►
From YouTube: ZFS First Mount by Mark Shellenbaum
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
B
B
Son
had
a
history
of
starting
and
stopping
file
system
projects
over
its
history.
There
are
at
least
four
or
five
different
attempts
that
had
been
done
over
the
years
and
and
ZFS
actually
started
out
as
what
they
called
the
Pacific
program,
which
was
kind
of
a
grandiose
plan
to
come
up
with
a
new
storage.
B
There
was
no
guarantees
we're
going
to
get
to
do
it,
but
that
there
was
this
really
good
possibility,
and
so
it
was.
It
was
enough
to
sort
of
lure
us
away
from
other
places.
We
were
working
at
the
time
and
so
then
finally,
Matt
and
Jeff
actually
got
the
prototype
up
him
up
up
online
and
and
Mark
Himelstein
finally
said:
yes,
we're
going
to
go
forward
with
it.
At
that
point.
B
Jeff
came
looking
for
some
other
people
for
to
work
on
this
and
he
needed
some
file
systems.
People
because
he
didn't
have
any
file
systems.
Experience
I
mean
he
didn't
really
have
storage
experience
at
this
point
in
time,
and
so
he
came
looking
to
mark
and
myself
because
we
actually
were
file
systems,
people
who
know
how
to
write
file
systems
from
scratch,
and
so
we
had
an
initial
kickoff
meeting
in
Broomfield,
Colorado
I
think
it
was
about
May
of
2002,
and
in
that
meeting
there
was
actually
people.
B
It
was
kind
of
kind
of
ironic
because
they
never
actually
been
worked
on
the
project
at
all,
but
they
had
to
be
there
and
during
that
meeting
it
was
all
about
Matt
and
Jeff
kind
of
presenting
the
architecture
of
what
they've
come
up
with
it.
At
that
time,
which
was
you
know,
it
was
not
it's
pretty
minimal
at
that
point
in
time
and
then
the
team
kind
of
got
broken
round
into
into
basically
the
California
team
in
the
Colorado
team.
B
Of
course,
I
mentioned
before,
we
only
was
actually
knew
how
to
write
file
systems,
and
we
need
to
get
this
whole
POSIX
layer.
It's
like
how
do
we
put
a
POSIX
file
system
on
top
of
the
DM
use,
object
store
a
few
months
after
that
pipe
out,
maybe
about
four
months
or
so
later.
Neal
Perrin
joined
the
team
to
actually
start
work
on
the
initials
ill
that
came
along
and
then
between
like
May
and
a
long
time.
B
We
had
to
wait
for
Jeff
to
to
get
busy
and
actually
port
the
spa
to
the
colonel
and
it
seemed
like
it
took
forever
you're
just
waiting
around.
So
in
the
meantime,
what
I
actually
went
off
and
did
was
I
actually
went
ahead
and
created
the
ZFS
kernel,
module
and
I
actually
had
you
know
the
whole
thing
in
there
I
can
unload
the
module
I
could
I
could
do
I
could
mount
a
ZFS
filesystem.
B
Even
though
I
had
no
ZFS
code
that
to
talk
to,
but
I
could
actually
mount
the
file
system
and
do
some
very
simple,
VFS
level
operations.
What
I
couldn't
do
was,
like
you
know,
read
any
kind
of
date
or
anything,
but
I
had
a
lot
of
the
framework
in
place
and
one
and
and
during
that
time
Matt
mark
and
I.
B
We
also
started
talking
about
what
we
thought
the
CPL
should
look
like
and
in
the
original
prototype
that
Matt
had
he
had
something
called
zdes
which
probably
not
have
you've
heard
about
before
I
stood
for
ZFS
directory
services.
So
we
at
that
point
we
rename
zdes
to
be
zapped
and
it
became
generalized
to
support.
You
know
multiple
number
of
integers
of
a
varying
size,
and
that's
that's
the
zap
that
you
all
know
today
and
then
finally
around.
B
B
It
was
simple
to
test
and
simple
to
implement
and
that
actually
became
very,
very
important
for
us
because
at
that
same
time,
Lori
Stevens
was
working
on
a
project
to
make
ufs
in
Solaris
multi
terabyte
and
due
to
the
nature
of
the
you
know,
thin
provisioning.
We
were
able
to
actually
test
MTB
ufs
on
top
of
a
Z
vol,
because
we
didn't
actually
have
enough
storage
to
have,
like
you
know,
a
two
terabyte
file
system
at
that
point
in
the
file
systems
team.
