►
From YouTube: Storage Configurator by Steven Umbehocker
Description
From the 2019 OpenZFS Developer Summit
slides: https://drive.google.com/open?id=0B_J4mRfoVJQRbDBWY0o4RmNuc2FXNDMySWZjd2t1WGlpdmkw
A
Alright,
so
our
next
presenter
is
Steve,
oomba,
haukur,
sorry,
I,
hope,
I
didn't
slaughter,
your
name
he's
the
CEO
of
OS
Nexus,
and
this
is
going
to
be
his
fifth
open,
ZFS
developer
summit.
So
today
he's
going
to
be
talking
a
little
bit
about
the
ZFS
work,
which
is
being
done
at
OS.
Nexus,
give
a
little
overview
on
some
of
the
topics
so
yeah.
Please
welcome
Steve
and
yeah.
B
Thanks
I've
got
a
presentation
today
on
a
tool
that
we've
developed
to
kind
of
help.
People
build
better
ZFS
solutions
at
scale
so
for
companies
that
are
building
stuff
that
are
at
the
rack
scale
we
for
many
years.
We
just
designed
everything
in
Visio,
so
you'd
get
to
be
able
to
see
the
whole
solution
and
then
the
end
user
and
the
distributor
and
everybody
could
go
and
take
a
look
at
a
picture
of
what
they
were
gonna
go
deploy,
but
that
had
a
lot
of
problems
with
it.
B
One
of
them
is
jet
your
it's
really
collaborative
working
with
the
hardware
supplier
and
the
customer
and
figuring
out
hey.
What's
what's
the
right
build
and
if
they
go
and
they
change
one
thing,
then
it
changes
usable
capacity
and
and
and
all
these
different
factors.
So
we
just
we
wanted
a
way
to
kind
of
eliminate
Visio
from
the
equation
and
to
just
just
make
it
a
lot
easier
to
come
up
with
something
in
minutes
rather
than
having
having
somebody
have
to
use
those
tools.
B
So
that's:
where
kind
of
made
a
big
list
of
things
that
was
like
how
do
we
get
rid
of
Visio
out
of
this
equation
and
and
just
make
it
turnkey,
so
it
needed
to
be
some
way
where
it
would
be
a
web
app,
but
also
that
people
could
just
go
and
share
their
designs
with
each
other
and
it
needed
to
cover
our
hardware
compatibility
list
and,
and
our
HCl
is
basically
every
major
vendor.
That's
out
there.
B
B
Granted
using
ZFS
for
so
many
years,
oh
yeah,
there
we
go,
and
so
I
was
trying
to
just
get
a
lot
of
that
knowledge
into
the
tools,
so
that
more
and
more
people
could
be
successful
in
deploying
ZFS
solutions
and
and
take
a
lot
of
the
the
mystery
and
fear
out
of
out
of
it.
And
so
there's
a
kind
of
the
two
main
things
is
right.
As
I
jump
in
I'll
show,
you
show
you
a
demo.
B
The
tool
here
in
the
minute
is
that
you
can
choose
a
use
case
and
you
can
choose
a
capacity
and
it
will
automatically
choose
a
lot
of
good
recommendations
for
that.
For
that,
given
use
case,
we
put
all
the
tools
online
and
there's
no
login
required.
So
you
can
go
to
the
the
tool
just
os
Nexus
comm,
slash
ZFS
designer,
and
that
will
take
you
right
right
to
the
page.
B
C
B
B
B
Can
adjust
the
slider
bar
to
go
meet
whatever
capacity
they
need
and
then
a
lot
of
the
common
use
cases
like
say
they
want
to
do
triple
parity
like
for
a
cold
archive.
They
can
now
just
start
putting
those
things
in
there
and
adjusting
it.
So
all
the
servers
that
we
go
in
and
and
test
ZFS
with
and
and
do
the
H,
a
failover
testing
are
all
in
here.
