►
From YouTube: ZFS past & future by Mark Maybee
Description
From the 2017 OpenZFS Developer Summit:
http://www.open-zfs.org/wiki/OpenZFS_Developer_Summit_2017
A
So
I
wanted
to
introduce
mark,
maybe
mark
mark,
has
never
worked
on
open
ZFS.
So
why
did
I
invite
him
to
give
the
keynote
for
our
conference,
so
mark
mark
has
been
with
us
since
the
very
beginning.
He
started
working
on
ZFS
16
years
ago,
when
it
was
like
I
think
he
was
the
third
or
fourth
person
to
work
on
ZFS
he's
been
working
on
it
ever
since
at
Sun
for
9
or
10
years,
and
now
at
Oracle
for
seven
years
he
worked.
A
He's
worked
on
ZFS
when
it
was
just
a
general-purpose
file
system
in
Solaris.
I
should
preface
this
by
saying:
he's
worked
on
a
ZFS
in
a
lot
of
different
contexts.
So
when
it
was
a
general-purpose
file
system
in
Solaris,
he
worked
on
the
ZFS
storage
appliance,
which
has
generated
over
a
billion
dollars
of
revenue
for
Oracle
and
he's
now
working
on
the
Oracle
cloud
storage
team,
which
is
operating
ZFS
at
a
massive
scale,
so
I
invited
Marc
to
come
speak
to
us
about
both
ZFS
has
passed.
A
B
The
other
branch
of
ZFS
I'll
call
it.
So,
although
one
your
schedule
says
the
this
is
talk
of
ZFS
past
and
future
I
broaden
that
to
say
it's
actually
ZFS
past
present
and
future.
This
is
really
gonna.
Focus
on
what
has
happened
is
the
FS
that
you
may
not
be
aware
of
in
the
Oracle
branch,
if
you
well.
So,
let's
start
with
a
bit
of
the
past
and
I'll
start
from
the
very
beginning
just
to
get
a
placemark.
B
But
I
want
to
sort
of
skip
ahead
quickly
from
there,
because
I
really
you've
heard
for
already
from
from
Jeff
I
know
and
from
Marc
Challon
bomb
in
the
past
and
I'm
sure
from
Matt
a
reminiscing
about
that
those
those
early
days
of
ZFS.
They
were
there
amazing
days,
but
what
I
think
you
may
not
have
heard
about
before
is
what
happened
in
the
history
of
ZFS
after
it
became
a
closed
source
within
Oracle,
so
Sun
Microsystems
was
purchased
by
Oracle
in
2010.
B
B
So
this
is
that
story
of
what
happened
to
ZFS
there
and
the
story
starts
with
fish.
I
know:
there's
some
people
here
from
fish
works
before
Sun
Microsystems
purchased
by
Oracle,
a
new
storage
product
was
developed
within
Sun
Microsystems
and
by
a
group
of
very
bright
engineers
who
wanted
to
take
the
technology
that
was
embedded
in
Solaris
and
build
a
premier
storage
appliance
around
that
technology.
It
was
designed
around
the
ZFS
filesystem
as
the
primary
service.
It
was
leveraged
the
capability
of
Solaris,
both
from
its
stability
perspective,
as
well
as
from
its
observability
perspective.
B
It
was
really
targeted
in
those
early
days
towards
sort
of
a
mid
tier
medium-sized
business,
and
that
was
where
we
were
selling
the
product,
and
at
that
point
and
when
Larry
bought
son,
one
of
the
first
things
he
announced
was
that
I'm
going
to
rebrand
this
thing,
which
was
in
the
backhoe
and
those
days,
called
the
7000
series
storage
appliance
to
be
called
the
ZFS
storage
appliance.
He
wanted
to
highlight
the
fact
the
ZFS
was
the
technology
in
it.
B
B
B
Infrastructure,
so
this
was
replacing
not
only
you
know
the
the
plug
development
storage,
which
is
actually
where
we
first
thought
of
rolled
in,
because
that
was
you
know.