B
And
so
then
you
know
and
then
we're
going
to
move
on
to
where
you
know
where
the
whole
zpl
was
born,
so
we
finally
get
our
spa
in
the
kernel,
and
so
at
this
point
in
time
you
know,
ZFS
is
actually
composed
of
two
different
kernel
modules.
We
had
ZFS
and
we
had
spa
now.
It
lived
that
way
until
about
three
days
before
it
got
integrated
into
Solaris.
It
was
a
mad
dash
to
merge
these
two
modules
together,
and
so
you
know
when
we're
looking
at
this,
we
know
we
want.
B
B
B
You
know
to
basically
look
at
UF
s
because
it
was
kind
of
you
know
the
gold
standard
as
to
what
a
file
system
should
look
like
on
solaris,
but
I
don't
want
to
you
know
reimplement
it
in
you
know
with
ZFS
interfaces,
and
so
I
was
looking
at
we're
looking
around
and
there
was
a
concept
and
you
this
may
surprise
you
that
pc
FS
had
it
wasn't
about
the
way
that
it
put
data
on
the
disk
or
anything
like
that.
It
had
to
do
with
the
way
it
used.
B
B
It's
like
I
want
to
get
rid
of
all
that,
and
so
we
basically
took
the
concept
of
multiple
bop
tables
from
PC
FS,
but
we
did
it
on
kind
of
steroids
where
we
did
everything
like
symlinks
and
all
kinds
of
other
objects,
another
one
of
our
things
we
wanted
to
do,
and
this
one
was
was
very
controversial
to
the
NFS
team.
Was
we
never
ever
wanted
to
support
bop,
RW
lock,
which
was
a
very
large
hammer
to
serialize
right
activity
from
the
cisco
layer?
B
Now
at
sun
we
went
to
a
number
of
projects
where
we
were
actually
trying
to
let
database
consumers
come
through
without
holding
the
right
lock
because
with
because
holding
the
right
lock
was
impeding
their
performance,
and
so
we
did.
We
didn't
want
to
have
that
same
problem,
and
so,
but
the
NFS
team
was
like.
Oh
what
we
depend
on
that,
and
so,
if
you've
ever
looked
in,
like
the
the
vatter
t
of
in
solaris,
you'll
see
that
there's
something
called
a
V
sequence
number.
B
That
was
the
compromise
we
had
to
make
with
the
NFS
team
to
allow
us
to
actually
do
this
and
that
the
end,
the
purpose
there
is
so
that
they
can
tell
or
not
the
attributes
have
changed
from
one
from
one
call
to
another.
Another
idea
that
we
had
was
we
wanted
to
always
have
an
extensible
zeno
door.
I
know'd
format
so
that
we
were
never
like
locked
in
like
like
ufs
or
any
other
file
system.
B
That's
pretty
much
out
there
in
the
world,
and
so
the
very
first
DPL
that
we
had
is
actually
entirely
built
around
zap.
So
there
was
a
zap
attribute
for
M
times.
C
time
sighs,
you
name
it
everything
was.
It
was
an
attribute
now
that
that
made
it
easy
for
us
to
very
rapidly
prototype
this
first
version
of
the
CPL
and
just
a
matter
of
a
few
months.
It
did
not
have
the
performance
characteristics.
We
really
wanted
long
term,
but
that's
how
we
actually
did
the
very
first
version.
B
B
It's
kind
of
unfortunate
that
we
had
to
do
that.
But
that's
what
we
did
and
the
Diploma
the
DLC
is.
It
causes
all
kinds
of
other
side
effects
that
are
really
not
desirable,
like
it
will
postpone
a
lot
of
delete
activity
from
being
performed
on
the
file
system
due
to
negative
caching
on
the
DLC,
the
directory
name,
lookup
cash.
B
B
It
was
actually
a
tremendous
test
suite
because
it
would
do
a
multiple
SS
ccs
get
operations
in
parallel
and
it
would
expose
all
the
kinds
of
races
that
you
have
and
look
up
in
an
active
in
a
file
system,
and
so
the
goal
here
was
we
were
actually
trying
to
work
on
that.
The
old
hello
message
was
going
to
come
out
into
the
Sun
engineering
team,
which
was
basically
what
Jeff
wanted
to
do.
B
Was
he
wanted
to
be
able
to
pull
down
or
bring
over
the
Solaris
date
command
and
compile
it
and
then
run
it
from
from
the
zpl?
And
this
was
like
around
Halloween
of
you
know:
2002,
I,
guess
and
I'm
having
a
problem
with
with
getting
the
to
bring
over
to
work
so
that
you
know
I'm
going
through
a
series
of
panics
and
I.
