►
From YouTube: ESPA Module 3B - Architecture Pt. 2
Description
Benjamin Højsbo (VP of R&D) with PiKNiK discusses Architecture Part 2 (of 2) at the Enterprise Storage Provider Accelerator (ESPA) bootcamp week that took place in February 2022. This discussion includes operating system, storage architecture and hardware, power and renewable energy considerations, and data center considerations.
Data is growing at an incredible speed and much of this data is archived and/or simply lost by enterprises. Our program will accelerate net-new Enterprise Storage Providers into the ecosystem, using a Web3 protocol with an impressive incentive model called Filecoin.
Learn more at:
Sign-up: https://web3espa.io
Landing Page: https://m.fil.org/espa-bootcamp
Follow ESPA:
Twitter: https://twitter.com/web3ESPA
LinkedIn: https://www.linkedin.com/company/web3espa/
A
Is
it's
not
really
that
much
about
that
part
about
the
hardware
itself,
so
we
are
still
in
the
architecture,
so
we
are
a
little
bit
more
high
level
than
just
going
down
into
the
like
the
the
different
cpus
and
everything
that
we
use
for
this.
Obviously,
I'm
also
into
that
you
know,
but
here
the
architecture
is
a
little
bit
more
high
level,
and
so
I'm
I'm
going
to
talk
about
a
couple
of
different
topics.
A
On
top
of
what
alex
just
talked
about
and
as
a
picture
shows
up
here,
storage
is
one
of
them.
So
basically
I
would
like
this
to
be
a
little
bit
more
interactive
because
generally,
when
we're
talking
about
stuff
like
like
storage
and
and
data
centers,
and
so
on,
you
know
this
is
also
something
that
is
really
much
up
to
you
to
figure
out.
How
would
you
like
this
to
look
so
so?
A
A
So
that's
one
thing:
the
other
thing
is
like
low
choose
does
not
have
a
gui,
so
the
only
one
where
you
would
interact
with
with
this
is
by
a
cli
so
being
into
the
space.
This
should
not
be
a
big
problem,
but
but
if
it
is
it's
at
least
good
to
know
that
this
might
you
know
if
you
come
from
a
like
a
window-centric
organization?
This
will
be
very
different,
so
the
the
thing
is
that
what
we
see
is
like
yeah.
A
So
I
kind
of
divided
this
into
two
different
areas.
So
one
here
is
about
the.
What
would
you
like
to
do
with
the
storage
provider?
So
now
we're
talking
about
the
servers
and
in
the
community,
we'll
probably
see
consensus
around
using
ubuntu,
and
you
can
obviously
you
know,
choose
another
distribution
if
you
like,
we
also
have
a
lot
running
on
on
debian.
A
The
reason
why
we
say
that
there
is
consensus
and
why
we
brought
it
up
here
is
like.
If
you
go
into
the
community
channel
and
try
to
debug,
why
is
it
doing
something
riding
errors
and
so
on?
You
will
see
a
lot
of
the
others
in
here
are
running
ubuntu,
so
at
least
for
starting
it
might
be
easier
for
debugging
to
to
start
there,
but
obviously
it
could
be
that
you
know
you
have
your
own
reasons.
A
Decisions
on
what
distribution
you
use
so
another
point
up
here
is
like
we
want
to
run
everything
headless,
meaning
that
when
we
install
when
we
do
the
installations,
we
won't
have
the
desktop
experience
the
the
gui
on
on
the
os.
There
are
a
couple
of
different
reasons
why
so
one
really
annoying
thing
about
about
using
using
linux
and
using
a
desktop
is
this
whole
thing
about
the
drivers
for
the
gpus
and
yeah?
A
Well,
it
also
updates
the
video
driver,
video
driver
recompiles
to
a
new
kernel
or
whatever,
and
guess
what
it
needs
to
restart
before
the
driver
works
again
and
what
happens
to
your
window
post
yeah,
you've
flown
it.
So
it's
like
that's
just
things
that
happens
on
this
that
you,
you
know,
take
that
into
consideration
right
away
that,
although
that,
for
some,
it
might
be
more
like
easy
just
having
the
desktop
you
can
you
can
google
things
from
there
and
just
type
it
in,
but
it's
like
in
reality,
it
might
just
make
more
issues.
