►
From YouTube: Kubernetes is Doomed! — Anne Currie 1.1.9
Description
The direction of travel of climate-intelligent hosting will spell the end for Kubernetes in its current form. Can the community save it? An overview of the issues currently being addressed for green hosting and what the big players are doing. Can Kubernetes keep up? Is it even possible for the orchestrator to survive in the new world?
A
B
B
A
B
Right
well,
my
name
is
anne
curry
and
I
have
been
in
the
tech
industry
for
depressingly
enough
nearly
30
years
now-
and
I
was
I
used
to
work
on
high
performance
c
backhand
servers
these
days.
I
do
something
a
little
bit
different
and
we'll
talk
to
talk
I'll
mention
that
later
in
the
talk,
but
my
aim
today
is
to
talk
to
you
about
the
direction
of
travel
of
operations
and
architecture.
B
Now
the
cloud
don't
worry,
don't
panic,
I'm
not
about
to
say
it's
the
cloud,
because
that
would
be
crazy
talk.
The
cloud
is
what's
completely
revolutionized
architecture
and
operations
in
the
past
10
years,
but
I
don't
think
that
that
in
and
of
itself,
not
not
directly
is
what's
going
to
be
pushing
our
direction
of
travel
in
the
next
decade.
B
I
think
in
the
next
day
decade
it's
all
going
to
be
about
the
climate
and
it's
not,
and
also
don't
worry,
I'm
not
about
to
tell
you
that
this
is
that
your
data
center
at
all
about
to
burn
to
the
ground
or
be
washed
away
into
the
sea
or
anything
like
that,
hopefully,
within
the
next
10
years.
That's
that
won't
happen.
B
But
what
will
happen?
What
we
know
will
happen
in
the
next
10
years
and
we'll
have
a
significant
effect
on
us
is.
That
is
how
the
cloud
is:
architecturally,
responding
to
climate
change
and
the
effects
of
climate,
because
for
the
past
five
years
or
so,
I've
been
involved
in
campaigns
to
move
the
cloud
over
to
sustainable
hosting
and
in
2020.
B
We
had
an
amazing,
an
amazing
step
forward
in
that
which
I'd
I'd,
love
to
say,
win.
But
let's
face
it.
It's
not
jeff
bezos
woke
up
one
morning
and
decided
he
was
going
to
do
it.
It
had
absolutely
nothing
to
do
with
anybody
else,
but
in
2020
all
of
the
major
cloud
providers
committed
to
being
carbon
zero
by
2030.
B
So
now,
in
nine
years
time-
and
this
isn't
some
kind
of
airy
fairy
half-assed
carbon
neutral
commitment
because
they're
all
there
most
of
them
already
there
already
so
google
cloud
has
been
carbon
neutral
since
2007,
which
is
an
extraordinarily
long
time.
Azure
we're
a
bit
later,
but
they've
been
they've,
been
carbon
neutral
since
2014.
B
C
B
Harder
target
to
meet
you
can
be
in
many
ways:
carbon
neutral
just
means
you're
tracking,
what
carbon
you're
emitting
and
then
you're
paying
for
offsets
of
one
form
or
another,
and
that's
it's
nice
that
you're
doing
the
tracking
and
it's
nice
that
you're
willing
to
shell
out
a
little
bit
of
money
to
buy
the
offsets.
But
offsets
are
not
that
useful.
What
we
actually
need
in
the
long
run
is
a
commitment
to
not
emit
any
carbon
as
a
result
of
your
operations,
so
carbon
zero.
B
Is
that
commitment,
no
carbon
from
electricity
and
no
carbon
from
hardware
production?
That
is
a
ridiculously
difficult
goal,
and
I
have
to
say
in
all
the
time
that
I
was
campaigning
for
this
and
and
all
of
us
a
lot
of
us
were
campaigning
for
this.
I
think
we
were
oh.
We
were
ridiculously
naive.
We
all
thought
that
once
the
cloud
committed
to
be
carbon
zero,
if
you
were
running
on
the
cloud,
then
you
would
just
become
carbon
zero.
B
I
don't
think
that
in
and
of
itself,
they
can
do
it
so
their
first
step-
and
this
is
really
good.
I
can't
criticize
them
for
doing
this.
Their
first
step
is
that
they
have
spent
huge
amounts
of
money
getting
more
renewables
into
the
into
the
national
grids,
so
they've
put
shed
loads
of
money
into
bill
either.
