►
Description
Don’t miss out! Join us at our upcoming hybrid event: KubeCon + CloudNativeCon North America 2022 from October 24-28 in Detroit (and online!). Learn more at https://kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects.
Does Green Software need Open Hardware? - Hosted by Open Compute Project - Moderator Cliff Grossner, Ph.D.; Kate Mulhall, Senior Cloud Software Engineering Manager, Intel , VP Market Intelligence, Open Compute Project (OCP) Foundation; Dinesh Marjrekar, D
By 2040 the communications technology sector it is expected to account for 14% of the world’s
carbon footprint — up from about 1.5% in 2007, and in 2019, researchers at the University of
Cambridge estimated that the energy needed to maintain the Bitcoin network surpassed that of
the entire nation of Switzerland. Clearly this cannot continue and software needs to be
evaluated by judging its performance on its energy efficiency as much as on traditional
parameters. On top of the software itself, is the IT infrastructure and data center facilities that
can make a huge difference on the ecological impact of software workloads. This panel will look
at best practices and new technology directions that can change the current and dangerous
ecological impact curve implied by the digitization of society.
A
That
is
not
actually
totally
true
at
the
ocp.
We
do
look
at
software
and
we
just
announced
a
new
strategy
at
the
ocd
around
hardware
software
coastal,
and
so
you
know
five
years
ago,
many
people
said
we
don't
care
about
the
visible
infrastructure
they're,
all
the
same,
giving
one
giving
10
million
million
doesn't
matter.
But
something
happened
about
five
years
ago
when
the
google
grain
project
took
off
and
all
of
a
sudden,
you
have
the
intml.
We
have
all
this
district.
A
B
A
Of
the
orchestration
layer,
it's
still
pretty
independent,
but
we've
started
to
have
discussions
now
about
updating
the
scheduler
for
kubernetes.
We
now
have
algorithms
in
it
that
look
at
what's
the
current
program
of
the
underlying
infrastructure
on
running
them
and
including
that
in
the
decision
making.
So
when
we
approached
cncf
about
six
months
ago,
without
wanting
to
be
here-
and
the
first
thing
chris
said
to
me,.
A
Since
we
have
quite
a
lot
of
work
going
on
in
sustainability
in
the
ocp
that
it
would
be
a
good
match
to
try
to
build
awareness
around
the
software
and
the
hardware
and
that
awareness
interaction-
maybe
maybe
at
some
point
in
the
future,
we
start
talking
about
applying
some
standardized
apis
where
software
could
then
probe
the
physical
infrastructure.
That's
running
in
the
in
the
data
center
and
understand
is
that
a
600
watt
gpu
card?
That's
in
that
server,
and
how
do
I
want
to
optimize
that
versus
something
else?
And
it's
not
stopping
there.
A
A
C
A
Take
the
innovations
that
are
coming
from
hyperscale
data
centers
because
they
have
to
compete
at
scale
like
the
problems
you
work
on,
writing
software
that
works
at
scale
and
make
those
available
in
an
everyday
product
filter
into
the
general
function.
That
is
lesson
in
that
show,
and
today
we
have
over
five
thousand
active
engineers
on
our
mailing
list.
E
Nice
so
my
my
name's
kate,
kate
malhole,
I
work
for
intel,
I'm
a
senior
cloud
engineering
manager
and
I
have
a
team.
We
have
two
focus
areas.
One
is
on
the
networking
side,
so
that's
really
about
efficient
packet
processing
and
the
second
team,
and
a
few
of
them
are
here
actually
in
the
audience,
are
on
the
resource
management
side
and
really
our
focus.
There
is
around
optimal
workload,
scheduling
and
really.
How
can
we
provide
the
energy
savings.
C
C
My
name
is
jaime
and
I
work
for
client
technologies.
This
is
a
german
company
offering
cloud
services
and
liquid
cool
solutions
and
waste
reuse
solutions
for
data
centers,
so
we
are
actually
in
both
worlds.
We
have
a
foot
in
each
of
one
of
them
and
this
is
important
to
know
actually
what's
what's
behind
your
code
guys.
So
if
we
could
maybe
rename
this
panel,
we
couldn't
say
what's
behind
your
code.
Besides
this,
I'm
active
at
the
ocp
in
the
ways
hit
reuse,
workstream
and
thank
you
very
much
for
an
invitation
for
being
here.