A
The
other
thing
here
that
I
put
in
is
also
that
we
want
to
keep
the
least
amount
of
packages
on
there,
because
you
know
having
hundreds
of
of
security
updates
and
updates
for
all
kind
of
packages.
You
that's
just
a
lot
of
noise
compared
to
that,
we
basically
only
want
to
run
the
central
services
needed
to
to
do
the
storage,
providing
thing
yeah
and
then,
basically,
the
issue
around
having
a
desktop
environment
is
like
this
needs
to
go
into
scripting.
A
So
in
the
ideal
world,
also,
when
you
start
with
your
first
minor,
you
don't
just
you
know,
sit
and
type
in
stuff,
and
you
know
chit-chat
with
your
miner.
Basically,
this
would
be
in
a
script
and
you
can
you
can
you
know,
take
it
from
there
and
and
punch
it
in,
but
you
basically
need
this
stuff
to
be
documented.
So
that's
also
why
it's
actually
not
a
bad
thing
that
we
have
to
run
everything
in
command
line.
A
If
something
goes
wrong,
we
need
to
know
you
know
what
happened
and
if
that
was
by
you
know,
clicking
around
setting
up
something.
It's
like
yeah.
We
can't
really
trace
back
what
was
going
on,
so
that's
really
yeah
it's.
This
is
just
to
say
that
you,
if,
if
you
come
from
a
gui
world
well,
this
will
be
different.
A
What
this
is
trying
to
say
is
like
if
we
just
take
it
up
one
level
and
say
like
from
an
architectural
perspective,
what
types
of
storage
do
we
actually
need
in
a
system
like
this?
I
kind
of
divide
it
into
three
different
types
that
has
like
a
different
specifications,
different
demands
and
if
we
just
take
it
by
the
left
side,
the
long
term
storage
does
not
need
much
speed.
A
So
we're
not
caching
any
data
here
we
don't
need
like
a
high.
I
o
for
for
read,
write
while
the
data
is
processed.
So
I
that's
why
I
kind
of
pointed
it
like
it.
Is
it's
not
high
speed?
Obviously,
it
would
be
high
capacity,
meaning
that
you
know
you
want
the
highest
capacity
for
the
for
the
cost
on
these
systems,
because
in
the
end
we
are
storage
providers
we're
going
to
as
a
storage
market
where
we
would
have
to
yeah.
A
A
Each
deadline
about
30
minutes,
so
two
partitions
could
take
between
five
to
fifteen
minutes
and
if
you
have
a
slow
storage
system
and
it
takes
about
15
minutes
doing
two
in
each
deadline.
You're,
like
yeah,
you
have
pretty
much,
no
yeah,
no,
no
error,
yeah
left.
So
it's
basically
that
that
is
why
you
know
you
still
have
to
have
some
performance
out
of
the
system.
A
So
the
long
term
storage,
the
issue
there
is
like,
probably,
if
you're
doing
a
ton
of
ceiling,
you
have
a
lot
of
data
ingesting
into
the
system.
At
the
same
time,
it
needs
to
do
window
post.
So
that's
basically
the
critical
task
right
there.
It's
like
you're
gonna,
get
that
query
on
hey.
I
need
to
do
my
window
post.
So
now
I'm
going
to
check
the
sectors
a
little
bit
of
random
data
here
and
there
a
lot
of
the
issues
that
we
see
around.
This
is
like.
A
It
can
be
an
issue
for
these
systems
that
you
ingest,
you
know
very
large
amounts
of
blocks
of
data
onto
the
system
and
then
the
window
post
process
comes
along
and
says:
hey.
I
just
need
to
pick
and
choose
a
little
bit
here
and
if
that
gets
delayed.
Well
it
doesn't
have
that
much
time.
So
if
each
of
these
small
queries
every
time
has
to
go
in
the
back
of
the
queue
to
wait
to
get
served.
Well,
you
end
up,
maybe
not
taking
50
minutes,
but
it
could
actually
be
that
it
takes
a
way
more.
A
A
Then
we
have
the
thing
in
the
middle,
which
is
like
the
the
the
market
storage.
So
the
market
storage
is
basically
where
the
deal
data
comes
in.
