►
From YouTube: Rust and sustainable IT - Benoit Petit - EuroRust 2022
Description
Benoit Petit
ICT Infrastructure Engineer at Hubblo
We know Rust is performant and can be used to build software that have a low footprint on IT resources. But how much does it relate to energy consumption? It this a crucial part of architectural choices for lighter applications (some would say "greener")?
How Rust could be used to assess the environmental impacts of ICT?
What are the environmental impacts of ICT anyway? Is electricity the only topic?
We’ll try to give decent answers to those questions and provide leads to dig deeper in those huge yet mostly unknown topics.
A
A
A
And,
as
you
may
have
guessed
it
there
are?
There
are
some
beige
maths,
some
benchmarks.
There
is
one
from
Portuguese
researchers
that
has
made
some
noise
like
six
months
ago
or
maybe
one
year
ago.
Actually
it's
more
it's
older
than
that,
but
because
they
publish
every
year
a
new
version,
every
time
it
has,
it
makes
some
noise.
A
It's
based
on
the
computer
language
Benchmark
game,
which
is
a
fixed
set
of
algorithms
that
you
can
find
on
Wikipedia
and
it's
just
a
comparison
across
all
those
algorithms
between
the
languages.
So,
as
you
guessed
it,
it's
a
based
on
fixed
tests,
because
the
algorithms
are
defined
defined
in
advance
and
there
are
no
other
real
variables
in
the
test.
But
anyway,
what
we
can
see
in
this
Benchmark
is
that
rust.
A
Most
of
the
time
is
pretty
well
ranked
in
terms
of
energy
time
to
complete
the
task
memory
consumed
most
of
the
time
it's
in
the
top
of
the
rank
not
always
at
first
in
many
cases,
C
or
C
plus
plus,
could
do
better,
but
it's
still
in
the
top
three
or
the
top
five,
depending
on
the
test,
you
will
look
at
and
if
you
look
at
the
aggregated
results,
it's
still
good.
Rest
is
in
second
position
in
energy
and
time
only
in
sixth
position
for
memory.
A
A
To
my
opinion,
we
should
know
not
go
so
fast
and
remember
that
the
choice
of
the
language
is
just
one
part
of
the
software
architecture,
and
that's
that's
in
in
this
idea
that
there
is
another
Benchmark
that
has
been
published
this
time.
It's
made
by
some
of
my
colleagues
at
webstar,
our
fellow
members,
Ewen
and
Jeremy
and
other
contributor
contributors
since,
but
they
started
the
project
in
the
beginning
and
in
this
Benchmark
they
included
other
architectural
charges,
compilers
interpreters.
A
So,
for
example,
on
python,
you
could
run
on
the
default
interpreter
or
there
is
by
pay
free
or
there
are
other
interpreters.
That's
one
of
the
variables
languages,
of
course,
databases
so
far.
I
think
they
only
have
compared
with
postgresql
and
Mexico
kgs,
whether
or
not
there
is
pagination
or
not
in
the
app
so
yeah.
Sorry,
the
the
basis
of
The
Benchmark
is
a
to-do
list
to
do
this.
A
A
A
A
A
Any
tip
for
the
next
rest
event,
maybe
next
time
you
could
make
you
enjoy
me,
because
that
presentation
is
very,
very
detailed
and
very
interesting.
At
least
it
was
very
interesting
to
me
so
again
the
stage
it
still
seems
great
as
a
residential
player.
So
far
so
good.
We
only
have
good
signs
that
rust
is
still
a
nice
choice,
spread
it
all
out
it
won't.
A
This
won't
change,
but
just
we'll
try
to
widen
to
have
a
wider
view
than
just
this
question
about
language
before
that
to
clarify
also
my
relationship
to
rest,
one
of
the
maybe
the
first
approach
for
me
about
environmental
impacts
of
HCT
on
the
environments.
It
was
okay.
What's
the
energy
consumption
of
my
server?
A
What's
the
energy
consumption
of
my
software,
that
kind
of
things
I
had
in
mind
already
at
that
time,
even
if
I
didn't
know
the
benchmarks
that
rust
a
load
for
performance
and
low
resources,
consumption,
so
I
went
for
it
when
I
tried
to
evaluate
this
energy
consumption,
so
I
started
a
project
to
measure
the
energy
consumption
of
servers
and
the
applications
running
on
the
servers
and,
of
course,
I
did
this
University.