F
Hi,
I'm
dinesh
dinesh
madraka,
director
of
innovation
at
sivo,
we're
a
cloud
native
hosting
provider
and
we're
running
ocp
kit
in
production
in
three
data
centers
around
the
world.
We've
got
presence
in
new
york,
london
and
frankfurt,
and
we've
been
running
this
for
two
years
now:
providing
kubernetes
clusters
for
tenants
to
use.
G
A
So
we've
got
a
pretty
pretty
well
designed
panel.
I'm
pretty
happy
about
that.
I
do
have
one
housekeeping
thing
before
we
kick
on
and
kick
it
off
is
that
we
do
have
a
little
contest
over
there.
So
if
you
put
your
business
card
in
or
fill
your
piece
of
paper
with
name
and
email,
we're
doing
a
little
contest
for
an
ipad,
so
we
can
get
a
little
present
for
being
here
today,
again
appreciate
everyone
being
here
at
4
30
in
the
afternoon.
A
Day
of
conference,
so
with
that,
let's
kick
off
the
discussion
and
first
question
that
the
panel
told
me
they'd
like
to
talk
about
is
really
asking
the
question:
what
is
green
software?
What
what
could
make
software
green?
If
anyone
were
to
think
about
it,
and
so
I
think
marcelo
said
he
would
kick
that
one
out
and
not
mistake.
It.
G
Yeah,
so
green
software
for
us
would
be
like
treating
everything
in
our
data
centers
because
we
control
the
full
stack
from
what
what
does
our
cooling
system
consumes?
What
is
our
power
mix
and
also
how
we
build
our
racks?
What
we
do
with
our
network
stack,
we
control
the
whole
thing,
so
we
can
also
do
stuff
like
automatically
have
software
turning
off
racks,
which
are
needed
or
just
use
the
servers
we
really
need
and
all
the
others
can
be
shut
off,
and
something
like
that.
G
E
E
What's
the
right
performance?
What's
the
right
from
sustainability,
and
I
think
that
that
is
a
it's
slightly
different
right
when
you
start
thinking
that
way.
So,
for
example,
if
I
was
to,
if
you
were
running,
if
you
were
running
a
workload
that
needed
a
that
had
a
high
performance,
say,
for
example,
high
computational
power
for
running
a
new
vaccine,
for
example,
you
might
be
thinking
about
where
you'd
locate
it
right,
because
it's
going
to
need
a
lot
of
power,
you
don't
want
to
just
go
on
the
maximum
power.
E
You
might
be
just
thinking,
okay,
I'll
put
it
in
a
data
center
next
to
the
solar
next
to
a
solar
farm
or
something.
So
it's
really
thinking
about
what
is
the.
When
is
the
right
time
to
run
it?
You
know
what
is
the
right
time
right,
duration,
because
you
know
some
of
these
workloads
just
run
forever,
because
we
haven't
really
been
thinking.
We
haven't
been
putting
sustainability
in
in
the
forefront
of
everything
we've
been
doing,
so
I
do
think
you
know
there
is
a
paradigm
change.
E
That's
required
around
the
screen,
software
in
terms
of
doing
the
right
thing,
doing
the
right
performance
and
intel
very
interestingly
yeah.
Actually,
it
was
just
last
month
they
have
ai
really
starts
coming
into
play
here,
so
they've
they're
actually
acquired
this
company
called
granular
and
it's
a
optimized
platform,
an
intelligent,
optimized
platform.
So
it.
E
It
really
allows
you
to
get
more
out
of
the
hardware,
so
it
can
reduce
your
cpu
utilization
with
all
this
kind
of
by
observing
the
applications
and
how
they're
performing
so
it's
kind
of
really
interesting.
So
I
I
do
think
that's
kind
of
interesting
when
you're
thinking
about
you
know
right
the
right
performance,
right
sustainability
and
then
there's
kind
of
this
ai
dimension
to
it
too.
F
Up
really
allows
workload
to
be
moved
around
a
cluster
or
a
data
center
or
site,
knowing
at
a
software
level
that
everything
will
recover
and
I
think,
we've
been
afraid
at
an
operational
level,
which
is
why
we've
kept
workloads
running
and
running
and
running,
because
we
don't
want
to
get
paged
overnight,
that
it's
gone
down.
It's
gone
offline,
whereas
kubernetes
will
let
it
come
back
online.
We've
got
things
like
chaos,
engineering
which
gives
us
confidence
to
shut
things
down
and
move
workload
around.
