►
From YouTube: Environmental Sustainability: An Energy-Aware Datacenter Operating Sy... Ying-Feng Hsu & Yuki Sogawa
Description
Don't miss out! Join us at our upcoming event: KubeCon + CloudNativeCon Europe in Amsterdam, The Netherlands from 18 - 21 April, 2023. 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.
Environmental Sustainability: An Energy-Aware Datacenter Operating System - Ying-Feng Hsu & Yuki Sogawa, Osaka University
Data centers are currently using 2% of the world's energy. With the increases in the scale of cloud edge computing systems and other services, data center power consumption is expected to increase about 15 times in the next 10 years. High-performance applications and services are always beneficial, but this does not mean that power consumption needs to increase as a result. This talk covers an energy-aware datacenter operating system (EADOS) that manipulates several types of open-source software to realize energy savings in distributed computing environments. EADOS uses Zabbix and Prometheus continuously collect thousands of data center operating parameters from all servers, sensors, and HVAC cooling systems. Those collected data are further optimized by Apache Kafka stream processing to produce event-driven, machine-learning ready data on a runtime basis, which facilitates the building of various machine learning models. A Kubernetes based workload optimizer use instructions from machine learning models to enable optimal Pods to Nodes assignment to realize data center energy efficiency and service performance. We will use a real data center with 200+ servers to demonstrate this concept to show how EADOS enhances business effects through a lens of environmental sustainability.
A
So
here
is
the
outline
of
my
talk
today
so
first
and
for
the
most
I
will
talk
about
a
little
bit
of
about
ourselves
and
I
will
talk
about
the
the
background.
The
background
of
this.
This
talk,
which
is
the
challenge
of
the
environmental,
suspense
challenge
of
the
power
consumption
on
cloud
Edge,
Computing
systems
and
I,
will
talk
about
the
detail
of
the
purpose
solution,
energy,
aware
data
center
operating
system.
We
name
it
eados.
A
So
at
this
point
you
might
be
interesting
about
so
what
kind
of
data
center
we
are
talking
about
and
what
how
much
data
power
consumption
reduction
we
are
talking
about
here.
So
all
this
will
be
demos
through
a
test
based
data
center
about
200
physical
servers,
so
this
project
is
initialed
and
sampled
by
the
needle
and
expect
to
develop
the
energy
saving
technology
from
osakan
University
and
several
other
joint
research
partners
So
based
on
this
oven,
Consortium
alignments.
A
So
I'd
like
to
share
about
it
our
Milestone
a
quit
a
little
bit.
Our
research
group
has
been
working
on
developing
power,
consumption
reduction
related
technology
in
the
data
center
in
2019
we
created
two
data
center
in
the
Osaka
site.
After
that
various
of
the
technology
has
been
conducted
carry
on
in
this
data
center,
such
as
the
immune,
Cooling,
cfd,
and
machine
learning
for
power,
consumption
prediction
and
data
center
power
reductions.
A
Recently
we
moved
our
work
through
a
workload
location
and
it
is
based
on
the
kubernetes
East,
and
this
work
was
accepted
in
the
cncf
last
year
in
North,
America
and
later
on.
It
also
be
reports
on
a
technical
block.
So,
as
we
know,
the
environmental
sustainability
of
cloud
computing
is
indeed
a
crucial
concept.
A
So
in
this
year
quarterly
we
start
our
NATO
project
and
almost
the
same
time
in
the
cncf
there
is
a
environmental,
substantial
ability.
Technical
advisor
group
Tech
group
was
create
inside
the
cncf,
so
this
Tech
group
was
created
to
support
and
assist
Cloud
native
related
project
that
related
to
Environmental
sustainabilities.
A
Type
of
application,
type
of
insurance,
infrastructure
and
type
of
cloud
so
usually
in
general,
business
related
application
consume
much
more
energy
than
content
manager.
Applications
and
physical
machine
comes
much
more
than
the
virtual
machine
and
for
the
type
of
cloud
public
clouds,
because
they
have
more
control
and
it
is
more
reliable
and
less
power
consumption
than
private
cloud.