So
when
I'm
downloading
an
online
file
from
a
client
who
wants
to
store
it
on
my
storage
provider,
it
needs
to
go
somewhere
and
basically
it's
just
the
raw
data
getting
cash
there
and
just
waiting
to
getting
ingested
into
the
ceiling
cluster
and
the
reason
why
speed
is
a
little
bit
higher
is
because
you
could
have
many
deals
coming
in.
A
At
the
same
time,
it
needs
to
to
check
the
data
it
needs
to
import
the
data,
so
that
is
processing
actually
running
the
data
through
and
then
need
to
export
it
and
send
it
over
to
the
at
peace
worker.
So
normally
for
something
like
this,
you
would
go
with
something
else
than
just
the
regular
hard
drives.
A
Basically,
because
you
have
a
lot
of
of
I
o
going
in
and
out
at
the
same
time,
but
it
doesn't,
it
doesn't
really
need
to
be
like
nvme
type,
because
it's
not
like
it's
not
really
extreme,
it's
more
like.
If
you
just
rely
on
hard
drives,
it
basically
might
get
overwhelmed
by
all
the
different
requests
at
the
same
time.
A
So
that's
a
little
bit
in
the
middle
and
basically
it
says
the
critical
task.
Well,
that's
obviously
true
to
be
sure
that
you
can
import
the
data.
If
importing
goes
too
slow,
then
you
might
you
know
starting
to
build
up
data
coming
from
the
client
which
can
be
imported
and
sent
off
to
the
ceiling
worker,
so
it's
kind
of
just
piling
up
and
we
don't
want
that.
A
The
last
time,
the
last
in
the
very
right
is
the
ceiling
scratch
space
and
alex
talked
about
this
before
so
the
scratch
space
is
kind
of
the
most
demanding
type
of
storage
that
we
have.
Basically,
what
happens
is
like
when
we
do
the
pc1
process.
The
reason
why
we
go
from
say,
32
gig
of
data
and
then
into
those
almost
500
gig
is
because
we
are
producing
in
the
pc
one
process.
We
are
producing
11
layers
of
the
data
and
that
basically
takes
up
like
11
times
those
300
and
yeah.
A
It
ends
up
being
330
gigs
of
data,
and
then
you
also
have
the
unsealed
copy,
and
you
have
some
other
caching
data
and
so
on.
So
it
also
adds
up
and
for
each
of
these
layers.
Basically,
what
happens
is
that
once
it's
done
with
one
layer
in
pc1,
it
needs
to
write
that
to
storage,
so
it's
gonna
wait
until
that's
written
before
it
can
continue
on
to
the
next
layer.
A
So
if
your
scratch
space
is
like
really
slow
say
that
you
have
like
one
sata
ssd
and
that's
all
well,
it's
just
gonna
sit
there
and
wait
until
all
that
data
was
written
to
the
drive
before
it
can
continue
on.
So
this
is
why
we
need
so
high
a
throughput
here,
and
this
is
why
you
would
probably
see
that
you
know
it's
not
only
nvme.
It's
actually
also
rated
like
stripped
nvme,
so
you
could
have
like
anything
from
a
couple
to
I
don't
know
eight
more.
A
You
could
have
a
whole
cluster
just
to
provide
like
ridiculous
amounts
of
nvme
throughput.
So
what
you
want
to
do
with
this
is
probably
do
some
kind
of
of
pooling,
meaning
that
these
pc1
processes
they
don't
end
at
the
same
time.
So
in
the
ideal
world,
all
of
these
layers
are
not.
They
don't
need
to
be
written
to
disk
at
the
same
time,
because
each
of
the
pro
you
know
jobs
are
running
at
different
times.
So
it's
like
you
want,
through
the
full
throughput
of
the
system.
A
A
So
this
is
about
the
the
long
time,
long
term,
storage
and
again
here,
I
just
put
it
into
three
different
boxes
and
again
this
is
like
this
is
totally
open.
I
would
be
glad
if
some
of
you
said,
like
hey
benjamin,
you
don't
know
anything
about
this.
I
have
a
way
better
model
than
this,
but
but
this
is
basically
just
to
say
well
what
are
we
seeing
out
there
and
what
is
the
kind
of
the
normal
ways
of
doing
this?
A
So
if
you
have
one
of
the
small
systems,
probably
where
we
start,
you
probably
just
have
direct
attached
storage
having
the
drives
inside
the
storage
provider,
computer
or
maybe
just
having
a
jbot
that
is
connected
to
you
super
easy.