C
C
So
we
had
to
move
actually
the
workloads
from
a
to
b,
to
z,
to
c
so
actually
in
the
base
of
our
businesses
to
schedule
this
workloads
to
to
move
them
depending
on
different
variables,
and
one
of
them
is,
for
example,
power
cost
because
we
do
waste
heat
reuse.
Another
one
is
the
needs
for
heating,
for
example,
if
we
have
the
possibility
of
moving
a
workload
to
helsinki,
then
we'll
do
that,
because
helsinki
needs
a
heating
power
long
time
of
the
year
so
and
so
on.
So
this
is
this.
C
F
That
topic
of
waste
heat
is
also
really
interesting.
Right
of
we
always
put
compute
in
data
centers,
and
then
we
pay
power
costs
to
cool
it.
If
we
could
find
other
places
to
to
keep
hardware,
could
we
put
it
and
heat
water
that
is
then
used
in
a
in
a
community
or
something
like
that?
So
it's
not
just
where
the
software
is
running
and
where
it
is
in
the
data
center,
but
understanding
where
the
data
center
is
located
and
what
else
we
can
do
with
the
energy
that
we're
using
yeah.
E
I
mean
that's,
that's
that's
a
really
good
point.
We,
we
there's
a
whole
kind
of
piece
around
emotion,
cooling,
that
intel
has
been
working
very
closely
on
last
year
with
this
company
called
supper,
and-
and
that
is
a
lot
of
that-
is
it's
for
the
next
generation
of
data
centers.
It's
this
precision,
immersion,
cooling,
fluid
cloud
that
they're
working
on
and
it,
and
it
really
is
about
that
really
trying
to
make
sure
that
the
the
energy
is
going
to
be.
E
A
One
misconception
that
maybe
to
ask
the
panel
to
address,
but
still
here
in
quarters,
you
know
if
I
try
to
be
sustainable,
I'm
going
to
have
to
give
up
on
something
I'm
going
to
lose
performance
somewhere.
I
have
to
do
this
and
I'm
wondering
if
we
can
talk
a
little
bit
about
the
direction
you're
going
in.
C
Not
really,
actually,
you
can
improve
performance,
sometimes
because,
with
liquid
cooling,
you
can
actually
push
the
the
hardware
to
to
follow
boundaries
in
the
temperature,
something
you
cannot
reach
with
actually
with
the
with
classical
cooling,
just
because
of
physical
constraints
of
of
the
air
water.
It
has
three,
I
think
it's
3
500
times
more
heat
capacity
as
as
air,
just
because
of
its
density,
and
it's
a
it's
a.
I
think
the
the
term
is
a
heat.
C
G
So
so,
in
our
platform,
I
think
we
can
have
a
lot.
We
have
a
lot
of
potential
to
save
energy
and
also
reduce
the
consumption
our
whole
company
has-
and
I
also
see
like
stacks
for
let's
say
developers
which
can
then
also
show
hey.
I
have
that
code
line
here.
That's
combining
or
using
a
lot
of
cpu
cycles,
burning
them
because
I'm
just
polling
stuff,
and
then
you
have
some.
G
G
So
for
us
that
would
be
a
good
tool
to
have
so
it's
nothing.
We
don't
have
a
tool
like
that.
It's
just
yeah
yeah,
so
yeah.
We
can
only
monitor
that
on
our
platform
that
we
now
consume,
like
20
kilowatts
more
than
before,
or
something
like
that.
But
we
will
not.
We
currently
don't
have
that
insights,
but
which
would
be
great,
so
you
can
directly
see,
oh,
that
app
got
updated
now.
It
consumes
like
50
kilowatt
more
than
before
would
be
nice
to
see
that
in
the
future.
F
Not
going
too
much
into
the
hardware
for
a
software
conference,
one
of
the
reasons
we
went
with
ocp
is
that
it's
not
about
using
power
it's
about
where
that
power
is
used.
So
even
things
like
having
a
single
shelf
that
is
doing
power
conversion.
That's
then
distributing
dc
to
all
of
the
servers
it's
about
using
the
energy
that
we're
consuming
efficiently
to
use
it
rather
than
doing
that
ac
dc
conversion
in
every
server
in
the
rack.
If
we
combine
it,
we
get
efficient
use,
and
I
guess
that
extends
to
other
areas
as
well.