Management
was
okay
with
some
instability
there
that
was
fine,
and
but
eventually
we
went
into
their
when
they
call
the
global
infrastructure
where
they
did
payroll,
etc,
and
that
was
towards
the
end
and
and
in
between
we
even
rolled
out
into
the
customer
infrastructure,
where
network
was
actually
selling
services
that
in
that
required
storage
to
customers
directly.
B
B
We
were
going
in
there
and
we
were
trying
to
force
into
an
ops
organization
that
was
intimately
familiar
with
NetApp,
a
new
model
of
storage,
I'm
sure
many
of
you
who
are
familiar
with
talking
to
customers
who
are
familiar
with
more
traditional
file,
storage
services
and
trying
to
tell
them
and
explain
to
them
how
ZFS
differs
from
what
they're
used
to
and
that
you
don't
need
to
do.
The
things
you
do
when
you
get
a
hot
spot
in
ZFS,
that's
a
good
thing.
B
But
on
top
of
that
this
was
a
new
product
and
there
were
issues
it
was
being
rolled
out
at
a
scale
had
never
been
seen
ever
seen
before,
never
been
rolled
out
this
game
before
and
it
was
growing,
it
was
evolving,
it
was.
It
was
becoming
a
solid
piece
of
software
and
hardware,
but
it
took
some
time
so
in
these
days,
Tiger
teams
became
a
frequent
occurrence
and
by
Tiger
teams,
retire
teams
in
the
store
side.
B
Otherwise
systems
would
lock
up
and
getting
his
nod
of
approval
and
going
through
and
applying
a
patch
that
said
all
right
limit
the
arc
to
this
size
now
for
awhile
until
we
get
this
bug
fixed
overall,
these
were
pretty
ugly
days
at
one
point:
I
remember
exactly
what
year
it
was.
It
was
probably
around
2012
or
so
the
VP
in
charge
of
the
operations
and
an
Oracle
called
meeting.
He
requests
a
meeting
which
she
requested
meeting
with
Larry
and
she
walked
in
with
a
deck.
B
That
said,
was
going
to
explain
to
Larry
why
ZFS
was
failing
in
her
environment
and
had
to
be
replaced
with
a
you
know.
Wasn't
it
up?
We
had
to
go
back
because
ZFS
was
not
meeting
the
need
there
and
famously,
at
least
within
our
team
Larry,
looked
at
her
and
said:
yeah,
there
is
no
plan
B,
you
will
go
with
ZFS.
B
B
Zfs
had
always
been
looked
at,
as
primarily
the
the
core
file
system
within
the
Solaris
operating
system
right
and
that's
what
we
spent
our
focus.
We
looked
at
what
Solaris
and
they
needed
from
us.
We
looked
at
what
its
use
cases
were
in
a
server
environment
and
we
looked
at
what
features
it
needed
in
that
environment
and
we
then-
and
we
were
constantly
developing
to
that
target.
B
B
Matt
depart
the
team
around
that
point
as
well
as
Jeff
and
George.
So
so
it
was
a
was
an
interesting
shift,
and
so
we
were
busily
trying
to
hire
up
some
some
replacement
talents.
We
managed
to
fill
the
gaps
with
about
three
or
four
people
to
one
there
to
fill
that
that
need,
and
even
there
I
don't
think
we
really
fill
the
need.
We
simply
patched
a
leak,
but
we
made
progress
and
we
focused
from
you
know:
development
focus
more
towards
a
stability
focus
within
the
team.
B
So
I
can
remember
having
a
conversation
with
a
director
at
that
point,
walk
in
my
office
saying
you
know,
your
bug,
counts
are
really
high
and
we
really
are
in
a
tight
place
here.
Can
you
shift
your
team's
focus
over
towards
picking
bugs
and
I
said
yeah
I
can
do
that
for
you
we
can
we
can.
We
can
do
that
and
in
fact,
if
we
looked
at
that
point
in
time
around
that
point
in
time
our
bug
churn
rate
was
very
high.
B
We
were
pulling
in
roughly
25
bugs
a
month
new,
bugs
a
month
being
filed
against
us
and,
and
we
had
to,
we
had
to
keep
on
top
of
it,
and
so
it
really
was
a
you
know,
fix
a
bug,
fix
a
bug,
fix
a
bug,
sort
of
mentality
for
at
least
a
short
period
of
time,
and
that
in
this
space,
as
they
creeped
in
from
the
use
case-
and
you
can
see
it
varied
quite
a
bit-
and
this
chart
showed
actually
skyrocketing,
at
least
at
one
point.