You
get
a
really
good
speed
when
you
do
that,
because
you
don't
have
a
any
complexity
from
networking
and
everything.
A
So
when
you
have
say
that
you
have
a
cfs
rate
set
up
and
this
needs
to
re-silver
well,
you
you
will
see,
at
least
in
the
community,
that
a
lot
of
storage
providers
using
that
has
issues
around
the
re-silvering
and
doing
window
posts,
because
the
problem
here
is
like
when
it's
resilient.
It's
basically
doing
a
lot
of
I
o
and
when
windowpost
comes
along
and
say
hey,
I
have
these
small
queries
and
so
on
again
it's
gonna
be
strangled
by
by
the
other
operations
being
done.
A
So
that's
one
of
the
issues
from
running
like
these
small
setups
and
basically
what
from
my
own
setup,
what
I
do
is
like.
I
have
to
go
in
and
basically
time
when
it's
doing
the
scrubbing
and
everything
to
keep
it
outside
of
when
you
have
your
window
post
windows.
So
basically
I
would
have
like
a
couple
of
hours
of
winter
post
and
then
I
know,
after
that
it
can
go
scrubbing,
but
if
it
scrubs,
while
it
has
to
do
window
posts
so
doing
those,
it
will
say,
that's
pretty
sure.
A
So
so
that's
something
that
you
see
in
the
small
systems.
Then,
when
you
start
to
move
up
into
the
bigger
sizes
like
a
multi-pip
and
yeah,
it
could
should
be
two
figure
pick
a
pip
as
well.
You
would
probably
see
people
using
nas
and
just
scaling
out
with
nasa
so
basically
having
nas
heads
having
multiple
jbots
on
each
nav
set
and
just
you
know,
keep
on
growing
like
that
and
again
yeah
you,
you
get
some
more
speed
because
you
you
distribute
it
now.
A
What
is
really
important
here
is
like
how
much
performance
degradation
do
I
have
while
it's
recovering
and
the
beautiful
thing
about
these
systems
is
when
you
can
recover
and
heal
and
do
window
post
at
the
same
time,
because
I
can
tell
you
that
for
some
of
these
one
of
some
of
my
fellows,
that
has
been
like
re-silvering
having
like
all
their
discs
or
all
the
spare
discs
fail.
Well,
one
thing
is
like
you,
you
get
like
super
nervous,
but
the
other
thing
is
like
you
know
that
this
is
gonna.
A
Take
past
24
hours
and
you're
gonna
miss
your
winter
post
again,
so
you,
you
kind
of
won
the
slashing
already
and
that's
just
by
design.
So
it's
like
that's
not
really
an
ideal
situation
knowing
that
yeah.
So
I
miss
a
disc
and
I
get
slashed
so,
but
but
you
know,
this
is
totally
something
that
you
can
design
your
way
out
of.
So
it's
like
it's
good
to
take
these
thoughts
into
consideration
already
now,
because
these
smaller
systems
tend
to
have
these
kind
of
issues
yeah.
A
So
this
was
just
to
say
that
you
know
you
can
you
can
probably
figure
out
where
you
want
to
be
on
this
and
it
makes
sense
not
to
start
out
by
just
doing
the
storage
cluster
right
away.
Probably
there
is
a
learning
curve
here,
but
obviously
before
you
go
and
build
something
that
needs
to
go
into
production-
and
you
know
make
it
big,
you
know
think
that's
right
from
the
beginning
yeah.
A
So
next
up
is
our
nvme.
I
didn't
take
the
market,
the
like
the
market
storage
in
here,
because
it's
a
little
bit,
it's
not
so
easy
to
define
exactly
what
that
should
look
like
right
now.
We
also
talked
about
we're.
Gonna
have
n,
plus
one
in
the
future.
So
probably
there's
going
to
be
some
kind
of
network,
that's
half
stars
as
well:
it's
not
right
now,
so
it
would
be
local.
A
So
it's
like
that
will
probably
be
a
little
bit
more
in
detail
in
the
future,
but
for
the
the
scratch
space
here
we
have
some
different
ways
of
doing
it
and
that's
actually
coming
a
little
bit
back
to
what
alex
said.