A
That
actually
leads
into
the
next
question
that
I
wanted
to
talk
about
about.
Why
open
hardware
could
make
a
difference.
Part
of
it
is
because,
in
terms
of
the
work
that
we
do
in
designing
hardware,
it's
designed
to
run
at
scale
and
at
scale
you
can't
afford
to
lose
even
a
little
bit
of
energy
anywhere.
You've
got
to
design
it
so
that
everything
gets
consumed
properly.
A
And
so,
when
you
look
for
open
hardware,
that's
designed
around
the
scale
there's
already
efficiency
built
in
that
doesn't
get
moved
into
other
infrastructure,
and
you
were
just
talking
to
a
little
bit
of
that.
The
other
issue
is
in
order
to
hook
into
control
systems.
You
know
looking
at
the
firmware,
that's
in
the
devices
and
the
lower
layers
of
software
that
I
showed
earlier.
A
A
C
Yeah,
okay,
thanks
lee
for
that
yeah.
Actually
I
don't
know
actually,
what's
your
background
guys
if
you
are
developers,
mostly
I'm
not
sure,
but
most
probably
so.
You
actually
write
your
code
and
the
first
layer
is
the
hardware
itself
so
open
hardware,
and
so
on
liquid
cool
I
mean
not
only
immersion
but
also
coldplay.
C
There
are
different
technologies
that
are
more
or
less
established
in
the
market
and
and
the
next
phase
is
the
facility
itself
right
so
and
everything
all
of
your
code
most
probably
go
to
the
data
center
data
centers
are
mostly
huge
places.
Sometimes
10
megawatts
of
power
are
concentrated
into
a
single
place.
C
The
10
megawatts
is
equivalent
to
the
power
consumption
of
20,
000,
households,
20
000
households,
and
this
takes
only
10,
000
square
meters
more
or
less
and
20
000
households
take
a
bit
more
than
that,
and
and
if
you
take
into
account
all
this
power
is
coming
inside,
it's
creating
a
value.
Actually,
it's
actually
moving
bits
at
the
end
of
all,
but
it's
creating
waste
heat
100
is
going
into
waste
heat.
So
this
is
a
garbage
where
you
have
to
pay.
C
You
have
to
invest
a
lot
of
energy
to
remove
this
energy,
this
garbage
and
instead
of
doing
that,
if
you
could
actually
put
that
into
a
distributed
heating
network
into
some
facility
industrial
facility
that
needs
it,
it's
a
huge
leverage
and
actually
you
can
monetize
that
waste.
So
this
waste,
actually
I
mean
again
numbers
I
mean
taking
spanish
numbers,
I'm
actually
from
from
spain.
So
I
know
the
spanish
numbers:
20
000
households,
power
consumption,
but
around
10
000
households
for
heating
consumption.
C
F
Was
just
going
to
add
that,
as
as
developers
and
you
know,
maybe
some
in
small
companies
slightly
larger
companies,
you've
probably
got
a
back
office
somewhere,
that's
running
hardware
for
your
development
environments
or
your
staging
environments,
and
you
think
that
that's
you
know
really
really
handy
to
have
that
kit
in
a
room
next
to
you.
But
you've
got
to
remember
that
you're
delivering
power
to
that
room
and
cooling
to
that
room,
whereas
the
efficiencies
that
you
get
of
moving
it
to
a
data
center
are
really
massive
from
an
energy
efficiency
point
of
view.
F
E
Yeah,
so
actually,
I
might
just
come
in
there.
So
we
mentioned
there
were
quite
a
few
software
initiatives
that
were
that
we're
driving
one
is
the
intelligent
workload
placement
the
telemetry
aware:
scheduler.
We
have
the
gpus
scheduling,
we
have
two
engineers
that
are
working
on
it
in
the
room
and
that
really
makes
sure
that
you're
putting
your
your
workload.
E
You
know
it's
like
that
intelligence
layer
on
top
of
the
native
kubernetes
scheduler,
so,
instead
of
maybe
running
a
workload
with
a
node
that
has
maybe
only
a
low
memory
or
you're,
and
it's
and
it's
on
a
hungry
one,
you're
you're
running
it
on
the
highest
one.
So
it's
that
kind
of
smart
decision
making
that
we're.
E
I
mean
we're
in
the
early
stages,
but
and
we've
got
lots
of
feedback
from
some
of
you
guys
there,
the
other
day,
which
is
going
to
really
help
evolve,
that
the
other
thing
is
the
cpus
and
and
how
that
how
that
works.