So
you
can
just
pick
a
different
server
or
pick
a
different
j-bot
and
put
it
all
all
together.
B
Here's
like
the
the
Seagate
they're
106
bays
and
then,
as
you
adjust
the
capacity
it
just
starts,
adjusting
the
design
it's
going
in
these
big
increments,
because
we've
got
this
set
to
maximum
or
minimum
of
four
jbods.
But
you
can
just
kind
of
give
it
the
full
range
and
then
it
will
go
in
and
adjust
that
in
quanta
store.
We
automatically
detect
the
jbods
and
do
a
lot
of
hardware
integration
to
show
it
in
the
UI
and
we
stripe
across
the
jbods
for
fault
tolerance.
B
So
if
you
lose
a
j-bot
you,
you
won't
have
any
downtime
to
the
pool,
and
so
some
of
these
settings,
like
the
large-scale
cold
archive,
automatically
pick
the
minimum
number
of
jbods
to
go
and
match
that.
So
some
of
the
other
things
is,
let's
say
you
know
your
system
integrator,
they
tell
you
it's
gonna
cost
X
amount
per
server
or
X
amount
per
drive.
B
You
can
start
to
put
that
data
right
in
here
and
then,
as
you
move
the
slider
and
adjust
things
it
goes
and
does
all
the
math
to
tell
you
what
the
price
per
terabyte
for
the
hardware
is
so
makes
it
real
easy
for
them
to
figure
out
okay
as
I
scale.
You
know
what
what
does
this
look
like
and,
as
you
can
imagine,
if
you're
doing
a
really
small
pool
with
a
lot
of
hardware,
the
price
per
terabyte
for
your
hardware
is
a
lot
higher
than,
as
you
start
getting
out
out
to
scale.
B
B
A
B
In
the
design
of
the
build
and
so
yeah
this,
this
utility
is
just
basically
eliminated
about
ninety
five
percent
of
all
the
Visio
work
that
we
that
we
do,
and
it
just
saves
a
huge
amount
of
time.
So
now
you
can
kind
of
think
about
different
ideas
and
and
run
them
by
the
end
user
or
a
system
integrator.
And
what
you
can't
see
here
is
at
the
very
top
inside
of
the
browser
window
is
a
URL
that
has
all
this
data
in
it.
B
C
B
Kind
of
futures,
for
this
is
well
before
that,
so
the
tool
really
helps
eliminate
a
lot
of
these
Franken
builds.
You
know
the
it's
terrible
when
you
see
someone
using
a
consumer-grade
SSD
for
just
about
it,
for
anything
and
and
it's
unfortunate,
sometimes
where
you
know
in
a
che
configurations.
Of
course,
everything's
got
to
be
SAS
and
you
want
it
to
be
sad
anyways,
and
this
just
starts
to
put
them
on
the
right
path.
So
the
design
tool
doesn't
let
you
pick
bad
choices
in
terms
of
Drive
types.
B
So
a
lot
of
these
these
problems
have
just
sort
of
go
away
as
long
as
we've
got
a
design
that
that
that's
that's
within
the
tool
and
then
so,
where
are
we
going
with
it?
So
performance
is
sort
of
the
next
phase
of
this
is:
how
do
we
go
and
quantify
the
performance
as
you
go
and
start
adding
more
and
more
hardware
to
a
build
and
your
goal
is
to
get
to
10
gigabytes
per
second.
Where
do
you
stop?
B
Do
I
need
to
buy
a
gazillion
dollars
worth
of
hardware
or
can
I
stop
when
I've
got
four
servers
and
four
pools,
and
so
what
we're
working
on
next
is
to
make
some
tools
to
quantify
sequential
performance
and
I
ops
and
then
start
putting
that
into
a
database.
That
can
then
let
us
start
building
some
equations.
So
the
goal
is
to
start
getting
synthetic
benchmarks
out
of
running
tests
on
enough
different
systems
that
are
similar,
that
that
we
can
start
to
say.