B
So
I
had
to
kiss
kind
of
slug
it
out
and
you
know
finally
finish
it
and
then
you
know
eventually
we
actually
achieve
that
on
a
day
he
wanted
and
were
able
to
to
send
out
our
Hallel
message
and-
and
you
know
and
all
was
well,
but
you
know
I
would
say
that
you
know
too
much
swear
I.
We
were
doing
this
work.
It
was
perhaps
the
most
fun
I
ever
had.
B
You
know,
as
a
software
engineer
at
the
very
beginning,
like
in
the
first
couple
weeks
like
Mark,
maybe
was
in
my
office,
you
know,
like
all
day
were
sitting
there
like
I'm
trying
to
get
like.
You
know,
simple,
read
to
work
and
then
once
we
finally
got
that
to
work,
then
we
kind
of
like
divided
up
the
Bob's
and
he
went
off
and
did
some
of
them
and
I
did
some
of
them.
B
B
They
won't
ever
like
saved
you
a
copy
of
it
off.
So
once
I
got
the
whole
thing
running,
I
actually
just
saved
it
all
off
as
a
giant
tar
file,
so
I
could
unload
it
whenever
I
wanted
to,
and
we
start
running
through
all
these
tests
and,
of
course,
you
know,
there's
a
number
of
things
that
are
failing.
It's
like
okay,
now,
all
right
this
doesn't
this
doesn't
work
I
mean.
Let
me
go
take
a
peek
and
see
what
you
have
fest
did
here,
so
I
go!
B
So
one
other
thing
I
think
I
should
I
should
mention
is
so
you
know
when
Jeff
put
this
code
back
I
spent
like
about
I.
Don't
know,
I
was
a
half
a
day
or
a
day
trying
to
figure
out
like
all
right.
How
can
I
use
this?
Because
at
this
point
in
time
we
had
no,
there
was
not
multiple
spas.
There
was
not
more
than
one
objects
that
it
was
just.
There
was
like
pretty
much
nothing
that
that
you
have
today
was
there
and
so
I.
B
B
We
never
wanted
to
have
a
file
system
that
you
know
had
to
go,
lay
out
I
nodes
or
anything
like
that,
but
it
was
just
you
know.
Basically
a
case
of
we
need
to
create
a
you
know.
A
root
directory
and
I
need
have
a
way
to
find
out
where
that
directory
is,
and
so
that's
where
things
like
the
master
node
came
up,
which
is
you
know,
it's
kind
of
ZFS
as
equivalent
of
a
super
block.
But
you
know
weh
weh
pared
down
from
what
it
would
have
in
it.
B
A
B
Really
I
mean
so
one
of
our
one
of
them.
He
want
to
know
if
there's
anything
that
I,
that
we
liked
in
ufs
and
there
were
number
things
that
to
dislike
about
ufs
one
of
them
was
it
had
a
series
of
really
really
hard
bugs
related
to
the
virtual
memory
subsystem
that
we
did
not
want
to
reintroduce
with
ZFS
another
one.
That
was
that
that
drives
me
crazy
to
this
day.
Actually
is.
It
is
a
file
system
built
around
macros
and
these
macros
are
all
lowercase
and
you're
sitting
there.
B
You
know
disassembling
code
and
you're
trying
to
figure
out
like
what
this
function
is,
and
you
can't
even
find
you
know
you're
looking
for
like.
Where
is
this
thing?
It's
like?
Oh,
it's
doesn't
even
exist
because
it's
a
macro
and
so
it
what
you
know
it's
like,
there's
there's
no
way
we
wanted
to
introduce
any
of
that
kind
of
nonsense.
Inside
of
ZFS
George.
B
B
You
know
that
things
like
that,
but
you
know,
as
far
as
like
the
way
the
the
CPL
was
put
together
and
probably
not
we
did,
we
did
actually
actually
walk
away
from
the
Zap
model
because
it
didn't
perform
real
well
and
we
went
to
a
fixed
format
version
for
a
while.
But
then
we
came
back
to
the
original
vision,
which
is
the
extendable
attribute
model.
B
So
the
question
is:
when
did
when
did
mirroring
and
raids?
Ii
came
in
so
ironically,
we
actually
had
more
advanced
v-dub
types.
At
one
point
we
had
concat
and
other
types
that
lived
for
a
while
raids.
Ii
came
in
like
we
put
back
ZFS
in
a
you
know,
Halloween
of
2005
raids,
II
came
in,
I
think,
like
october,
one
it
came
in
really
really
late.
Mirroring
had
been
there
for
a
couple
years.
I
think.