So
the
first
one
is
like
what
happens
if
we
just
direct
attach
the
nvmb
in
each
server.
So
basically
just
the
disks
are
in
there.
You
have
a
pc
one
server,
so
that's
the
one
in
the
top
on
the
yeah.
It's
in
the
top
of
all
the
pictures.
A
Let's
say
it's
a
dedicated
pc
one
worker
and
it
just
has
nvme
inside
so
it
will
do
the
pc
one
job
right.
There
then
the
box
beneath
it
has
gpus
inside
you
can
do
the
pc2
and
c2.
So
it
needs
to
move
over
those
500
gigs
of
data
to
do
the
the
pc2
process.
What
happens
is
I
give
these?
If
this
is
not
like
shared
storage
in
any
kind
of
way,
lotus
will
know
and
well
it
will
look
locally
first
and
say:
hey.
A
A
A
I
think,
10
sorry
goes
about
one
gig
a
second,
so
it's
basically
utilizing
a
full
10
gig
network
connection
for
one
for
one
of
these
jobs
and
obviously,
if
you
have
then
many
of
these
machines
and
many
of
these
just
running
well,
you
can
see
that
it
starts
to
pile
up
if
each
of
them
actually
needs
almost
a
10
gig
throughput
to
to
move
it
through.
So
this
can
easily
you
know
start
to
yeah
to
get
become
a
bottleneck
right.
A
Then
we
have
another
way
of
doing
it.
So
this
is
something
that
alex
actually
has
been
doing
for
quite
a
while,
but
you
could
actually
also
just
say
like
instead
of
having
any
disks
in
your
pc2
worker,
then
you
could
basically
just
use
it
over
network
and
just
have
a
network
path
to
your
pc,
one
worker,
because
that's
where
your
data
is
so
now
logos
does
not
do
any
get.
A
So
the
great
thing
about
that-
and
you
can
also
see
that
up
there
and
the
like
the
colors
here-
is
like
we
are
getting
less
wear
from
that
process,
because
the
big
problem
about
having
that
get
process
is
like.
Basically,
first
we
generated
500
gigs
of
data,
and
then
we
move
it
over.
So
we
rewrite
all
those
500
to
another
set
of
so
in
terms
of
where
you
know,
that's
a
really
inefficient
process
and
we're
already
using
very
expensive
nvme
drives
so
like
having
the
double
up
of
of
ride
on
these
drives
is
like.
A
Sorry,
software
defined
storage
to
build
some
kind
of
a
cluster
on
the
nvme,
not
easy
to
do,
but
definitely
something
that
can
that
can
give
really
good
throughput
righty.
So
I
also
have
some
other
topics
that
we
need
to
go
through,
but
that
was
basically
the
part
about
that.
So
I
want
to
move
into
some
other
topics
that
we
need
to
to
look
at
from
an
architecture
perspective,
power
and
renewable
energy
is
one
of
these,
so
actually
it
falcon
in
itself.
A
So
so
what
what
we've
been
doing
in
in
picnic,
for
example,
is
like
we
are
also
on
this
crypto
climate
accord,
together
with
protocol
labs
and
other
companies
there.
What
we're
trying
to
accomplish
is
we're
trying
to
get
away
from.
You
know
all
the
bad
kind
of
the
bad
perception
there
is
about
proof
of
work.
I
don't
personally
see
falcon
as
being
proof
of
work.
I
do
know
that
we
have
a
quite
intensive
sealing
process
reason.
A
A
We
need
to
seal
that,
but
what
we're
also
seeing
is
like
these
ceiling
speeds,
go
down
and
down
and
down
really
rapidly,
and
probably
we
will
also
have
new
features
that
can
do
even
you
know,
faster
speeds
will
probably
also
have
new
hardware
that
can
speed
up
this.
In
the
end,
the
goal
in
terms
of
this
would
be
that
you
have
a
a
very
small
impact
from
the
ceiling
in
itself.
A
A
A
A
A
I
think
it
will
be
almost
like
a
like
industry
standard
now
that
you
have
something
on
that
and
probably
you
can
go
to
some
kodo
side
and
they
will
say
yeah.
We
use
you
know
100
green
energy,
defective.
That's
it
one
of
the
things
that
we're
working
on
in
falcon
green
is
actually
saying:
that's
not
good
enough.