But
overall
we
did
get
a
handle
on
it.
B
You
know
the
ZFS
team
got
handled
on
the
bug
situation.
We
even
managed
to
reduce
our
overall
bug
counts.
We
were
actually
at
at
a
high
of
about
400
outstanding
bugs
and
the
ZFS
software
product
at
the
time
in
2010,
when,
when
we
started
this,
and
within
two
years
we
had
marriage
to
lower
that
to
about
two
hundred
bugs,
and
even
with
an
incoming
rate
of
roughly
twenty
five
a
month,
we
were
maintaining
about
two
hundred
bugs
about
sending
bugs,
and
so
it.
A
B
From
from
that
perspective,
and
we
kept
on
top
of
the
bug
situation
pretty
effectively
and-
and
we
weren't
just
doing
bugs,
so
we
were
still
doing
future
development
now.
Some
of
you
know
the
old
timers
would
probably
be
a
little
bit
frustrated
by
the
fact
that
they
were.
There
was
a
lot
of
bug,
fixing
going
on
and
feature
development
was
a
little
bit
on
the
side,
but
you
we'd
snuck
it
in
and-
and
we
were
able
to
do
more
and
more
as
time
went
on
as
we
got
as
the
product
stabilized,
and
things
got
better.
B
We
were
definitely
doing
more
and
more
of
that
and
and
performance
was
also
dramatically
improved
of
these
years.
When
fish
works
first
introduced
sort
of
an
aside,
the
the
ZFS
storage
appliance
or
the
7000
series
storage,
they
had
introduced
it
with
this
concept
of
the
16
metrics
I
think
it
was
16.
B
I
might
have
the
number
wrong,
but
it
was
a
a
observation
that
you
know
there
were
these
industry
standard
benchmarks
out
there,
but
they
weren't,
really
representative
of
storage
and
they're
kind
of
being
exploited
by
the
storage
vendors,
and
we
don't
really
want
to
play
that
game.
Instead,
we're
gonna
find
our
own
set
of
metrics,
which
are
more
representative
storage
and
really
represent
what
you
need
and
what
you
want
to
see
in
a
storage
product
that
made
sense,
but
it
fell
flat
because
customers
didn't
care
about
our
16
metrics.
B
But
we
were
actually
pleasantly
surprised
when
we
eventually
decided:
okay,
let's,
let's
go
get
into
the
game
and
let's,
let's,
let's
play
that
game
and
I'll
get
to
a
slide
here
in
a
second.
It
shows
that
it
shows
you
our
blood
counts.
So
this
is
a
graph
that
actually
tracked
blood
counts
between
what
July
2008
through
July
2012,
and
you
can
see
that
we
had
that
spike
of
around
400
outstanding
bugs
in
2010
and
and
slowly
chipped
away
at
it
and
by
early
2011.
A
B
I
mean
it
obviously
fluctuates
up
and
down
over
time,
and
these
are
a
smattering
of
some
of
the
new
features
I
know.
Many
of
these
will
probably
look
familiar
to
you.
I
know
that
some
of
these
have
been
also
implemented
in
the
open,
ZFS
environment.
I.
Think
this
highlights,
though,
that
this
is
a
different
trunk,
so
things
did
proceed
a
little
bit
differently
over
here,
similar
directions.
We
took
our
difference.
Some
features
that
might
look
the
same
are
probably
significantly
different
in
the
way
they're
implemented
within
the
product.
B
I
think
some
of
the
more
interesting
features
on
this
list
are
the
asynchronous
data
set
deletion.
So
we
moved
a
deletion
into
the
backgrounds,
a
stickers
process
that
was
important
to
us
from
a
product
perspective,
a
storage
plants
perspective
because
we
were
going
into
environments
where
there
was
this
demand
that
deletion
B
isn't
aeneas.
So
I,
don't
know
all
of
you
who
are
familiar.
All
of
you
are
familiar
with.