So
the
project
is
named
scaffold:
it's
an
open
source
project.
A
Of
course
it
has
now
about
15,
active
contributors,
I
guess-
and
it
runs
in
several
companies-
machines
that
are
very
good
contributors
for
us
to
clarify.
What's
the
the
role
of
Scaffolding
in
the
monitoring
stack,
it's
just
a
data
collection
group,
so
it's
a
metering
agent.
Then
we
send
the
data,
of
course,
to
a
software
that's
made
for
storing
data
like
Primitives
or
other
time
service
database,
and
then,
of
course,
you
could
visualize
your
metrics
in
graph
anal
and
in
visualization
tool.
A
I
make
a
very
quick
demo,
a
lot
of
all
the
stacks,
but
at
least
a
refunded
as
well
a
sample
so
just
about
the
Matrix
of
one
machine.
So
pretty
simple
example:
we
can
see
that
we
have
the
power
usage
of
the
machine
so
to
be
precise,
there
I'll
explain
that
just
after,
but
we
are
based
on
a
certain
technology
to
have
this
data.
So
we
we
miss
a
part
of
the
power
usage
of
the
machine
here
and
I.
A
A
So
this
is,
as
we
said
about
our
usage,
then
you
can
aggregate
with
data
to
have
energy
consumption
Matrix
on
a
given
time
frame,
whether
it's
three
hour
to
one
year
and
then
it's
pretty
easy
to
calculate
that,
and
then
we
can
zoom
on
the
power
usage
of
each
software.
If
we
zoom
on
each
graph,
we
see
that
it's
the
power
usage
of
its
process,
that's
attributed
to
that's
related
to
this
software,
so
I
won't
go
into
details
here.
A
It
was
just
to
clarify
that
my
relationship
to
rest
in
this
area
is
also
because
it
seems
we
can
build
tools
that
help
us
to
evaluate
our
impacts
that
are
pretty
low
on
resource
consumption,
which
is
great.
So
of
course
it
runs
on
Rust
and
to
explain
a
little
bit
about
the
data.
As
I
said,
we
are
based
on
rainbow,
so
it
stands
for
running
average
power
limit.
A
This
is
a
brief
overview
of
the
features
of
scaffold
that
let
the
point
here.
Maybe
we
just
can
export
the
data
in
different
monitoring
tools
or
time
series
database,
and
we
can
run
it
in
different
contexts.
Just
remember
that
the
virtualization
layer
is
one
of
the
biggest
pain
points
so
so
far
we
can
have
powers
power
measurements
in
Virtual
machines,
but
only
on
qmu
KVM,
because
we
only
developed
that
support
at
this
time.
A
So
once
again
we
could
say:
okay,
that
is
great.
We
can
build
a
helpful
tools.
We've
arrest
to
help
us
understand
our
impact,
which
is
part
of
the
reduction
in
process,
but
let's
take
a
step
back
so
far
we
talked
about
performance,
electricity
usage.
So
that's
interesting.
That's
part
of
the
the
topic,
but
that's
not
all
of
it.
A
It's
supposed
to
be
between
two
and
four
percent
of
the
greenhouse
gas
emission
globally.
But
what's
interesting
is
the
trend
because
all
of
these
estimations
they
talk
about
a
pretty
big
increase
so
that
in
2025
we
may
have
a
between
six
and
eight
percent
of
the
original
gas
emissions
in
the
world
due
to
act.
So
that's
not
nothing
and
more
than
that,
the
impact
of
ICT
are
not
just
about
bringing
those
gas
emissions
in
this
table.
A
We
have
some
of
the
impacts,
not
all
of
them,
but
me,
for
example,
primary
energy
consumed.
That's
the
First
Column
greenhouse
gas
emission
that
the
second
one
gwp
stands
for
global
warming,
potential,
water
scarcity,
electricity
consumption
resources,
consumption
and
when
we
talk
about
resources
consumption
in
ICT,
most
of
the
time
we
talk
about
minerals
and
methods.
A
So
that's
the
first
point.