A
So
now
I
will
explain
the
technical
We,
Run
The,
debating
of
this
architecture.
I
would
like
to
use
a
video
processing
service
as
an
example,
so
it
is
starting
from
the
great
box
on
the
upper
left.
Then
the
proposed
Eid
always
is
reside
in
the
middle.
Then
there
is
a
cloud
computing
resource.
The
hardware
part
in
the
blue
Prague
so
inside
the
orange
block
layer
are
actually
free
of
A
and
B
and
C.
A
So
for
the
process,
one,
when
the
video
processing
service
start,
it
sends
to
service
definition
to
the
server
a
and
in
the
same
time
we
collect
the
power,
consumption
and
other
information
from
a
server
B,
so
all
works
together.
Information
will
be
sent
to
the
server
C
and
Microsoft
service
Arrangement
at
air
conditioning
also
including
into
the
service
So,
eventually
service.
We
evaluate
the
air
conditioning
of
power,
server
power
and
the
response
type
of
information
based
on
the
evaluation
function,
and
then
they
extend
the
new
parameter
control
parameter
back
to
the
cloud
environment
So.
A
Eventually,
everything
is
optimal,
optimized
and
can
be
reduced.
The
total
power
consumption
in
the
black
cloud
environment.
So
by
repeating
this
kind
of
five
process,
sequentially
for
of
each
load,
energy
saving
can
be
realized
as
a
result.
In
addition,
power
consumption
and
the
response
time
model
can
be
flexibly
controlled
based
on
the
service
rate
requirements,
so
this
kind
of
rates
come
to
a
marketing
will
lead
to
high
Return
of
the
business
and
also
the
environmental,
substantial
ability.
A
So
now,
let's
talk
about
the
actual
information
of
this
process
eados
and
how
it
works
through
the
goal
of
the
data
center
energy
reduction.
So
I
will
talk
about
starting
from
the
data
collection
and
how
we
built
the
machine
learning
model
and
how
we
use
the
Machine
model
to
build
the
workload
location
and
eventually
reach
the
goal
of
the
data
center
energy
reductions.
A
So
we
built
our
testbed
data
sensor
in
different
structure,
and
that
is
why
we
want
to
ensure
the
purpose
of
eidos
Works
in
different
environment.
As
an
example
of
these
two
text,
Data
Center,
as
you
can
see
from
the
diagram
they
have,
they
are
different
and
they
have
different
kinds
of
cooling
airflow.
So
we
focus
a
lot
for
the
air
conditioner
optimization
because
it
is
the
largest
single
unit
come
soon
about
25
to
40
percent
of
total
power
consumption
in
the
data
center.
A
A
So
the
way
we
collecting
the
data
in
in
this
style,
so
this
is
a
example
of
one
of
our
data
center.
It
comes
contains
about
200
servers,
so
we
collect
all
of
kind
of
operating
parameters
which
is
around
4000
per
minute
and
we
use
the
premises
and
zapix
to
collect
the
data
so
because
of
this
kind
of
big
data
so
put
in
the
mini
scale.
Sometimes
there
is
a
duplicated
and
some
data
description
issue
or
we
need
to
be
fixed.
A
All
of
this
data
can
be
categorized
to
be
three
category.
One
is
the
temperature
sensor,
so
it
tells
about
different
temperature
inside
the
data
center
and
we
also
collect
the
detailed
parameter
of
the
air
conditioners
because,
as
I
mentioned,
it
is
the
most
important
device
in
the
data
center.
Most
of
data
is
actually
comes
from
the
server
because,
for
example,
there
are
200
servers
and
each
server
has
8
to
16
fan
so
200
times
16.
A
So,
as
you
can
image
this
kind
of
Big
Data,
when
we
want
to
use
it
to
build
a
different
prediction
model,
it
will
be
time
consuming
because
the
origin
data
is
correct
in
real
time.
I
need
the
BB
pre-processed
before
it
can
be
used.
So
recently
we
are
starting
to
move
in
to
use
the
Kafka
platform
and
by
using
the
cop
car
stream
faucets
and
the
ksql
DB,
so
everything
can
be
wrong
in
a
real
hand.