B
Cfs
know
that
deletion
is
a
is
we
we
actually
in
the
short
period
of
time
you
wait
we're
actually
going
through
and
cleaning
out
all
the
space
maps
and
and
freeing
up
those
blocks
before
we
turn
back
to
control
back,
and
that
was
just
too
long
to
wait
for
these
customers,
even
though
we
still
had
to
do
that
work,
and
so
we
implanted
an
async
destroy
that
that
made
them
happy,
even
though
the
space
didn't
would
come
back
over
time
that
they
didn't
care
about.
That.
B
We
also
did
a
lot
of
work
and
a
lot
of
the
features
up
here
are
all
around
performance,
so
changes
to
improve
the
block
allocator
a
significant
rewrite
of
the
arc
to
change
a
number
of
the
elements
of
its
architecture.
It
certainly
has
the
same
functionality
as
I
had
before,
but
it's
it's
organized
in
a
fairly
significant
way.
These
days,
the
improved
Numa
locking
we
discovered
in
a
number
of
places,
as
we
scaled
things
larger
and
larger.
B
We
were
at
this
time
scaling
because
we
were
not
also
I
should
point
out
not
just
developing
for
the
appliance
which,
by
the
way,
was
continually
increasing
its
processor
power,
but
we
were
also
continuing
to
develop
for
SPARC
architecture,
and
that
was
pushing
us
to
our
limits
in
a
lot
of
places
with
its
multi-threaded
behaviors.
So
we
were
having
to
scale
our
locking
to
be
able
to
keep
pace
with
the
processor
and,
what's
amazing
is
we
could
I
mean
it's
amazing
to
see
some
of
our
performance
numbers
on
on
some
of
the
SPARC
hardware?
B
It's
it's
it's
sad
that
it
isn't
deployed
more
in
that
environment
and
then
sequentially
silver
was
added
some
time
ago
and
and
IO
throttle
control,
which
I
think
might
be
interesting
to
some
it's.
It
was
put
in
place
again
from
a
storage
appliance
perspective
to
allow
us
to
be
able
to
throttle
specific
client
access
cloud
access
to
specific
data
sets
and
it's
primarily
used
by
our
customers
in
a
VM
environment,
where
you're
deploying
VMs
to
a
lot
of
clients-
and
you
want
to
avoid
the
noisy
neighbor
problem.
B
So
you
can
place
throttles
across
all
your
VM
say.
This
is
how
much
bandwidth
you're
allowed
your
max
max
out
at
so
no
one,
no
single
client
can
dominate
all
right
so
back
to
to
performance
around
2014.
We
sort
of
made
our
first
foray
into
this
SBC
landscape.
We
actually
released
24
7420
model
and
published
numbers
I
think
he's
one
of
the
first
numbers
we
published
in
this
space.
I
thought
looking
back
through
my
history.
This
is
your
list
when
I
could
find
anyway.
It
could
be
that
we
had
some
earlier
numbers.
B
B
So
we
were
very
pleasantly
surprised
that,
with
you
know
appropriate
amount
of
tweaking
to
things
that
we
could
be
very
competitive
in
this
space,
and
so
we,
from
this
point
on
pretty
much
routinely,
would
publish
results
that
put
us
near
the
top
of
the
heap
for
storage,
and
could
you
do
so?
Actually,
our
recent
numbers
in
at
least
the
SPC
to
inspect
ifs
benchmarks,
are
very,
very
respectable
all
right.
B
Finally,
in
this
sort
of
historical
context
of
what
was
happening
between
2010
and
today,
Oracle
has
moved
strongly
into
cloud
I,
don't
know
if
you're
aware
of
this,
but
Oracle
now
claims
to
be
pretty
much
a
cloud
company
it
despite
I,
think
claims
early
in
2010,
I,
remember
Larry,
saying
he
didn't
know
what
this
cloud
thing
was.
It
wasn't
gonna
be
important,
but
but
Larry's
nothing
adaptable,
and
so
we
started
at
Oracle
with
software-as-a-service.
B
This
was
really
an
outgrowth
of
service
services
that
there
okla
been
providing
actually
for
a
long
time.
Remember
those
Austin
data
center
slide.