A
So
one
problem
about
doing
the
the
renewable
energy
credits
is
like
you
can
have
data
centers,
saying
guy.
I
run
100
percent
of
hydro
power
because
I'm
paying
I'm
paying
for
that.
Well
guess
what
right
now
is
you're
not
importing
entity
from
there.
So
obviously
you
usually
using
something
else,
and
you
see
this
a
lot.
A
That
kind
of
the
problem
is
that
you
have
a
decoupling
before
between
the
the
consumption
at
the
data
center
and
the
what
the
providers
are
doing
so
like
having
100
wind,
energy
or
or
solar
energy,
and
so
it
doesn't
really
make
totally
sense.
So
one
of
the
things
that
that
we're
working
on
here
is
what
we
call
like
the
energy
tokenization
and
actually
also
you're
using
blockchain
technology
for
that.
A
But
if
you
tokenize
your
energy
consumption,
you
could
basically
have
a
power
provider
saying
like
hey,
I'm
producing
energy
now
who
wants
to
buy
it
and
then
you
would
have
on
the
other
side
data
centers,
just
saying
like
yeah,
I
want
to
consume
one
kilowatt,
one
megawatt
whatever,
and
that
basically
goes
into
these
little
buckets
and
if
I'm
not
able
to
buy
any,
you
know
renewable
energy
anywhere.
Well,
that
would
go
into
my
accounting
and
then
I
could
you
know
say
like
for
90
of
the
time
I
would
be
able
to.
A
You
know,
run
on
in
green
energy,
the
rest
of
the
time
yeah.
There
was
nothing
so
I
basically
had
to
revert
back
to
to
more
black
energy
and
then
it's
all
about.
So
what
can
you
do
about
this
in
the
early?
You
know
in
this
early
phase?
Well,
obviously,
you
could
think
about
putting
your
data
center
somewhere
or
putting
your
choosing
a
co-locator
or
putting
your
data
somewhere
where
you're
near
to
one
of
these
sources,
where
you
actually
can
get
a
green
energy
and
not
a
place
where
it's
like
100
coal.
A
So
that's
one
idea.
You
could
also
look
into
building
like
a
off-grid
on-grid
systems.
Off-Grid
will
probably
be
pretty
difficult
to
run
for
most
and
it's
not
really
a
big
advantage.
So
I
think
in
the
community
we
probably
look
a
little
bit
more
at
the
on
on
on
grid,
because
that
is
more
flexible,
but
basically
making
these
kind
of
solutions
to
make
this
way
more
transparent.
What
is
actually
going
on,
and
one
of
the
big
concerns
here
is
like,
if
we
just
do
what
we
do
today,
it's
basically
saying
yeah.
A
I
bought
some
energy
over
there,
so
it's
like
I'm
all
good
and
in
reality
yeah.
You
got
your
coal
power
plant
right
here,
which
is
producing
the
energy
for
you,
so
you
yeah,
you
can
say
that
you're
buying
it
over
here,
but
that's
not
what's
going
on.
What
we
do
want
is
like
start
to
couple
this,
so
we
can
actually
work
with
real
green.
A
A
If
you
could
use
batteries
to
basically,
you
know
flatten
out
the
usage,
that's
probably
where
you
know
that
could
be
really
interesting
and
from
a
storage
provider
in
itself
say
that
I
put
like
a
smart
battery
next
to
my
miner.
What
I
could
do
is
actually
every
time
that
I
am
not
able
to
buy
green
energy
on
the
grid.
I
could
basically
revert
over
to
battery
and
say
yeah
the
next
two
hours
I
will
just
run
on
batteries.
Once
I
have
green
energy
again,
I
can
jump
back
on.
A
So
these
are
all
ideas.
I
think
it's
just
important
to
think
about
it
like
early
on
now,
because
you
know
you
haven't
been
spending
like
millions
on
a
specific
data
center,
maybe
or
you
might
need
to
build
into
new
sites.
So
it's
like
when
you
have
all
these
considerations.
Well,
this
is
probably
one
of
them.
Nothing
here.
Funny
thing
you
could
also
look
into.
Is
I
that's
something
I
do
personally?
But
what
can
we
do
with
all
the
waste
heat?