There
are
multiple
impacts,
so
we'll
talk
about
that
just
after
the
other
point
is
that
we
see
in
this
table
that,
with
those
estimations
it
seems
that
that
the
user,
the
end
user
devices
other
the
the
part
of
ICT,
that's
the
most
impactful
are
the
other
parts
you
know
in
this
model
is
networks
and
data
sensors.
A
So
from
that
we
can
say
that
as
individuals
and
it
professionals,
we
already
have
a
responsibility
to
extend
the
lifetime
of
our
devices
as
much
as
possible
by
as
less
new
products
as
possible
and
maybe
by
refurbished
products,
repair
our
devices
screens,
laptops,
smartphones
and
so
on,
so
that
that's
the
first
point
and
I
think
it's
very
important
because
it
includes
everyone,
it's
not
just
about.
Are
we
in
ICT
or
are
we
end
users
of
ICT?
It
concerns
everyone.
A
A
But
yet,
if
we
sum
up
the
data
from
the
previous
table
for
Data,
Center
and
network,
so
let's
say
infrastructures
that
are
needed
to
provide
a
service
infrastructure
still
represents
56
percent
of
the
electricity
consumed
by
act,
37
of
the
greenhouse
gas
emissions
emitted
by
Act
and
40
of
primary
energy
consumeration.
So
it's
not
nothing.
It's
pretty
important,
still
important,
even
if
user
devices
are
as
well.
So
as
it
professionals.
A
A
So
it's
the
same
as
if
I
was
driving.
A
car
I
would
burn
for
social
work
locally,
so
I
emit
with
gas
emissions
locally,
which
means
it's
in
Scotland
direct
emissions
in
80.
If
you
don't
manage
a
data
center
yourself,
most
of
the
time
you
don't
have
much
to
account
for
in
scope.
1.,
you
will
be
much
more
interested
in
script.
Two
and
three
so
sculpture
is
about
indirect
emissions.
A
A
So
I'm,
indirectly
emitting
greenhouse
gases
and
Scott
free,
it's
all
the
rest
and,
what's
all
the
rest,
is
it's
most
of
the
time
what's
hidden
in
ICT,
so
the
first
phases
of
the
life
cycle
of
a
product
which
means
extraction
of
the
raw
materials
manufacturing,
Transportation
installation
as
well
and
in
the
end,
the
end
of
life
of
the
the
hardware.
Maybe
the
recycling,
which
can
have
an
impact
as
well
and
keep
in
mind
that
recycling
in
ICT
is
very,
very
bad
and
that's
globally.
A
So
if
we
apply
this
way
to
account
the
greenhouse
gas
emissions
to
a
data
center,
it
could
look
like
this
with
a
very
simple
simplification
of
data
center
bits
to
illustrate
so
in
in
green.
We
have
scope
one.
So,
as
we
said
previously
in
this
example,
it
would
be
just
the
the
fuel
tank
and
the
backup
generator
that,
let's
say,
related
to
scope.
A
One
scope
2
is
in
yellow,
so
all
the
ICT
devices,
compute,
storage,
Network
and
so
on
are
of
course
concerned
by
Script
2,
but
also
all
the
other
devices
that
consume
fqc,
so
cooling
systems,
lighting
fire
off
stop
systems
and
that
kind
of
thing
the
the
electricity
range
is
also
is
in
in
yellow
because
we
will
see
that
after.
But
you
also
need
to
know
what's
the
carbon
intensity
of
the
electricity
you
consume
and
then
scope
free
in
blue.
A
Actually,
it's
everything
because
everything
has
been
built
at
some
point,
so
everything
has
emitted
greenhouse
gas
station
at
some
point,
the
building
included
and
you
you
can
see
how
sometimes
it
makes
evaluations
for
free
missions
have,
and
that's
maybe
why
many
many
companies
still
excludes
Capri
from
their
carbon
report.
That's
pretty
sad
because
even
if
it's
not
always
known
scope,
3
is
is
in
many
cases
the
the
worst
part
of
the
impact.
A
A
So
we
base
our
work
on
this
and
we
try
to
apply
this
methodology
and
if
we
make
the
matchup
between
the
two,
that's
very
simple,
so
scope
3
in
LCA,
it's
extraction
of
raw
materials,
manufacturing,
transport,
installation,
end
of
life
and
scope.
2
is
usage.