I
showed
you
earlier
there
on
that
slide.
Was
this
comment
that
it
was
supporting?
You
know
thousands
or
actually
at
least
hundreds
of
of
customer
applications.
B
Oracle
has
for
a
long
time
supported
hosting
customer
database
and
middle
and
middleware
within
data
centers
for
customers,
and
fact
that
was
some
our
biggest
headaches.
That
VP,
who
went
to
Larry
I,
try
to
kick
us
out
of
the
data
center
was
was
primarily
concerned
with
the
customer
environment
because
that's
where
she
was
getting
beat
up
every
time
we
went
down
for
a
period
of
time
within
that
environment
that
impacted
customers,
and
so
there'd
have
to
be
these
awkward
phone
calls
to
the
customer
saying
yes,
I'm.
B
Sorry,
your
services
went
down
for
and
hours
while
we
dealt
with
us
whatever
issue
that
had
come
up,
so
you
know,
Oracle
evolved
that
into
or
maybe
just
rebranded
data,
perhaps
into
software
as
a
service.
All
right
so
saw
fro
service
was
run
primarily
on
engineered
systems.
The
Exadata
system,
if
ever
you
are
familiar
with
that,
it's
a
piece
of
hardware
specifically
designed
to
run
database,
but
it
was
also
coupled
with
a
need
for
storage
and
that
storage
was
provided
by
ZFS.
So
a
ZFS
was
deployed
quite
early
in
that
environment.
B
For
the
storage
needs
of
the
SAS
environment,
this
is
still
the
fastest
growing
part
of
Oracle's
cloud,
as
far
as
I
know
is
the
SAS,
because
of
course,
this
is
what
Oracle
does
best
its
database
and
middleware
its
fusion
in
around
2013
Oracle
purchase
company
called
nebula.
Nebula
was
infrastructure
for
platform
as
a
service.
They
deployed
that
in
around
20,
fifteen
I
think
early
2015,
if
I
recall
correctly-
and
this
is
still
what
nimbala
sorry,
what
what
Oracle
offers
as
paths
and
it's
cloud.
It's
based
off
of
that
nebula
technology.
B
This
is
the
second
fastest
growing
piece
of
Oracle's
cloud
and
probably
some
point
will
overtake
SAS
as
as
its
premier
service
in
this
environment.
This
is
platform.
So
what
the
needs
here
for
storage
is
primarily
once
lends
to
populate
the
instances,
and
so
ZFS
is
the
source
of
those
all
those
ones
in
this
environment.
So
OPC,
really
the
OPC
by
the
way
is
Oracle
public
cloud
is
is
run,
is
firmly
on
top
of
ZFS
storage
at
the
moment.
B
So,
let's
fact
transition
to
the
present
and
talk
about
what
it
looks
like
now
in
in
the
world
of
Oracle
and
ZFS.
The
most
current
ZFS
storage
is
the
GS
five
series.
As
you
can
see
this
is
you
know.
The
key
to
our
success
has
been
very
large
memories.
We
were.
We
were
for
the
first
out
there
to
pump
up
the
size
of
our
memory
from
you
know.
B
B
B
Remember
that
I
told
you
early
that
GFS
when
we
had
a
small
foothold
in
the
market
in
the
early
days
in
2010.
Well,
today,
we
are
trusted
by
most
fortune
500
top
fortune
100
using
cloud
financial
services,
telco
semiconductor
oil
and
gas
media
entertainment
companies
across
the
board,
all
right.
So
it's
it's
doing
very
well.
As
I
think
Matt
mentioned
the
the
sales
of
this
product
over
its
current
lifespan
have
been
over
a
billion
dollars
and
it's
maintaining.
B
B
So
it's
not
even
an
issue
of
yes,
we've
tuned
this
way
or
that
way
to
address
a
specific
market,
although,
admittedly,
we
do
have
tunings
for
specific
markets.
Ok,
the
reality
is
that
it's
still
a
fully
functional,
very
broad
spectrum
file
system,
storage,
appliance
supporting
a
very
broad
spectrum
of
workloads.
B
B
What
does
that
supposed
tell
me
I'm
done
in
10
minutes.
Are
you
yanking
me
out
of
here?