B
Building
wind
farms,
solar
farms
themselves
and
piping
the
the
power
directly
into
their
own
data
centers,
but
more
commonly
they
have
signed
up
to
be
guaranteed
purchases
of
power
coming
from
third-party
wind
farms
and
solar
farms,
either
directly
with
the
the
companies
building
those
farms
all
with
the
within
that
with
the
local
grids
and
they've
spent
a
lot
of
money
on
this.
So
over
the
past
decade,
google
has
commonly
been
the
biggest
non-government
purchaser
of
renewable
power
in
the
world.
B
If
the
clouds
decide
that
they're
all
going
to
be
the
run
on
on
sustainable
power,
then
all
they're
going
to
do
is
suck
all
the
sustainable
power
out
of
the
grids
and
they
won't
be
any
for
anybody
else.
And
it's
you
know,
there's
it's
a
zero-sum
game.
We
haven't
improved
things
at
all,
but
in
fact
that's
not
what
they're
doing
they're
they
are
guaranteeing
that
more
renewable
power
goes
into
the
system.
So
that's
all
great.
B
So
that's
all
good
and
I
kind
of
in
my
old
dream,
my
old
dream
world,
where
I
was
feeling
very
positive
about
this.
I
thought
that
would
be
all
that
was
required,
but
that
was
really
because
I
hadn't
properly
thought
things
through,
because
there
is
a
significant
problem
with
the
renewable
power.
Well,
some
renewable
power
there's
a
significant
problem
with
wind
and
solar,
and
that
is
that
sometimes
it
rains,
sometimes
it's
even
the
night,
wind
and
solar.
Sometimes
it's
not
windy.
B
Wind
and
solar
are
naturally
variable
available,
variably
available
power
sources.
They
are
not
pumping
out
power
to
you
all
the
time.
You'll
there'll
be
times
when
there's
tons
of
power
and
times
when
there
is
no
power
or
very
little
power
now,
and
that
isn't
the
way
that
data
centers
work,
most
data
centers
need
power
all
the
time.
Your
applications
mostly
need
to
run
all
the
time,
and
there
are
places
where
you
could
have
data
centers,
where
that
would
be
the
case
and
the
power
would
still
be
renewable.
B
I
mean
obviously,
if
you're
running
it's
the
reason
why
we
it's
the
reason
why
we
still
have
fossil
fuels
after
all
of
these
years
and
we've
known
for
you,
know,
half
a
century,
that's
about
climate
change
and
about
the
greenhouse
gas
effect,
but
fossil
fuels
are
so
damn
good.
They
are
reliable,
they're,
cheap,
they're,
easy
to
transport.
B
This
there's
a
reason:
it's
not
craziness
that
we've
we've
been
so
reluctant
in
moving
to
renewable
power
power
sources,
and
there
are
places
in
the
world
where
you
won't
have
to
worry
so
much
about
this.
There
are
places
where
you
could
put
your
data
center,
where
you
would
have
renewable
power
that
was
reliable.
B
So
maybe
you
could
be
in
france
where
there
were
loads
of
nuclear
power
or
you
could
be
in
iceland,
where
they've
got
loads
of
geothermal
or
you
could
be
in
canada
where
they've
got
loads
of
hydro,
but
that
isn't
everywhere
and
that
isn't
where
most
data
centers
are
because,
let's
face
it,
most
data
centers
are
in
the
east
coast
of
the
usa.
So
there
is
there's
always
there
is
going
to
be
a
huge
problem
with
reliable
sourcing
of
power
for
data
centers.
B
A
B
Because
I'm
not
one
of
these
communist
climate
activists,
I
think
that
capitalism
works
quite
well,
but
it
has
to
be
incentivized.
If,
if
we,
if
the
government's
subsidized
that
subsidizes
bad
energy
production,
then
there's
there's
no
reason
why
anyone
should
do
anything
about
it.
There's
no
reason
why
we
should
start
to
adapt
to
a
world
of
variable
power.
So
I
really
hope-
and
I
think
that
it
will
be
the
case
that
we'll
get
electricity
pricing
more
widely.
B
They
have
it
in
spain
at
the
moment,
but
that's
that's
kind
of
the
the
forerunner
of
this
kind
of
thing,
so
it'll
be
useful.
It'll
be
helpful.
We
think
it's
all
inevitably
coming,
but
it's
probably
not
going
to
solve
absolutely
well.