The
one
I
wanted
to
call
out
just
because
we're
talking
about
idle
servers
is
the
power
manager
and
two
engineers
from
from
that
are
here
too.
E
So
really,
some
of
that
is,
is
changing
the
frequencies
of
the
cause
and
you
can
get
like
a
15
to
30
percent
improvement
when
you,
when
you
scale
the
cause
up
and
down
the
other
thing,
we're
working
on
we're
kind
of
hoping
that
we'll
be
able
to
release
it
in
the
next
coming
months,
is
really
around
peace,
states
and
sea
states.
So,
as
you
say,
you
know,
if
you
have
lots
of
idle
servers,
I
mean
we
talked
about
in
the
in
the
keynote
of
this.
E
E
You
can
save
a
lot
of
power
there.
You
can
also
we're
working
with
c
states.
You
can
also
put
them
into
kind
of
sleep
modes,
and
it
just
means
that
they're
not
switched
on
and
they're,
not
they're,
not
as
well.
They
are
on,
but
they're,
not
they're,
not
using
the
same
power
and
we've.
I
was
chatting
to
trisha
there
the
other
day,
and
you
know
we
we've
with
it's
probably
around
about
30
to
40
we're
expecting
in
terms
of
power
improvements
there,
which
is
going
to
be
huge.
A
H
That
correlates,
would
you
suggest
that
putting
your
workload
in
a
place
like
helsinki
regularly
becomes
placed
old,
I'm
assuming
to
the
additional
hops
that
the
request
needs
to
take,
because
every
server
on
the
way
also
consumes
energy
right,
and
the
small
stuff
also
adds
up
to
a
big
number.
At
the
end,.
E
Actually,
you
asked
the
other
day,
which
I
just
wanted
to
answer,
because
I
don't
know
whether
I
answered
it
correctly
at
the
booth.
You
asked
about
reuse
of
hardware.
What
are
we
doing
with
reuse
of
hardware?
It's
interesting
as
well.
Yeah
yeah,
it's
actually
a
really
good
question,
so
you
come
up
with
really
good
questions
and,
if
you,
if
you
power
down
your
calls
right
with
kind
of
the
power
map,
you
know
in
what
we're
working
on
right
now,
you're
actually
going
to
increase
the
longevity
of
your
hardware,
so
I
think
yeah.
A
B
A
A
H
A
B
I
A
G
If
you
then
have
one
connections,
they
consume
a
lot
more
power
because
the
sfps
in
there
consume
a
lot
of
more
power
because
the
lasers
need
more
power
because
of
the
distance,
so
I
think
there
it
can
add
up,
but
there
you
also
need
to
think
about
you're,
not
the
only
one
using
it.
So
you
need
to
then
condition
it.
How
much
bandwidth
do
I
need?
G
C
G
G
Yeah,
so
from
our
perspective,
we
are
currently
looking
and
gathering
that
data,
so
we
haven't
did
and
we
have
a
dedicated
team
inside
our
company
just
looking
for
it,
and
they
are
now
adding
up
and
also
checking
where
we
can
reduce
our
power,
where
we
have
issues
with
our
power
consumption
and
also
how
we
can
have
more
power
off-grid,
because
that's
currently
most
of
the
power
which
we
are
consuming
is
coming
from
the
grid
and
yeah.
That's
something
our
company
is
currently
looking
into.
A
And
we'll
conduct
the
audience
questions
I
wanted
to
come
back
to
some
of
the
discussions
we're
having
and
see
if
we
can
make
it
a
little
bit
more
concrete
around
hardware,
software
and
co-design
right.
So
if
we
were
to
imagine
a
world
a
year
or
two
from
now
right,
based
on
the
fact
that
we
have
both
hardware
and
software
knowing
that
people
panel,
what
do
we
see
happening
are
apis,
something
that
could
matter
and
could
become
standardized?
A
F
Where
it
was
the
some
of
the
things
kate
was
saying
about,
the
the
peace
state
sea
state
stuff
is
interesting.
It's
the
first
time
I've
I've
heard
about
it,
but
I
know
in
in
sivo
we're
trying
to
get
performance
at
the
moment
and
utilization.
So
we've
been
looking
at
how
we
do
that
scheduling
and
how
are
you,
configuring,
just
the
cpus
in
the
bios?