A
So
I
have
this
really
funky
idea
that
if
you
are
actually
locating
your
data
center
close
to
district
heating
systems,
you
can
use
these
kind
of
heat
pumps
like
I
have
a
picture
of-
that's
probably
pretty
expensive
in
that
size,
but
it
also
comes
with
smaller
sizes,
but
the
thing
is
like
basically
think
about
it.
If
you,
if
you
say
hey,
I
want
to
actually
be
a
like
a
like
a
a
heat
provider,
so
I
want
to
provide
heat
into
a
district
heating
system.
A
A
So
it's
it's
at
least
an
idea,
and
I
think
this
will
be
coming
there
in
the
future.
Look
at
your
server
park
as
a
fancy
radiator.
But
again
this
needs
to
be
considered.
You
know,
where
is
the
data
center,
because,
obviously,
if
you
put
it
somewhere
where
you
have
no
customers
for
buying
your
waste
heat,
it
makes
no
sense
right.
A
A
So
from
what
I
heard,
this
is
actually
producing
a
surplus
of
energy
compared
to
what
a
switch
here
is
using
as
vegas,
but
obviously
you
come
down
into
these
issues
like
okay.
So
obviously
this
is
not
producing
energy
24
7.
So
what
you
do
when
it's,
when
you
don't
have
the
sunlight
here,
batteries
could
come
into
that
in
the
future,
so
that
one
of
the
solutions,
I
think,
is
currently
being
worked
on.
Obviously
it
could
be
really
cool,
going
totally
off
grid
and
basically
just
saying
that
yeah.
A
A
A
So
alex
said
that
you
have
about
30
seconds
to
do
a
window
per
winning
post,
that's
basically
the
compute.
That's
basically
you
proof
of
work
that
you
do
so.
The
gpu
is
running
for
about
one
second
to
do
your
winning
post,
that's
basically
just
taking
a
little
bit
of
data
and
hashing
it
together
with
some
other
data.
A
The
data
points
here
would
be
stuff
like
the
random
data
that
you
get
from
the
dram.
It
would
be
your
minor
power,
so
your
storage
power,
but
basically
that
hashing
that
goes
on
there.
That
is
basically
when
you
win
the
right
to
to
to
minor
block
and
mining
a
block.
Here
is
the
same
as
all
the
other
blockchains
that
you
basically
put
messages
into
a
block
and
you
distribute
it
out,
and
everyone
else
is
looking
at
your
winning
post
and
say:
yeah.
A
Okay,
this
guy
got
it
and
they're
signing
off
on
that
so
same
process,
but
that
looks
way
more,
like
proof-of-stake
compared
to
proof-of-work.
So
you
could
see
the
work
process.
Here
is
the
sealing
process.
It's
not
mining
blocks.
We
don't
spend
almost
any
energy
on
actually
mining
blocks
and
keeping
that
whole
thing
running
and
also
the
proving,
I
would
say,
running
the
gpu
for
three
minutes
for
each
partition
and
deadlines.
A
That's
not
really,
you
know
doing
a
lot
of
waste.
So
actually
it's!
Basically
you
could
just
look
that.
Look
at
that
as
like
a
consistency
check
as
well.
So
that's
that's
the
green
stuff.
One
thing
is
to
look
at
what
kind
of
a
tier
do
you
want
to
be
in
the
data
center
yeah?
We
talked
about
that.
You
actually
also
have
t4
and
five,
but
then
again,
we
also
talked
about
right
now.
Filecoin
is
probably
mostly
being
used
for
archival
of
data,
so
obviously
your
your
clients.
A
They
would
like
to
have
you
available
when
they
need
to
store
data
with
you.
If
they
asked
to
retrieve
a
file,
they
would
probably
also
like
some
someone
to
respond
at
the
other
end,
so
downtime
is
obviously
not
good,
but
on
the
other
hand,
this
really
comes
down
to
what
kind
of
client
a
relationship.
A
What
kind
of
clients
you
want
to
attract,
and
I
think
it
makes
sense
to
look
at
these
tiers,
probably
does
not
make
sense
to
move
way
up,
because
you
again,
you
have
to
think
about
this-
will
come
down
into
what
is
the
cost
of
your
storage
and
if
you
don't
have
any
customer
who
wants
to
pay
for
it,
because
you,
you
know,
got
the
highest
price
in
the
in
the
market
yeah.