It
probably
will
solve
things
for
us
in
the
long
run,
but
we,
it
won't
just
transparently
solve
it-
for
us,
it'll
incentivize,
to
make
incentivizers
to
make
the
kind
of
operational
and
architectural
changes
that
we'll
need
to
make
to
take
advantage
of
variable
electricity
pricing.
B
B
So
from
solar
from
wind,
because
that
really
is
the
work
the
workhorse
I
would
personally,
I
would
love
it
if
we
were,
if
we'd
all
gone
to
nuclear-
and
we
just
didn't-
have
to
worry
about
this
kind
of
stuff,
but
we
didn't
and
we're
not
likely
to
have
quite
some
time.
So
I
think
we
have
to
grok
the
world
that
we're
in
is
one
in
which
the
power
will
be
variably
available.
B
I
know
that
there's
loads
of
storage,
everybody's,
desperate
for
storage,
but
all
of
the
storage.
All
of
the
storage
solutions
at
the
moment
are
sub-optimal
they're,
expensive,
they're,
not
they're,
not
a
like-for-like
replacement
for
oil,
even
even
hydrogen,
which
is
kind
of
that.
The
and
shell
would
like
to
believe
was
the
like
flight
for
replacement
for
oil
is
not
as
good,
it
will
cost
more.
So
I
think
we
have
to
accept
that
we're
going
to
be
handling
some
variable
availability
of
power.
B
So
how
are
I
mean?
I
wonder
how
much
they'd
thought
about
this
really
thought
this
through
before
they
committed
to
their
zero
carbon
deadline.
Last
year,
some
of
them
certainly
had
google
certainly
had,
but
I'm
seeing
a
lot
of
activity
now
it
suggests
so
that's
that
they've
upped
their
game
a
little
bit
and
they're
trying
a
little
bit
harder
on
this.
B
So
they
produced
a
very
interesting
paper.
It
came
out
in
june
or
july
about
the
next
generation
of
scheduling
and
programmatic
orchestration.
Now
this
is
kubernetes
conference.
So
we
all
know
that
google
have
been
the
kings
and
queens
of
programmatic
orchestration,
the
forerunner
to
kubernetes
and
they've,
used
it
to
physically
shift
encapsulated
tasks
around
in
their
data,
centers
to
increase
server
density
and
use
less
energy
to
do
the
same
work,
which
is
all
great
for
for
climate
change.
B
But
it's
not
enough.
We
need
to
actually
have
a
bit
of
a
paradigm.
Another
paradigm
shift
in
how
we
do
this
and
at
their
next
stage
I
think
they're
trialing
at
the
moment
is
more,
is
the
trying
harder
to
shift
tasks
in
time
and
they
call
this
temporal
displacement
and
because
they
are
googlers
and
they
do
like
their
sci-fi
and
it's
interesting,
because
what
they're
doing
is
in
many
ways
very
similar
and
just
an
extension
of
what
they've
done
in
the
past.
B
What's
what
their
schedulers
and
then
their
old
programmatic
orchestration
did
and
in
some
ways
it's
the
complete
opposite.
So
the
similarity
is,
it
requires
encapsulated
tasks
that
are
well
labeled
and
they
move
them
around.
It
just
happens
to
be
moving
them
in
time
rather
than
just
in
in
location.
So
you
know
we
live
in
a
4d
world.
Why
not?
B
But
in
some
ways
it's
the
exact
opposite
of
what
they
were
doing,
because
what
they
were
doing
before
was
they
were
trying
to
pack
their
workloads
onto
machines
as
tightly
as
possible
so
that
those
machines
were
maximally
used.
They
got
really
high
server
density,
really
high
utilization,
and
that
meant
everything
was
more
efficient
and
it
meant
that
you
got
more
out
of
your
machine
during
its
lifetime.
So
you,
you
didn't
use
up
too
much
electricity
and
your
hardware
lasted
longer.
B
But
the
idea
is
less
energy
used
when
there's
no
good
energy
to
use.
Now
there
are
various
things
that
they
need
in
order
to
accomplish
this.
So,
as
I've
said,
they
need
good
orchestrators,
which
they've
got
schedulers,
which
they've
got
that
actually
encapsulated
tasks
which
they've
got,
which
is
all
good,
but
you
also
need
tasks
that
are
lower
priority
and
not
as
urgent
that
don't
have
quite
as
high
an
sla
on
as
a
rocket
and
sla
and
when
they're
running
they
don't
have
to
be
run
immediately.
B
They
can
be
delayed
deferred.