F
E
Know
just
I
I
just
I
just
caution
when
we
do
talk
about
performance
performance,
I
and
I
I
kind
of
said
that
right
at
the
start,
you
know
I
do
think
we
really
need
to
think
about
the
right
performance
and
the
right
and
that's
sustainability.
It
is.
It
is
a
change,
it
is
a
change
of
way,
but
but
we
need
to
also
make
it
easy
for
engineers
here
as
well
to
be
able
to
do
that
and
right
now
you
know
we're
certainly
in
intel
trying
to
drive
as
much
as
we
can.
G
Leverage
to
build
stuff,
which
then
is
more
sustainable.
I
saw
some
talks
about
other
schedulers
which
were
also
sustainable
or
where
you
can
say.
I
need
like
this
many
cpus
in
the
next
in
the
next
days
for
my
batch
processes
or
something
like
that,
what
they
do
with
san
now
and
there
are
a
lot
of
people
just
looking
to
schedule
us
and
I
think,
if
we
have
more
data
where
we
can
decide
on
for
a
scheduler
that
gets
easier
also
to
make
a
more
sustainable
workload
happen.
G
So
I
also
consume
in
like
pdus,
you
have
in
the
data
center
and
every
all
of
that
data
could
be
available
as
a
new
api
in
kubernetes
so
that
you
can
hook
in
as
a
developer
and
say:
oh,
this
bdo
is
not
as
efficient
as
the
other
one,
and
that
is
then
in
some
form
of
number,
and
you
can
say:
okay,
the
whole
stack
I'm
running
on
is
now
68
percent
efficient,
but
I
have
one
in
something
like
helsinki,
which
is
89
percent
efficient.
So
I
can
better
run
there.
If
my
workload.
E
G
E
In
terms
of
energy,
if
you're
moving
your
workloads
across
the
network,
so
that
is
something
that
needs
to
be
like
location
counts
right,
it
counts
in
terms
of
energy
usage.
So
that's
something
we
definitely
need
to
be
thinking
more
about.
I
know,
there's
you
know
that
there
are
a
lot
of
people
thinking
about
how
do
we?
How
do
we
measure
this
stuff?
You
know
what
what
do
we
need
to
put
in
place
from
a
software
perspective,
so.
G
B
F
It's
also
something
we've
got
to
make
make
easy.
I
mean
we're
at
a
conference
of
7
000
people
and
yet
we've
not
filled
7
000
people
into
this
room.
Quite
quite
rightly
because
it's
a
very
niche
topic
at
the
moment.
So
there's
a
lot
of
education
either.
We
need
to
do
to
make
everyone
care
and
spend
the
time
into
developing
code.
That's
efficient
or
making
these
scheduling
decisions,
or
we
as
a
small
group,
need
to
make
it
very
easy
for
everyone
else,
just
to
get
it
for
free
if
they
use
kubernetes.
C
In
my
opinion,
also
we're
talking
about
totally
separated
worlds.
Sometimes
we
we,
we
think
about
the
facility
as
facility
infrastructures
as
a
very
strange
world
world
from
the
whole
software
and
cloud
infrastructure
right
and-
and
our
experience
in
all
these
years,
to
be
also
in
both
of
the
worlds
is
very
positive.
So
we
we
actually
in
our
monitoring
system.
We
gather
information
from
the
whole
facility
and
also
from
the
servers
and
everything
flows
in
the
same
system.
C
So
actually
we
measure
at
the
at
the
core,
the
temperature
we
measure
the
pressure
outside
as
well,
and
so
with
this
also,
we
can
actually
manage
the
the
flow
movement
and
the
workflow
movement
as
well.
I
mean
at
the
end.
This
is
also
something
difficult,
usually
because
facility
managers
are
at
one
side
and
the
hardware
managers,
or
even
the
the
the
programmers
are
on
the
other
side,
and
they
don't
pick
each
other,
and
this
is
actually
a
gap
that
we
need
to
to
feel.
A
hopeful
initiative
as
ocp
can
can
actually
feel
that.
A
Actually,
what
I'm
hoping
is,
after
today,
some
of
the
people
in
this
room
might
reject
to
us
and
understand
how
we
might
engage
with
some
of
my
counterparts
at
the
osdp
that
are
looking
at
sustainability.
That
measure
it
looking
at
different
ways
to
bring
bridgette
hardware.
Software
divide,
and
you
know,
benefit
from
have
software
that
can
make
the
right
decisions
based
on
knowledge
of
hardware.