That's
as
simple
as
that
in
simple,
in
certain
tools
or
documents
you
may
find
other
naming
one
is
embodied
impacts,
so
embody
impacts
is
scope
free.
So
extraction,
manufacturing
and
so
on
and
operational
impacts.
Its
usage
also.
A
So
let's
talk
one
last
time
with
electricity
to
to
be
a
bit
more
complete.
If
we
zoom
on
the
data
centers
consumption
of
electricity,
the
estimations
today
are
pretty
different.
We
have
estimations
of
200
terawatts
power
of
electricity
consumed
by
data
center
in
the
world.
We
have
estimation
of
800
terawatt
hour
of
that
data
center
would
consume
in
a
year
in
in
the
world,
so
wide
range
of
estimations.
That's
not
very
surprising,
because
the
data
is
very
hard
to
get.
A
But
what's
interesting
is
that
even
the
the
estimation
that
that
take
the
lowest
hypothesis
so
200
watt
hour
in
2020,
for
example,
most
of
them?
They
don't
talk
about
a
predicted
reduction
of
the
energy
consumption.
They
more
talk
about
a
huge
increase
for
some
of
them.
They
predict
multiplication
by
free
of
this
energy
consumption
by
20
by
2030.
A
It's
not
necessarily
easy,
because
producing
electricity
is
not
a
simple
process.
We
have
to
get
energy
in
another
form,
which
is
called
primary
energy.
First,
we
have
to
transform
that
energy
to
make
electricity.
Then
we
have
to
transfer
that
energy
on
a
an
electrician
electricity
Network,
so
that
in
the
end
we
can
power
up
servers
and
have
applications
and
so
on
in
the
process
we
lose
a
lot
of
energy.
When
we
build
dhsc,
the
power
generation
loses
a
lot
of
energy.
A
A
In
the
machine
itself.
The
power
supply
will
lose
energy
end,
funds
will
will
consume
some
of
the
energy
in
the
end.
In
this
example,
don't
take
the
numbers
for
what
they
are
is
just
a
order
of
magnitude,
but
we
could
say
that
most
of
the
time
we'll
have
maybe
10
percent
of
Energy
Efficiency
regarding
the
the
energy
that
we
have
actually
consumed
in
its
primary
form.
A
So,
let's
say
when
we
consume
10
watts
for
a
server.
We
have
consumed
actually
100
watts
of
primary
energy
in
the
beginning
to
estimate
that
overhead,
at
the
scale
of
the
data
center
itself
of
the
building,
there
is
a
metric,
that's
called
the
Pue.
You
may
know
it
so
it's
power
usage
Effectiveness,
it's
just
a
ratio
between
the
energy
consumed
by
all
ICT
equipments
in
the
building
and
the
energy
consumed
by
everything
in
the
building
and
the
closer
stitches
to
1.0
the
best.
A
It
is
1.0
periods
the
don't
exist,
but
there
are
some
actors
on
the
market
that
are
able
to
have
pretty
good
pus
in
1.2
1.3
or
something
like
that
and
we
talked
before
about
carbon
intensity.
So
what
I
presented
you
just
before
was
about
how
much
energy
we
consume,
but
when
you
want
to
evaluate
your
scope
to
emissions,
so
scope,
2,
remember,
usage
emissions.
A
You
need
to
know
what's
the
intensity,
how
much
greenhouse
gas
station
I
emit
for
one
kilowatt
hour
of
electricity
produced,
and
you
have
some
tools
today
that
are
pretty
nice
to
have
this
an
idea
about
that,
depending
on
the
country
where
your
services
are,
one
of
them
is
electricitymap.org.
That's
an
interactive
map
where
you
can
zoom
on
each
country,
have
the
electricity
mix
and
how
it's
composed,
how
it's
impactful
and
so
on.
It's
also
an
API.
So
it's
pretty
nice
to
for
every
evaluation
process.
A
A
The
other
approach
is
called
Market
based,
so
there
it's
not
the
emissions
of
the
electricity
that
we
that
is
produced
it's
the
emissions
of
the
accuracy
that
is
about,
and
that's
that
could
seem
like
nothing.
It's
a
huge
difference
because
actually
not
it's
not
the
electricity,
that's
actually
consumed
or
produced
that's
taken
into
account,
and
you
can
include
in
the
calculation,
then
Market
mechanisms
like
warranty,
guarantees
of
Origins,
renewable
electricity
certificates,
ppas
and
so
on.