B
They
can
start
to
do
that
and
not
have
to
really
go
run
their
own
performance
tests
and
stuff
like
that,
it's
not
gonna
perform
its.
You
know.
Performance
testing
is
a
black
hole,
I
mean
when
you
get
down
into
it,
there's
so
many
different
parameters,
but
what
we're
looking
for
is
to
have
it
land
somewhere
in
the
ballpark
so
that
they
have
a
starting
point.
That's
that's
gonna,
get
them
a
something
it's
it's
within
within
their
the
right
parameters
for
their
their
needs
and
use
case
so
kind
of
futures.
B
Okay,
I'm
gonna
buy
this.
It's
gonna
get
cabled
like
that.
There's
there's
no,
no
mystery
to
it!
So
we're
gonna
open
source
the
utility
as
well
and
put
it
on
github.
It's
not
out
there
yet,
but
I'm
trying
to
refactor
the
code.
I'm,
not
real,
proud.
It's
not
I'm,
not
a
JavaScript
guy,
but
trying
to
make
it
a
little
cleaner
before
we
put
it
out
there
yeah.
So
that's
it
any
questions.
C
B
It's
all
just
vanilla,
JavaScript,
it's
actually
not
too
much
code
and
I've
been
just
sort
of
trying
to
refactor
it
a
little
bit.
So
it's
a
bit
more
readable
I
had
some
help.
I
hired
out
the
kind
of
the
starter
seed
at
that
project
and
then
kind
of
took
it
over
about
a
year
ago
and
and
just
been
chipping
hacking
away
at
it
here
and
there,
but
yeah.
B
C
B
We
don't
today
we
capture
a
lot
of
logs
that
that
give
us
everything
from
smart
information
and
things
like
that,
and
it
lets
us
kind
of
analyze
the
configurations
that
are
out
in
the
field
and
see
when
people
have
like
misconfigured
things
like
set
it
to
sync
disabled
or
something
like
that
for
a
test.
And
then
they
forget
to
turn
it
back
to
sync
standard
before
production
we've
seen
that
happening
before,
but
but
no
we
don't.
We
don't
have
like
the
Backblaze
type
analysis
of
the
drive,
endurance
and
MTBF
stuff
yeah.
C
B
It
doesn't,
but
that's
a
great
idea
have
to
check
a
chat
with
you
afterwards,
and
that
would
be
a
good
thing
to
have
the
it
does
have
in
there
a
see
if
I
can
get
back
to
back
to
it.
It's
got
a
thing
where
you
can
indicate
how
much
you
expect
is
gonna,
be
the
overhead,
so
you
can
start
to
reserve
some
extra.
So
if
you
think
that
there's
gonna
be
a
lot
of
padding
happening,
so
you're
gonna
get
a
lot
of
wasted
space,
then
you
want
to
over
build
the
config
by
X
percent.
B
Yeah,
so
this
that
that
bit
there
on
the
on
the
reserved
free
space
lets
you
just
kind
of
adjust
that
based
off
of
what
you
think,
you're
gonna
need
extra
for,
padding
and
reserve
drive
slots
is
more
for
kind
of
making
the
hardware
configuration
bigger
intentionally,
because
if
you,
you
know
reserved
a
bunch
of
drive
slots,
it's
going
to
need
to
add
more
more
kit.
But
that
might
be
intentional.
Maybe
maybe
the
users
got
all
the
hardware
upfront
and-
and
they
want
to
just.
B
To
an
existing
configuration
they
can
just
do
that.
We
constantly
add
to
the
back-end
database
for
this
new
hardware,
so
as
Seagate
and
Western
Digital
and
all
the
major
manufacturers
come
out
with
their
new
jbods,
those
are
going
into
the
tool
pretty
quickly
and
and
we
yeah.
So
we
do
our
best
to
keep
up
up
up
with
all
that
yeah.