They
can
be
delayed
until
the
sun
comes
up
and
they
have
that.
But
we
don't
always
have
that.
So
that's
one
of
the
things
that
they
pointed
out
in
this
in
this
paper
is
that
they
can
run
this
and
they
can
get
this
to
work
and
actually
that
they're
well
I'll
put
links
to
the
paper
in
my
twitter
feed.
B
So
you
can
have
a
read
if
you
want,
but
their
initial
results
are
a
bit
rubbish
but
as
they
point
out,
it's
their
first
go
and
they
expect
it
to
get
better.
And
we
hear
in
the
grapevine
that
the
results
are
getting
much
much
better
as
they
move
on
as
they
move
forward.
But
they
can
only
run
all
of
this
on
their
own
workloads.
They
can't
run
it
on
the
google
cloud
and
the
reason
is
that,
within
their
own
workloads,
they
know
a
lot
about
those
workloads.
They
know
which
ones
can
be
deferred.
B
They
know
an
awful
lot
about
about
how
urgent
they
are.
But
when
you
move
to
you,
when
they
look
at
the
vms
in
the
google
cloud,
those
are
just
black
boxes.
They
have
no
idea
what
they
can
defer.
So
they
can't
do
any
of
this
clever
stuff
on
that,
and
that
is
going
to
be
an
issue,
because
this
temporal
displacement
is
going
to
be
required.
If
we're
going
to
handle
variable
power.
B
B
You
you've
wrapped
your
application
in
in
it
effectively
in
an
enabled
aws
spot
instance,
and
you
said,
run
it
when
you
got
the
chance
that
gives
aws
and
aws
orchestrators
and
and
schedulers
the
chance
to
balance
that
and
run
it
when
they
can
and
take
account
of
everything
else.
That's
going
on
in
the
system,
so
their
schedulers
have
more
information
than
your
local
kubernetes
scheduler
would
do
and
you've
told
it
that
you've.
B
Given
a
lot
of
information
about
that
task,
you
said:
look
I
don't
mind
when
it
runs
just
make
sure
it
runs
at
some
point
and
as
a
result
of
that
you're
paying
a
lot
less
money,
and
that
reflects
the
fact
that
actually,
these
boss
instances
help
aws
and
the
other
providers
also
provide
them
to
make
their
systems
more
efficient
and
it
cuts
the
cost
for
them.
It
has
a
a
a
lower,
marginal
marginal
cost.
B
It
seems
a
little
a
little
perverse
for
me
to
say
that
kubernetes
is
doomed.
Given
that
kubernetes
is
a
programmatic
orchestrator
with
scheduling,
but
I
think
there
are
some.
There
are
a
couple
of
really
quite
terminal
problems
with
kubernetes
at
the
moment
when
it
comes
to
this
new
world
of
advanced
temporal
shifting
scheduling
the
one
of
them
this
is
potentially
not
even
the
main
one
is
it's
that
there
is
not
enough
variety
of
tasks.
B
So,
if
you're
running
kubernetes,
the
likelihood
is
you're
using
it
to
manage
and
enterprises
worth
of
workloads
and
google
themselves
point
out
in
their
paper
that
that
isn't
really
enough
tasks
to
to
get
really
good
utilization
to
get
the
efficiency
up
to
the
level
that
they
need
in
order
to
meet
this
zero
carbon
deadline
is
they
need?
They
need
massive
variety
of
tasks
of
tasks,
sizes
task,
priorities,
task
urgencies,
and
you
just
don't
have
that.
So
that's
a
bit
of
a
problem.
B
Your
workloads
are
probably
not
labeled
enough.
You
probably
don't
have
enough
higher
priority
and
low
priority
workloads
anyway,
but
even
if
you
do
they're
not
sufficiently
labeled
and
but
the
first
one
will
probably
be
very
hard
to
fix
the
second
one
is
fixable,
especially
10
years
to
do
it.
The
third
one
is
totally
fixable,
but
you're
gonna
need
to
start
demanding
it.
B
B
They
are
too
inefficient
and
if
a
tool
is
going
to
be
running
all
the
time
as
as
these
tools
nest,
necess
by
necessity,
have
to
do,
those
tools
are
going
to
have
to
be
super
efficient
and
at
the
moment
that
has
not
been
a
requirement,
but
it
will
be
over
the
next
year,
10
years,
it's
going
to
become
an
absolute
requirement
and
you
need
to
be
pushing
back
to
your
suppliers
and
saying,
okay,
I
want
service
messages.