All
right
life
is
good
right,
life
is
is
pretty
good.
Life
is
definitely
not
bad,
but
I
see
warning
signs.
The
warning
signs
I
see
is
that
cloud
and
cloud,
storage
or
particulars
are
race
to
the
bottom.
All
right
when
you
deploy
storage
and
cloud
you're,
trying
to
sell
it
at
a
fraction
of
the
cost
of
really
what
it's
worth
in
some
places,
but
the
ZFS
storage
appliance
is
an
enterprise
storage
appliance.
B
It's
designed
to
supply
it
and
applies
needs,
and
so
it's
priced
above
what
you'd
normally
place
cloud
storage
at
now,
I
think
we
can
do
some
things
to
address
that
and
you
can
reconfigure
your
storage
to
deal
with
that.
I'll
talk
about
that
in
a
little
bit,
but
at
the
moment
I
think
you
know.
Larry
is
focused
on
growing
his
crop,
his
cloud,
but
he's
also
aware
that
he
needs
to
be
priced
competitively
if
he
wants
to
make
the
kind
of
profits
he
likes
to
make.
I
think
that
there
is
you.
A
B
Real
concern
that
we
could
be
priced
out
of
that
as
a
storage
system.
If
we're
not
careful,
also
in
the
Oracle
cloud
ZFS
is
not
exposed
right.
It's
the
technology
that
red
that
sits
behind
it,
but
customers
never
see
ZFS.
They
never
use
the
efest
directly.
It's
worth
providing
the
Lunz
for
PA's
is
providing
the
the
access
to
the
files
for
SAS
environment,
but
these
environments-
you
don't
go
in
there
and
address
that
directly.
You
don't
see
that
as
ZFS
storage,
you
just
see
it
as
your
storage.
It
happens
to
be
supported
by
ZFS.
B
It's
not
an
option.
Well,
that's
not
entirely
true,
but
it's
not
an
option.
Most
of
the
instances
deployed
in
past.
We
actually
do
support
a
Solaris
instance
where
you
could
put
a
ZFS.
It's
just
said
you
could,
if
you
threw
some
work,
I
suppose
put
a
ZFS
instance
into
your
Linux
environment.
We
take
some
work
on
your
part,
I'm,
not
sure
how
you
go
about
it
exactly,
but
ZFS
is
also
not
being
deployed
in
the
object
storage
service,
part
of
it
cloud.
B
Well,
again,
we're
strong
right
now
in
that
world,
but
there
is,
you
know
the
fact
that
Oracle
is
refocusing
its
interest
into
the
cloud
and,
as
such,
it's
gonna
be
potentially
I.
Fear
less
interested
in
selling
the
enterprise
directly
in
the
enterprise
itself
is
changing
its
focus.
It's
less
interested
in
acquiring
more
storage
for
us
data
centers
and
more
interested
in
moving
to
some
kind
of
cloud
environment.
B
Where
you
have
you
know,
probably
some
combination
of
private
and
public
cloud
supporting
their
infrastructure
and
so
I
fear
that
that
market
may
be
shrinking
on
us
ultimately
I.
My
fear
is
that
ZFS
is
a
path
to
come
niche
technology
all
right
and
this.
This
concerns
me
because
I
I
want
to
see
ZFS
everywhere,
I
think
it
should
be
everywhere.
It
is
the
best
technology
out
there
for
storing
your
data.
There's
no
reason
why
it
shouldn't
be
out
there
everywhere,
so
does
give
us
that
future.
B
Well,
absolutely
I
mean
this
set
of
companies
represented
here
are
certainly
a
part
of
that
future,
and
an
Oracle
is
certainly
strong
in
that
space,
but
I
think
there
are
things
that
could
happen
and
should
happen
to
make
sure
the
ZFS
continues
to
progress
and
grow
and
become
stronger
first
of
all,
Z
of
s
and
Linux,
and
for
you
all,
and
you
get
too
excited
here-
I'm
the
ghost
of
Christmas
future.
This
is
not
what
will
happen.
This
is
what
could
happen
within
Oracle.