I
mean
being
to
some
extent
hardware
aware
software
is
scheduled
to
kubernetes,
so
yeah.
That's
one
thing
that
we're
hoping
we'll
see
today.
C
Yeah
so
already
talked
about,
of
course,
liquid
cooling
solutions
and
the
waste
heat
reuse
already
talked
about
that
about
the
amount
huge
amount
of
energy
actually
is
needed
for
data
centers
and
wasted
and
another
another
thing
I
also
have
talked
about.
Actually,
we
have
talked
about
everything
already,
so
I'm
just
going
through
that.
So
is
this
orchestration
tool
so
moving
loads
from
a
to
b
to
c.
Actually,
we
started
also
answering
your
question.
C
We
we
proved
that
at
the
beginning,
when
we
had
these
distributed
racks
in
the
cellars
of
houses,
because
one
neighbor
was
going
in,
you
know
in
holiday
and
turning
off
the
heating
system.
So
at
that
time
we
realized
hey.
We
have
to
move
this
to
a
different
house,
because
otherwise
our
cloud
clients
won't
get
their
services.
So
actually
we
should
be
able
to
monitor
those
those
facts
and
be
aware
of
those
facts.
C
So
this
start
start
a
process
in
our
minds
to
develop
this
open
source
software
called
krake
and,
of
course,
release
into
the
into
the
community.
The
idea
in
the
future
will
be
not
only
helsinki
and
valencia,
but
maybe
northern
hemisphere
southern
hemisphere,
summer
winter
day
night.
So
all
of
these
climatic
changes
that
actually
make
our
data
centers
to
behave
in
a
different
way.
G
C
C
This
also
is
a
way
actually
restricting
looking
at
the
efficiency
to
be
more
green
in
an
indirect
way
yeah.
So
it's
running
into
kubernetes
as
well,
when
the
coordinate
event.
So
this
this
lifecycle
management
tool
is
running
on
kubernetes
is
much
more
efficient
using
the
the
containers
for
that
yeah.
E
E
Yeah,
no
I'm
just
I'm
just
thinking
about
just
software.
I
know
we've
talked
about
scheduling
what
we
haven't
talked
about
as
observability
and
some
of
the
metrics
you
get
right.
So
you
know
we
I
think
I
mentioned
some
of
the
things
we're
driving
we're.
Also.
We
also
have
a
group
working
in
the
metrics,
so
they're
working
with
collect
d
telegraph
and
some
of
the
areas
that
they
just
to
provide
you
with
that
granular
level
of
metrics,
so
that,
if
you're
interested
in
your
power
consumption
you
can
you
can.
E
G
Yeah
we,
we
are
satellite,
knows
not
so
many
decisions
as
you
as
we
need
to
provide
the
service
in
germany.
So
somehow
we
need
to
run
our
stuff
in
germany.
So
the
only
way
for
us
is
to
reduce
our
consumption,
which
is
very
high,
currently
and
also
terminate
traffic
earlier
in
our
network,
and
that's
the
only
ways
we
can
reduce
our
footprint
and
also
get
more
green
energy
into
our
stack.
E
Yes,
absolutely
what
a
wonderful
question.
Thank
you
for
asking
it!
So,
yes,
absolutely
everything
I've
I've
spoken
about
today
is
all
open
sourced
from
the
telemetry
aware,
scheduler
to
the
gpu
aware
scheduler
to
the
power
manager
to
the
the
metrics
that
we
have
available.
So
you
can
learn
about
your
power
consumption.
E
Everything
is,
is
open
sourced
and,
as
I
said,
a
lot
of
this
stuff.
We
are
kind
of
in
early
days
that
there's
a
huge
amount.
We
can
do
there,
as
I
had
mentioned
yesterday,
like
the
possibilities
here
are
and
especially
when
people
start
looking
at
it.
We
also
have
you
know
like
a
guide
anyway,
as
well,
just
to
help
people
install
and
play
with
some
of
the
stuff
too.
So
a
lot
of
the
engineers
working
on
that
are
here.
So,
if
you're,
if
you're
interested,
please
please
chat
with
them.
E
F
K
And
then
we're
also
rolling
out
recommenders,
for,
as
somebody
was
asking
earlier,
which
is
the
most
efficient
data
center
to
be
deploying
your
workflows
to
we've.