For
example,
we
create
a
many
different
of
Kafka
topics.
A
A
Okay,
so
after
the
collecting
the
data,
that's
what
I
want
to
mention.
We
put
a
lot
of
effort
for
machine
learning
technique
and
because
the
accurate,
accurate
person
power
prediction
model
will
be
lead
to
the
result
of
optimal
task
location
So.
Eventually,
with
the
accurate
and
reliable
power
prediction
model,
the
test
allocation
can
be
realized
and
achieved
the
power
consumption
reduction
of
the
data
center,
so
I
think
sorry.
Next
I
would
like
to
have
a
sokawa.
A
B
Thank
you,
Professor
Sue
I
would
like
to
talk
about
pre-consumption,
Update
Center
and
its
production
model,
air
conditioning
equipment
per
consumption
and
server
equipment
for
consumption
account
for
most
of
their
power
consumption.
In
a
data
center
to
optimize
per
consumption
near
death
Center,
it
is
most
important
to
reduce
the
upper
consumption
of
these
devices.
B
Another
operation
that
can
be
performed
on
the
server
equipment
is
to
determine
which
servers
to
press
the
workers
on
to
optimize
per
consumption.
In
your
data
center,
it
is
necessary
to
know
how
power
consumption
changes
when
operations
such
as
controlling
air
conditioning
and
server
task
placement
are
performed.
B
B
In
addition,
since
the
per
consumption
of
air
conditioning
is
greatly
affected
by
the
outside
temperature
and
humidity,
this
information
is
also
used.
Since
detailed
information
on
air
conditioning
equipment
is
not
available
to
the
server
administrator.
The
Cyber
CPU
utilization
server,
ambient
temperature
and
static
pressure
difference
between
the
front
and
back
frames
are
used.
B
The
static
pressure
difference
is
information
equivalent
to
the
amount
of
Afro
through
the
server
since
per
consumption
is
a
characteristic
that
varies
from
device
to
device.
A
pre-consumption
model
must
be
created
for
each
device.
However,
once
a
model
is
built
for
one
device,
it
can
be
used
universally
for
that
device,
so
the
model
is
built
using
only
information
that
can
be
easily
obtained.
B
I
will
discuss
how
we
built
the
model
I
just
described
on
kubernetes,
we
have
deployed
mfo,
which
allows
us
to
easily
compare
the
results
of
our
experiments
using
mfo.
The
results
of
the
experiments
can
be
checked
and
shared
from
a
browser,
and
the
results
can
be
aggregated
from
a
python.
Script
is
equipped
with
the
standard
function
to
record
the
results
of
experiments
in
ml4,
so
it
is
possible
to
conduct
sufficient
experiments.
B
Working
with
mfro
with
a
very
small
amount
of
code
file
is
shared
by
each
node,
such
as
experimental
data
and
training
data
are
placed
on
the
file
server
and
download
it
to
each
bot
using
FTP.
When,
therefore,
model
training
is
started,
the
file
server
is
also
used
as
strange
for
experimental
data
that
mfo
refers
to
by
establishing
such
an
environment
in
a
laboratory.
We
are
able
to
easily
conduct
various
experiments
in
the
research
process
in
the
unified
model.
It
is
also
very
useful
for
sharing
the
results
of
experiments.
B
Comparing
models
using
pieces
can
be
done
with
just
a
few
written
programs.
This
graph
shows
the
schedule
of
errors
in
each
model
when
training,
data
for
CPU
utilization
and
then
temperature
static
pressure,
difference
and
per
consumption
for
each
server
in
this
experiment.
Models
are
created
for
each
service
data,
so
the
number
of
models
that
must
be
created
is
large
and
requires
the
use
of
multiple
computers.
B
A
A
little
bit
of
time,
but
I
will
take
another
couple
minutes
to
go
to
the
FAQ
time
to
go
to
the
couple
more
slides,
so
so
I
want
to
jump
to
the
how
we
built
the
workload
locations.