It's
probably
not
a
good
idea,
but
at
least
that's
that's
something
that
the
you
can
look
into.
A
Other
considerations
on
the
data
centers
in
general.
If
you
have
current
facilities
already,
do
they
need
to
be
changed
into
doing
this?
You
would
see
most
of
the
falcoin
storage
providers
in
data
centers.
It
looks
like
totally
normal
enterprise
hardware,
so
rack
servers
pretty
much
everywhere.
Some
of
them
obviously
need
to
have
gpus.
You
will
have
a
ton
of
jbots
storage
servers
but
yeah.
It
will
look
probably
like
most
other
data
centers
and
nothing
special
there.
A
A
A
You
know
there's
a
lot
of
different
things
coming
into
it,
and-
and
my
advice
here
would
just
be
like-
take
that
into
the
consideration,
because
in
the
end,
with
falcon
green,
it
could
actually
be
that
when
you
are
to
be
selected
as
a
start
provider
for
a
client,
they
would
actually
look
at
these
data.
Like
are
you
running
green
energy?
How
much
are
you
using?
Do
you
have
any
other
features?
A
I
think
it's
important
to
say
that
when
you
think
about,
if
you
have
a
client
that
would
require
you
to
have
a
sock
two,
it's
not
enough
that
you
just
go
to
colo
and
say:
oh,
you
got
your
suck
too
sort
of
okay,
so
I
just
put
my
servers
here
and
now
it's
all
good.
Well,
it's
it's
all
changed,
so
obviously
you
would
need
those
guys
to
be
it
but
think
about
all
your
own
processes
as
well,
and
we
talked
about
all
this
stuff
that
needs
to
go
on
lotus.
A
It's
not
a
safe
place,
so
you
need
a
lot
of
processes
to
make
sure
that
you're
not
losing
client
data
that
you
are
not
exposing
client
data,
all
that
stuff,
it's
pretty
common,
it's
nothing
special
to
falcone,
it's
more
like
yeah!
You
definitely
need
to
to
think
about
that
physical
security
also
very
important
one
of
the
issues
that
we
see
around.
If
you
really
want
to
go
into
the
the
kind
of
the
the
extreme
dream
like
building
something
with
the
where
you
have
your
own
solar
panels
or
whatever
it's
like.
A
If
you
want
something
totally
custom,
it
can
be
really
hard
when
you
also
want
to
do
the
enterprise
part,
because
if
you
need
to
build
a
data
center
which
is
like
fully
up
to
like
enterprise
standards
at
the
same
time,
you
place
it
at
a
location
where
it
can
have
all
these
like
features
of
green
energy
and
stuff,
like
that.
Well,
it's
like
that
can
get
a
little
bit
complicated.
A
So
one
of
the
things
we
have
been
looking
into
like
it's
like
you,
go
to
the
code
and
say:
hey:
can
I
put
solar
panels
on
top
of
your
your
building
like
whoa,
and
the
other
thing
is
like
hey?
I
only
want
to
use
a
green
energy.
Can
we
have
like
special
metering
on
the
power
that
I'm
using
like?
No,
it's
not
really
wired
like
that.
So
you
want
us
to
do
like
new
cabling
through
the
whole
building.
A
A
Also,
when
we
go
into
like
redundancy
in
general,
like
a
redundant
power
supply
redundant
internet
connections,
you
need
all
this
stuff.
Obviously
so
so
you
can't
just
put
your
data
center
somewhere
in
the
mountains,
where
you
have
some
natural
cooling
and
everything,
because
probably
you
would
need
a
you
know,
a
really
long
fiber
to
go
there
and
you
would
need
multiple
going
in
different
directions.
So
so
it
could
be
yeah.
A
It
is
worth
taking
into
consideration.
You
know:
can
you
early
early
on
think
this
into
into
the
whole
process
yeah?
So
I
know
it
was
like
a
bit
of
different
topics.
We
got
here
and
we
are
going
to
hardware
later
on.
I'm
not
doing
it.
Joe
is
doing
it,
but
obviously
I
will
be
very
excited
about
it,
but
but
I
hope
that
this
gave
you
kind
of
some
of
the
more
like
high
level
ideas
about
what
is
needed
for
for
doing
a
for
doing
the
storage
provider.
Part.