Had
that
in
preview
for
a
while
and
the
feedback
again
is
the
customers
are
very
much
picking
more
efficient
data.
So
it's
having
a
good
effect.
I
E
E
No
no,
but
I
mean
these
are
all
things
to
you
know
this
whole
thing
around
carbon
awareness
software.
This
is
where
we
need
to
start.
This
is
exactly
this
kind
of
thing
that
we
want.
People
start
thinking
about
right,
so
so,
because
you've
got
you've
got
so
many
really
good
ideas
there.
So
how
do
we
start
thinking
about
this?
How
do
we
change
what
we've
been?
You
know
just
thinking
about
that
performance.
Thinking
about
that
throughput
to
thinking
about
okay.
So
now
you
know
where
is
where
is
how
do
we
make
something?
E
F
It's
also
worth
worth
noting
as
well.
We're
always
at
the
moment
seem
to
be
talking
about,
assuming
that
all
of
the
the
code
and
the
workload
that
is
being
run
on
hardware
is
the
right
coding
decisions
I
mean
taking
blockchain
at
the
moment.
Is
that
the
most
efficient
way
we
as
a
community
need
to
be
starting
to
move
our
computing
towards?
F
We
know
how
energy
efficient
or
non-energy
efficient,
even
blockchain
workload
is
and
adding
things
to
blocks
and
proof
of
stake,
work
proof
of
proof
of
work
it
is
and
if
we
are
starting
to
move
as
an
industry
towards
making
this
a
standard,
is
this
really
the
most
energy
efficient?
Is
this
the
most
responsible
way?
We
should
be
moving
the
web?
So
there's
a
lot
of
things
that
you
can
do
when
you
go
back
into
your
company
and
you're
designing
how
we're
going
to
take
our
product
to
the
next
technology
level.
F
C
F
E
Yes,
yes,
so!
Yes,
so
that's
actually
what
we're
that's
our
our
phase?
One
of
our
power
manager
was
released
right,
but
it
didn't
have
that
p-state
c-state
piece
in
there
that
we're
planning
in
the
next
coming
months
to
release
the
next
version
of
the
power
manager
with
with
that
in
there,
so
that
people
will
be
able
to
to
use
it
right,
but
you're,
absolutely
right.
It
has
been
around,
but
for
some
reason
it
has.
We
haven't
really
been
thinking
about
it
from
a
sustainable
perspective,
so
that
that
will
that
functionality
will
come
in.
L
A
E
Yeah
and-
and
you
know
the
fact
that
even
hopefully
now
people
who
are
working
in
the
open
source
projects
will
be
putting
that
in
as
an
agenda.
Well,
what
are
we
doing
for
sustainability
right?
They
may
not,
hopefully
when
they
go
back
because
they
may
not
have
done.
This
right
is
okay.
We're
going
to
have
this
as
an
agenda
item
we're
going
to
we're
going
to
start
brainstorming.
How
can
we
we
get
a
better
design?
How
can
we
get
better
architecture
but
absolutely
great
idea
around
the
tag.
I
G
Yeah,
I
think
it
could
also
be
something
like
a
validation
so
that
you
have
some
workload
which
always
consumes
the
same
power
or
something
like
that,
and
you
run
it
in
different
cloud
providers
and
then,
in
the
end,
have
an
audit
who's
reporting,
false
numbers
things
the
same
with
cars,
some
report
better
mileage,
other
ones,
not
so
yeah.
I
think
there
needs
to
be
trooping
to
prove
it.
I
F
Yeah
there's
also,
the
idea
of
the
ocp
is
an
open,
open
source
hardware
project.
So
you
would
like
to
hope
that
people
are
checking
things
and
if
you've
got
one
company
providing
a
design
and
claiming
some
specs
that
is
verified
by
a
competitor
and
as
a
as
an
open
organization,
it
should
almost
to
a
certain
extent,
be
self-validating,
because
if
someone
released
something
and
it's
false,
everyone
else
is
going
to
shout
about
it.
F
I'd
probably
say
it's
something
that
we
have
to
just
push
to
our
employers
and
say
look
this
is
this
is
not
on.
This
is
not
something
we
want
to
continue
supporting
and
yes,
there
might
be
this
bug,
but
actually
it's
almost
technical
debt
right
in
in
an
organization
and
like
you
were
pushing
it
for
a
bug,
fix
we've
got
a
push
to
say
no,
this
is
no
longer
efficient.
We
can
save.