C
B
These
here,
in
order
to
see
like
the
information
on
the
the
estimated
price
per
terabyte
on
the
hardware,
you've
got
to
go
and
put
in
some
numbers
here.
Otherwise
you
won't
see
those
calculations,
so
the
NA
that
you
see
down
there
in
the
bottom
is
is
going
to
start
to
show
now
so
like
this
build
here
you
can
see
the
the
price
per
terabyte
comes
out
to
like
44
bucks,
and
that's
that's
because
we've
got
and
not
so
optimal.
Here
you
see
it
come
down
to
39
this.
B
This
was
an
eye-opener
in
terms
of
price
per
terabyte
on
the
hardware
I,
don't
know
if,
if
you
guys
track
it
closely,
but
the
price
per
terabyte
right
now,
street
price
is
about
25
bucks
and
so
price
for
a
petabyte
of
just
the
drives
is
about
twenty-five
thousand
dollars
so
sort
of
all.
In
for
a
hardware
for
a
petabyte
of
storage,
it's
less
than
forty
thousand
dollars.
Today,
it's
more
like
around
thirty
five,
and
then
you
know
you
factor
in
distributor
stuff.
It's
all
it's
all
all
different,
but
the
it's
it's
pretty!
B
It's
pretty
incredible,
and
next
year
the
eighteen
terabyte
drives
are
going
to
come
out.
That's
going
to
provide
further
downward
pressure.
You're
gonna
see
thirty
thousand
dollars
a
petabyte
in
2020,
so
that
that's
gonna
be
great.
We
do
a
little
bit
of
work
on
shingle
magnetic
recording
drives
those
aren't
quite
those
aren't
ready
really
for
primetime.
We've.
We've
done
some
tests
with
that
with
ZFS,
but
we
haven't
trouble
like
expanding
the
pools.
B
That's
actually
two
servers
right
there
with
all
nvme
drives,
so
we've
been
doing
a
lot
of
work
with
Western
Digital,
it's
a
fantastic
unit,
and
you
know
they
kind
of
designed
that
all
nvme
box
as
something
for
just
doing
high
performance
virtualization,
but
actually
as
a
high-density
box.
It's
fantastic
because
you
get
these
24
Bay's
of
all
dual
ported
nvme
that
we
can
use
in
a
high
availability
configuration.
B
B
Yeah,
you
know
the
D
raid.
This
was
something
I
just
wanted
to
make
a
contribution
to
last
year,
because
you
guys
had
some
some
some
discussion
about
D
raid
and
what
I
want
to
do
with
the
D
raid
map.
Calculator
is
make
it
so
that
you
can
say
these
groups
of
drives
are
in
this
source,
make
subgroups
of
the
drives.
B
So
you
can
say
each
of
those
jbods
represents
so
you've
got
four
groups,
and
now
the
D
raid
can
stripe
across
those
knowing
that
it
needs
to
make
the
sort
of
the
many
raid-z
stripes
on
the
interior
fault
tolerant
across
the
jbods.
It
doesn't
do
that
today,
but
once
we
do
that,
then
we'll
look
at
adding
D
raid
into
quanta
store
and
and
and
doing
that,
but
and
then
we
can
add
it
to
to
the
storage
layouts.
But
we
just
have
to
experiment
it
with
it.
Some
more
yeah.
C
B
Yeah
it
just
that
the
thing
of
it
is
is
that
the
skill
set
at
the
resellers
and
distributors
for
software-defined
storage
generally
goes
down.
Year-Over-Year,
there's,
there's
much
more
specialization,
and
so
what
this
does
is
it
just
empowers
all
those
folks
to
really
start
doing
design.
We
review
it
and
everything,
but
it
just
lets
the
conversations
just
flow
much
more
freely,
because
you
know
there's
no
more
Visio,
nobody
has
to
have
Visio
installed
and
it
and
they
get
to
see
exactly
what
they're
getting
and
exactly.
What's
in
the
hardware
build
we
don't
sell
any
hardware.