with,
and
we
intend
to
add
a
few
of
those
too
to
this
suite.
So
please
check
it
out
and
such
as
workloads
to
us.
A
Ok,
so
next
I'm
going
to
move
from
the
language
performance
portal
to
workloads.
So
this
is
our
BCG
vision.
This
is
the
data
center
group
at
Intel,
so
our
vision
as
the
future
is
going
to
have
thousands
of
clouds
with
millions
of
services
connecting
billions
of
devices
there
may
be
micro
services,
like
you,
saw
this
this
morning
and
all
kinds
of
cloud.
It
would
be
like
Azure
cloud
and
we're
working
very
closely
with
Microsoft
on
some
of
these
workloads,
and
it
could
be
lots
of
other
other
clouds
too.
A
So
this
is
an
open
source
workload
and
it's
it's
a
series
of
workloads
that
we
want
to
build
using
different
customer
feedback.
We
have
heard
from
customers
that
server-side
rendering
with
reactant
node
is
very
important
to
us.
We
intend
to
add
that
and
we
intend
to
add
other
use
cases
and
you
can
also
contribute
right.
So
this
is
an
open
source
workload,
so
you
can
submit
pull
requests
or
you
can
fork
and
you
can
add
your
favorite
kind
of
like
workload.
A
The
rigor
of
workload
that
we
have
brought
to
spec
will
expect
gbb
2005.
We
are
trying
to
bring
to
the
node
community
so
that
you,
you
also
have
a
very
precise
measuring
period
and
a
standard
like
warm
up
and
also
reporting
on
the
score,
and
also
what
is
the
latency,
the
99th
percentile
of
the
latency,
all
these
kinds
of
like
details
that
the
workload
will
actually
print
for
you
at
the
end
of
the
workload
run.
A
So
what
what
does
this
workload
have?
In
it
today,
it
has
one
use
case,
which
is
an
employee
information
service,
so
we
are
currently
using
MongoDB
as
the
database,
but
you
could
use
other
data
is
to,
and
we
have
three
parts
of
the
workload
one
is
called
the
monolithic
mode
where
you're
really
running
on
one
CPU
right,
then
we
very
often
see
that
lot
of
workloads
that
people
are
deploying
or
using
just
a
single
CPU,
and
we
do
encourage
every
one.
That's
running
on
servers
that
have
several
course
right.
A
Without
you
making
a
lot
of
code
changes
to
your
to
your
application,
and
then
we
also
have
a
micro
services
mode
which
allows
you
to
restructure
this
workload
into
a
set
of
micro
services,
so
that
you
can
actually
see
how
the
performance
of
that
is
and
all
these
are
driven
through
like
a
client
and
there's
also
something
that
we're
contributing
to
the
community.
It's
a
multi-threaded,
Python
client
that
drives
because
things
like
Apache
bench,
we
felt
was
not
like
adequate
for
driving
this,
and
so
we
developed
this
this
client
for
you.
A
Okay,
how
does
this
perform?
So
this
is
not
yet
1.0
it's
0.9,
so
we
want
to
give
some
time
for
the
community
to
give
feedback
and
to
also
take
input
from
a
variety
of
other
stakeholders.
So
this
I'm,
showing
you
this
first
graph,
which
is
a
generation
to
generation
speed-up,
so
Intel
using
our
tick-tock
methodology.
We
have
been
creating
like
new
products
every
year
right
and
these
products
are
Xeon
servers
and
our
newest
product
is
the
Broadwell
EP
server.
A
It's
a
two
socket
server
that
44
course
and
then
our
previous
generation
is
has
full
server,
and
you
may
not
know
all
this
distinction,
but
if
you're
clearly
running
on
the
cloud,
I
encourage
you
to
find
out
from
your
cloud
provider
what
what
CPU
is
this?
We
are
seeing
customers
that
are
running
four-year-old
systems
right,
whereas
we
have
actually
doubled
or
quadrupled
the
the
performance
in
the
last
four
years.
So
this
is
showing
this
node
DCE
is
workload
performing
really
well,
it's
showing
like
30
percent
improvement,
and
you
have
22%
more
course
right.
A
It
moved
from
36
cores
to
44
course,
so,
22
percent,
more
coarse,
but
you're
getting
almost
30
percent
speed-up
and
on
a
single
CPU.
We
are
getting
like
20
percent
speed
up
with
our
newest
hardware
and
our
next
generation
product
that
will
come
out
next
year
is
going
to
do
even
better
than
this
one,
and
this
is
kind
of
like
showing
how,
if
you
are
deploying
just
a
single
node
instance
versus
you,
are
actually
deploying
Lawrence
instances
on
all
the
cores
that
are
actually
available.
What
kind
of
performance
can
you
actually
expect?
A
A
We
have
44
cores
and
you're,
showing
a
43
X
improvement,
going
from
a
monolithic
to
a
clustered
mode,
and
then
the
last
I
want
to
share
with
you
is
what
about
the
software
right.
So
the
software
continues
to
evolve,
and
so
we
ran
just
last
week
on
the
node.js
software,
the
latest
one,
six,
nine
one,
and
we
saw
that
there's
a
twenty
four
percent
improvement
right,
which
is
really
good
to
be
able
to
adopt
the
newest.
A
A
A
So
how
can
we
use
something
sort
of
like
faster
there,
and
so
has
a
library
called
the
DPD
K
library,
which
is
the
data
plane
development
kit
that
allows
the
internet
to
directly
write
into
the
into
the
caches
right
without
needing
to
do
a
kernel
switch?
And
then
you
can
actually
use
a
user
level.
Tcp
stack,
and
so
we
have
a
prototype
of
that
and
we
would
love
to
work
with
you
to
contribute
that
to
the
to
the
community
to
solve
some
use
cases.
A
We
think
that
it
can
have
really
good
performance
improvement,
and
so
that's
a
new
innovation
that
we
are
doing
and
we
want
to
share
it
with
you
and
the
community.
We
were
hoping
to
have
the
code
already
so
that
you
could
play
around
with
it,
but
I
look
for
that
very
soon.
Okay,
so
summary
is
that
we
are
committed
to
the
optimization
of
the
node
stack
right
across
a
single
platform
and
also
on
the
cloud
and.
A
So
our
call
to
action
is
to
visit
our
booth
and
work
with
us
to
accelerate
node
on
ie
and
I.
Have
a
couple
of
minutes
so
I
can
take
some
questions.
So
thank
you
all.
If
you
have
a
question,
you
can
come
up
to
the
microphone
here
and
and
ask
your
question.
So
this
is
no
js'
at
Intel.
Innovations
in
the
open
source
and
I'm
Suresh
in
was
my
twitter.
Handle
is
at
s
Suresh.
Sorry,
it's
not
a
Shiva
I
desperation.
A
B
A
A
It
does
not
preclude
because,
as
long
as
the
network
driver
can
be
virtualized-
and
we
have
the
DP
DK
team
has
worked
with
things
like
the
Zen
virtual
machine
to
also
enable
like
those
Knicks
right,
literally
virtualized.
So
but
we
have
not
actually
tested
that
out.
That's
one
of
the
feedback
we've
gotten
and
we'll
that's
something
that
we
will
look
at
addressing,
but
it
should.
There
is
no
technical
challenge
that
we
need
to
solve.
Okay,.