B
An
add-on
bolt
on
piece
of
Linux
all
right
and
in
fact,
from
the
ZFS
perspective,
I'd
like
to
see
us
look
at
Linux
as
a
core
supportive
platform,
so
that
every
feature
in
every
bug
that
we
develop
targeted
to
Linux
as
well
as
Solaris,
but
targeted,
so
that
Linux
is
always
up
to
date.
It
should
be
a
focus
all
right,
because
if
it
becomes
a
focus
and
becomes
part
of
core
Linux
and
I,
and
ideally
the
default
file
system
and
Linux,
then
that
would
be
a
huge
battle.
A
B
B
That's
hard
to
say:
Oracle
was
very
protective
of
his
IP,
but
we've
had
conversations
about
it
and
I
think
there
is
a
possibility
there.
Another
place
where
I
think
ZFS
needs
to
focus.
Some
energy
is
in
the
persistent
memory
space
all
right
until
recently
released
their
3d
crosspoint
memory,
it's
nearly
as
fast
as
DRAM,
and
it's
fast
enough
that
you
don't
have
to
stall
IO
is
too
going
I
owe
to
it
it's
cheaper
than
do
you
Ram.
B
B
This
is
the
model
of
a
that
Intel
envisions
one
access,
their
persistent
memories,
ZFS
fit
so
nicely.
There
is
the
file
system
in
the
middle
right
now,
but
that
just
means
it
can.
It
can
exist
in
that
ecosystem.
Well,
I
want
to
see
us
move
is
to
the
right
the
PMM
aware
file
system,
because
then
we
can
leverage
this
technology
all
right.
We
need
to
be
able
to
access
block
storage,
media,
the
axpy,
m'as
block
storage
at
the
sorts
of
Layton
sees
that
we're
all
capable
there
efficiently
in
ZFS.
B
So
we
need
to
optimize
our
a
path
that
ultra-low
latency.
We
need
to
support
Dax
direct
access
to
2
p.m.
I'm,
using
memory
mapping
resistant
memory,
exposing
it
to
the
application,
which
almost
certainly
means
pre,
allocated
storage,
which
raises
the
question:
how
do
you
do
check
something
in
that
world?
Can
you
check
something
in
that
world
I'll?
B
Leave
that
as
an
exercise
for
all
you
all
to
figure
out
and
I
think
when
you
leverage
PMM
in
the
technology
of
the
earth
itself,
specifically
within
the
Ark
and
the
Zil,
persistent
l2
arc
or
sorry,
precision
l1,
our
caching
I
think,
is
really
important.
How
many
minutes
did
my
allowed
to
go
over
and
ultra-fast?
B
Intend
logging
I
think
of
both
really
interesting
applications
of
persistent
memory
that
can
be
for
direct
utilization
by
ZFS,
as
opposed
to
exposing
it
to
application
there
and
finally
ZFS
in
the
cloud
everything
is
moving
to
cloud
these
days,
so
ZFS
needs
to
be
there
too
I
mean
I've
talked
about
ZFS
being
supporting
the
Oracle
cloud
at
the
moment,
but
I've
also
talked
about
the
fact
that
it's
at
risk
in
that
environment
I
also
talked
about
the
fact
that
it's
not
exposed
in
that
environment.
I.
B
Think
we
need
to
get
to
the
point
where
we
are
the
solution
for
the
cloud.
All
right
object.
Storage
should
be
based
on
ZFS
I
ZFS
is
good
at
supporting
cheap
drives.
I.
Remember
know,
if
you
remember
that
far,
maybe
you
have
you've
seen
early
versions
of
ZFS.
The
last
word
that
Jeff
bond
wick
and
Bill
Moore
used
to
give
their
dog
and
pony
show
around,
but
there
was
a
slide
on
there
and
I
forget
under
which
section
I
think
it
was
around
the
comparison
of
ZFS
to
some
between
various
configurations
of
hardware.
B
But
there
was
a
tag
line
on
the
bottom
said:
ZFS
loves,
cheap
storage
and
we
eventually
removed
that
line
least
I
did
out
of
the
pre
stations
I
gave
because
we
weren't
selling
cheap
storage,
but
the
important
part
of
it
was
that
it
was
there
to
say
if
your
storage
is
not
reliable.