So
we
extend
the
kubernetes
schedule
to
One
schedule
and
we
extend
the
low
balance
to
our
low
balances.
So
in
this
sense
the
power
consumption
can
be
controlled
in
the
priority
and
also
resulting
in
time
can
be
controlled
in
advance.
So
we
can
also
be
balanced
between
the
power
consumption
and
the
responding
type.
A
So
this
is
the
how
it's
collecting
the
data
and
build
machine
learning
in
the
wrong
time.
So
we
have
set
up
the
experience
as
the
demo.
So
in
this
experience
it's
based
on
the
data
center
with
200
servers
and
it
is
for
using
a
object,
detection
applications.
So
we
convert
our
work
schedule
and
what
will
be
together
with
comparison
with
the
default
code.
A
A
So
in
this
talk
we
talk
about
the
challenging
of
the
increasing
power
consumption
of
cloud
Edge
Computing
system,
and
we
propose
the
energy
aware
data
center
operating
system,
so
the
best
result
from
our
system
by
using
the
object,
detaining
example
is
with
successful,
reduce
13
of
the
power
consumptions.
So
the
last
but
not
least,
environmental
substantial
ability
is
just
starting
to
gain
a
lot
of
attention
inside
the
cncf
community.
A
Reducing
the
carbon
hydrogen
is
a
social
responsibility
that
everyone
needs
to
be
involved,
so
that
is
what
we
are
here
for,
and
we
will
continue
to
work
closely
with
the
cncf
community
to
achieve
the
final
product
of
this
Eid
OS.
So,
finally,
I
would
like
give
a
special
thank
you
for
our
project
support
sponsor
NATO
for
the
great
support
of
This
research
project.
Okay,
thank
you
for
joining
us
today.
So
if
I'd
like
to
take
any
question
from
here.
D
D
A
A
Okay,
so,
as
you
know,
we
focused
on
the
data
center
two
data
centers
in
different
structures.
So
this
is
the
first
step
and
even
though
different
data
center
is
the
power,
consumption
is
all
from
the
server
air
conditioner
and
surrounding
area.
So
we
focusing
on
the
both
server
part
and
air
conditioning
parts
so
that
we
can
ensure
the
Opera
country
consumption
can
work
in
different
environments
and
assign
you
my
thinking
about,
sometimes
the
not
in
most
of
time.
We
cannot
control
the
air
conditioner
in
the
most
data
center.
A
This
is
the
isp's
responsibility,
but
this
project
is
dedicated
for
different.
So,
even
though
you
are
using
the
housing
or
call
location,
you
can
still
control
your
own
server,
even
though
you
cannot
control
the
air
conditioner,
but
you
can
still
detect
the
surrounding
ambulance
temperature
of
the
server
and
that's
why
you
can
build
automatic
your
services.
But
if
you
are
the
service
provider-
and
you
all
also
run
the
applications
like
Google
Amazon,
that
kind
of
big
player,
then
you
can
reduce
more
power
consumption.
A
A
Okay,
this
is
actually
a
very
good
question,
so
the
walkthrough
actually
is
the
application
from
the
user
request
and
in
in
the
past
we
starting
from
very
beginning,
we
use
the
stress
comment
which
come
soon
it
stretch
comment
is
the
heavy
duty
task
which
comes
with
all
the
CPU
usage,
and
so
you
you're
right.
We
needed
to
know
the
application
first
and
we
need
need
to
manipulate
the
different
application
in
advance,
but
I
think
some
application,
for
example,
is
like
a
business
oriented
content
management.
They
have
a
similar
Behavior.
A
So
that's
why
we
focus
on
building
many
different
machine
models
in
the
future.
We
want
to
have
more
advanced
way
to
build
machine
learning
models
based
on
different
categories
of
application,
different
process
of
the
software
applications.
So,
even
though
there
are
a
lot
of
different
applications,
but
I
think
we
believe
there
still
can
be
categorized.
So
that's
why
we
can
build
different
categories
of
machine
learning
models
and
our
Focus
will
try
to
build
a
machine
learning
model
as
quick
as
accurate
as
possible.