A
So
I
won't
have
the
time
to
go
into
details
into
those
Notions,
but
keep
in
mind
that
at
least
my
opinion
is
that
it's
grooming
machine
period-
and
there
are
there-
is
much
more
to
say
about
that.
But
I
won't
do
it
today
and
just
to
sum
up
about
15
slides
about
that,
because
it's
a
huge
topic,
it
opens
the
topic
to
other
Notions.
A
First,
a
leadership
because
we
talked
a
lot
about
electricity
and
carbon
and
I
said
I
want
to
emphasize
that
it's
not
the
whole
topic
about
abiotic
resources
depletion
so,
as
I
said,
abiotic
resource
depression
in
ICT
most
of
the
time,
it
means
how
much
minerals
and
metals
we
are
extracting
from
the
ground
and
consuming
I
took
one
number
from
French
report
preview.
Recent
that
says
in
2020,
so
I'm
not
sure,
but
it's
a
report
from
2022,
but
the
data
may
come
from
2020..
A
So
when
we
look
at
that,
this
Frame
could
say:
that's
not
so
much
I
could
imagine
it.
It
was
a
much
more
than
that.
But
actually,
what
we
have
to
understand
here
is
that
when
we
say
900
and
so
tons
of
metals
in
digital
products,
it
doesn't
mean
we
have
extracted
950
tons
of
materials
from
the
ground.
A
A
This
is
a
representative,
a
drawing
of
the
process
from
extracting
raw
materials
from
the
ground
to
having
I,
don't
know
the
English
word,
but
cathode,
so
just
a
metal
bar
that
you
couldn't
send
to
manufacturers
so
that
they
build
projects
after
that
and
the
process
is
pretty
complex.
First,
you
extract
the
materials.
A
A
In
this
example,
we
take
copper,
which
is
one
of
the
metals
we
need
for
ictn
other
fields.
If
you
look
at
the
first
item
on
the
left,
we
see
that
the
mineral
in
the
ground
is
just
concentrated
at
0.4
to
2
percent,
so
we
are
away
from
the
100,
and
so
there
is
a
huge
process.
Then
we
need
to
smash
the
materials
extracted
so
that
we
get
the
pieces
of
minerals
that
are
more
concentrated.
A
This
consumes
a
lot
of
energy
and
emits
a
lot
of
greenhouse
gas
emissions.
Then
there
is
chemical
extraction
to
get
rid
of
the
minerals.
We
don't
want
and
just
keep
the
one
we
want.
We
can
when
just
parenthesis
on
this,
which
means
that
we
will
get
rid
of
minerals
that
actually
are
not
interesting,
but
it
also
means
that
we
will
get
rid
of
minerals
that
are
also
interesting
to
the
industry
because
of
economic
factors.
A
Sometimes
we
lose
minerals
that
have
our
of
any
value,
because
the
mind
is
not
made
for
those
metals
and
is
not
intended
to
extract
those
Metals.
So
there
is
a
huge
loss
here
as
well
and
when
we
say
chemical
extraction
you
imagine,
that
means
a
lot
of
chemical
products
and
it
pollutes
the
soils,
the
water,
the
air
and
so
on.
It
leads
it
leaves
full
areas,
very
big
areas
and
all
Stephen
talks
very
well
about
that.
That
are
completely
unlivable.
A
So
then,
how
do
we
evaluate
those
impacts?
Those
embodied
impacts,
whether
it's
greenhouse
gas
emissions
or
abiotic
resources,
depletion
or
something
else?
At
some
point,
you
need
data.
What
are
the
data
of
impacts
of
a
device
or
a
component
or
something
else
so
you're
going
to
have?
You
could
use
databases,
for
example?
A
A
So
what
we
try
to
do
to
make
do
data
more
accessible
and
to
democratize,
evaluation
and
reduction
of
the
impacts
is
that
we
created
an
open
database
on
GitHub
webstar,
which
is
actually
aggregation
of
the
data
the
manufacturers
publish
about
the
impact
of
their
product.
We
just
made
made
them
more
accessible
by
having
all
of
them
in
the
same
place.