There
are
a
lot
more
efficient
lawns
that
are
being
offered
to
me
now.
B
B
You
need
to
be
architecting,
architecturally
thinking
in
terms
of
high
and
low
priority
tasks,
and
I'm
using
that
literally
high
and
low
priority
tasks,
not
high
and
low
importance
tasks
it
it's
a
classic
of
management
that
you
have
important
tasks
and
urgent
tasks
and
that
they
are
not
the
same
thing
just
because
a
task
is
low
priority.
It's
not
urgent,
doesn't
mean
it's
not
important.
B
Make
sure
that
you
can
do
that
and
encourage
your
developers
to
start
thinking
in
terms
of
designing
architecting
to
use
spot
instances,
because,
with
a
spot
instance
it's
it's
that
they
can
be
incredibly
efficient
and
they
can
be
used
by
the
cloud
providers
to
to
to
help
them
move
towards
those
those
zero
carbon
goals
in
a
way
that
stuff.
That's
that
you're
managing
yourself
that
they
can
never.
They
can
never
do
that
as
well.
B
So
I
said
distributed
systems
work
well
with
if
you're
splitting
tasks
out,
if
you
are
carving
application,
carving,
microservices
off
your
monolith,
start
thinking
in
terms
of,
can
you
carve
off
applications
that
have
a
consistent
sla
associated
with
them
in
the
right
now?
Spots
are
the
most
obvious,
but
I
think
there
will
be
more
sophisticated
spots
in
the
future
that
that
may
be.
That
will
allow
you
to
say
well.
I
need
this
to
run
within
the
next
hour.
B
I
need
this
to
run
three
times
a
day,
a
bit
of
a
a
halfway
house
between
spots
at
the
moment
and
systems
that
that
you
just
have
entirely
under
your
own
control.
Think
about
what
regions
some
regions
are
much
more
energy
have
much
more
sustainable
energy
in
them
than
others.
We
talked
a
little
bit
about
that
earlier
targets.
Your
efficiency
improvements
at
things
that
have
to
that
are
urgent,
that
you're
going
to
have
to
run
like
that.
B
That's
where
you
really
need
those
efficiency,
improvements
and
efficiency
improvements
are
a
pain
in
the
neck
and
take
ages
and
yeah,
so
so
target
them
all
the
stuff
that
you
can't
stick
in
a
spot,
because
if
you
can
stick
it
in
a
spot,
it
doesn't
need
the
efficiency
improvement
as
much
and
make
sure
you're
using
the
engine
devices
and
if
you're
gonna
be
lacking
power,
those
devices
have
power.
If
they
don't
have
power,
then
they
won't
be
logging
into
your
systems.
So
almost
by
definition,
the
device
has
some
power
to
work
with.
B
B
B
They
are
more
effort
and
you
do
have
to
think
about
them,
but
you
know
it's
quite
a
fun
thing
to
think
about
crazy
stuff
like
that
and
because
actually
in
10
years
time,
that
is
the
situation
you
will
be
in
I'm
afraid,
unless
we
completely
discover
some
amazing
new
storage
technology
during
that
time,
which
I
think
is
unlikely.
So
that's
that's
your
main
takeaway
from
this.
The
second
takeaway
is
my
job
these
days,
as
well
as
talking
to
conferences
and
doing
techy
things.
B
A
That's
brilliant
thanks,
ann
I'd.
Do
a
quick,
a
quick
recommendation
for
the
book
as
well,
because
I've
read
it.
It's
very
good.
I'm
working
my
way
through
the
rest
of
the
series,
so
I'm
very
excited
we
hadn't
got
any
questions
posted
in
the
slack
channel,
but
I
did
have
one
question
so
I'll.
Just
ask
my
own
question.
You
mentioned
about
efficiency
of
tools
and
you
specifically
called
out
service
measures
as
being
pretty
inefficient.
A
B
There
isn't,
although
it
is
one
of
the
things
that
so
I'm
involved
in
the
green
software
foundation,
the
the
microsoft
and
linux
foundation
group
to
start
giving
advice
on
on
how
to
write
greener
software
and
and
how
to
operate
greener
software,
and
this
is
one
of
the
things
we'll
be
looking
into.
So
what's
what
should
you
be
doing?
You
know,
what's
putting
putting
a
little
bit
of
pressure
on
the
on
the
writers
of
these
tools
to
make
sure
that
they
are
more
energy
efficient
in
the
future.