Zfs
can
put
that
add
that
reliability
to
that
storage,
you
don't
need
to
have
enterprise-class
storage
sitting
down
there
for
ZFS
to
be
able
to
utilize
it
and
still
be
as
effective
as
if
it
was
because
we
will
discover
problems.
B
We
will
correct
those
problems
and
so
ZFS
could
be
and
should
be.
You
deployed
in
these
object
environments
where
you
do
want
to
preserve
your
data,
despite
what
the
cloud
is
actually
providing
you
right
now
and
I
think
it
could
be
really
interesting
to
look
at
how
ZFS
could
be
employed
in
an
object,
storage
environment
as.
B
But
again
that
requires
a
fair
amount
of
tweaking
of
ZFS
to
make
it
effective
that
environment
I
had
some
ideas
there,
but
we'll
see.
So
in
summary,
ZFS
is
proven
technology.
It's
extremely
successful
in
the
market
today,
both
the
enterprise
in
a
cloud
space,
but
it
needs
to
keep
moving
forward.
All
right.
It
needs
to
be
the
filesystem
for
Linux
needs
to
be
on
top
of
all
the
new
storage
technology
like
persistent
memory,
and
it
needs
to
become
the
go-to
technology,
go
to
storage
on
the
cloud.
B
B
It
is,
it
is
like
left
reference
links
under
Linux,
that's
where
we
would
hook
it
in,
as
in
fact
it's.
The
rough
link
system
call
is
what
we
use
to
to
trigger
it.
And
yes,
it's
it's
essentially
taking
a
file
and
creating
a
new
version.
The
file
where
we
have
block
pointers
all
point
to
the
original
file.
B
B
B
Zfs
I
think
it's
fair
to
say
that
the
storage
appliance
excels
as
as
an
ad
server.
It's
very.
The
combination
of
ZFS
technology
and
NFS
technology,
which
also
came
out
of
Sun
Solaris,
is
is
very
good,
and
so
our
ability
to
serve
up
Nazz
from
the
appliance
is
as
stellar,
but
we're
also
very
effective
at
serving
up
block
and
so
within
the
cloud
environment.
B
When
I
talked
about
Paz
and
and
says
in
the
SAS
environment,
it's
all
essentially
file
level,
services
and
the
paths
environment,
it's
all
essentially
block
services
and
and
we've
tuned
over
the
years.
The
block,
I
scuzzy,
primarily
their
service,
is
very
effective,
very,
very
fast.
So
so
in
the
markets
as
a
whole.
I
would
say
that
we
have
probably
sell
more
strongly
in
the
end
nas,
but
we
are
also
have
a
pretty
good
foothold
in
San,
the
the
main
hold
back
in
the
san
environment.
B
So
the
main
stumbling
block
is
in
selling,
for
example,
competitively
against.
Something
like
some
EMC,
for
example-
is
that
our
failover
times
are
still
in
the
range
of
20
seconds,
whereas
the
top-tier
sand
environments,
they
want
to
fail
over
time
sub
5
seconds,
and
we
just
aren't
there.
Yet.
You
know
questions
right.
A
B
B
B
Yep,
yes,
yeah
so
and
that's
interesting
question
I
mean
it:
yes,
why?
Why
doesn't
Oracle
participate
more
and
for
the
open
communities
and
leverage
the
the
the
the
brain
trust
there
and
and
the
resources
there?
That
was
the
question
so
effectively
and
the
question
the
answer
is
that
you
know
Oracle
is
a
very
again
I
mentioned
earlier,
very
tight
with
its
IP.
B
It's
very
unwilling
to
play
in
that
space,
as
you
well
know,
and
I
think
that,
despite
the
potential
economic
benefits
of
being
more
strong
in
that
community,
it
culturally
has
a
very
difficult
time
participating
there.
Now.
It's
funny,
though,
because
you
look
at
their
Linux
well,
community
and,
and
that
does
participate
there,
but
that's
mostly
a
take.
It's
not
much
of
a
give
and,
and
so
I
think
I
think
if
you
could
build
up
a
model
where
it
was
mostly
take
Oracle
would
bite.
But
it's
that
giving
part.