A
Even
if
the
evaluation
is
not
perfect
in
your
company
at
some
point,
then
how
could
we
make
evaluation
easier?
There
are
tools.
Existing
I've
presented
here,
mostly
open
source
tools
in
the
center
bloody
status,
mostly
on
the
right
and
the
carbon
Cloud
providers,
carbon
calculation
tools
in
the
bottom.
The
tools
don't
help
to
evaluate
the
same
thing,
so
you
may
have
to
match
some
of
them
to
have
your
evaluation.
In
your
context,.
A
I
just
want
to
show
you
a
very
brief
demo.
I
think
I
have
two
minutes
left,
I
hope.
First,
one
is
data
Visa,
that's
the
database.
I
told
you
about
before,
so
you
have
impacts
of
devices,
for
example,
if
I
zoom
on
laptops,
it
will
give
me
an
average
on
all
the
references
of
equipments
we
have
in
the
database.
We
see
here
that
scope
3
in
blue
is
more
important,
so
we
see
that
manufacturing,
let's
say,
is
more
important
than
usage
in
terms
of
impact.
A
So
that's
data
Vista.
Another
thing
we
did
is
we
made
this
an
API
that
will
give
you
metrics
of
impact
of
your
servers
and
services.
Those
are
not
from
manufacturer's
data.
This
time
it's
based
on
Paradise
impact
metrics
and
what's
a
die
when
you
build
an
edge
component
in
the
manufacturing
process,
you
have
cylinders
of
mid
of
metal
oxidation
that
you
will
cut
in
slice
and
do
slices.
Those
disks
are
called
Wafers.
A
So
then,
when
you
have
this
API,
you
have
answers,
but
it
doesn't
make
everything
you
still
have
to
collect
your
own
inventory
of
devices
to
know
what
questions
to
you
have
to
ask
to
the
API,
and
you
are
still
using
hypothesis
on
energy
consumption.
If
you
don't
measure
it
so
we
start
all
the
projects
on
top
of
the
API.
A
One
of
them
is
blue
agent
super
agent.
So
sorry,
I
didn't
translate
that
brilliant
is
a
tool
that
will
scan
for
all
the
hardware
specifications
of
a
server
then
ask
the
right
question
to
the
API,
about
impacts
of
manufacturing
of
those
components
and
then
build
the
total
impact
of
the
machine
on
your
side.
It
will
be
based
on
scaffold
to
have
the
energy
consumption.
This
way
you
can
have
a
global
estimation
of
the
impact
on
all
the
steps
of
the
life
cycle
and
I
will
show
you
briefly
and
then
I'm
done.
A
One
example
of
dashboard
that
you
could
have
here.
You
have
the
global
impacts
of
the
machine
where
the
the
tool
is
running
so
embeddy
Dimensions,
embedded
abiotic
resources,
depletion,
embeddy,
primary
energy
consumed
and
operational
emission
operational
liability
crisis.
Depression
switch
usage,
some
graphs
about
power,
consumption
and
because
it's
related
to
a
CI
pipeline.
A
A
A
A
Another
thing
is,
if
you
heard
about
sentences
like
one
gram
of
CO2
invested
in
ICT
enables
to
avoid
10
gram
of
CO2
that
would
have
been
emitted
without
ICT
or
other
sentences
are
ICT
enables
to
reduce
by
up
to
20
of
GSG
emissions
of
the
other
sectors,
keep
in
mind
that
do
sentences.
They
come
from
industrial
and
companies
reports,
it
come
from
Consortium
like
jsma
or
Jay-Z.
They
are
Consortium
of
companies
that
have
huge
interest
in
ICT
because
they
build
ICT
or
they
or
ICT
is
part
of
their
core
business
model.
A
So
be
careful
with
this.
There
is
no
scientific
paper
that
says:
ICT
has
a
positive
impact,
it
doesn't
mean
it
doesn't
have
one
and
you
can't
have
one:
maybe
it
has
a
room
to
play.
I
just
think
it
has
to
be
defined
and
we
shouldn't
rely
on
something.
That's
not
scientifically
based
to
say
ICT
by
itself
already
has
a
positive
impact.
We
don't
know
that,
after
that
you
have
resources.
I
hope
you
will
have
decides
to
go
further
and
I.
Just
thank
you
for
your
attention.