►
From YouTube: Keynote: Ceph as Part of the Data Infrastructure for Zoned Storage - Jorge Campello De Souza
Description
Keynote: Ceph as Part of the Data Infrastructure for Zoned Storage - Jorge Campello De Souza, Sr. Director, System and Software Technologies, Western Digital
Zoned Block Devices is a recently introduced new category of storage devices introduced to address the needs of large-scale data infrastructures. In this presentation we will describe Zone Block Devices and how Ceph is well positioned to be a part of the data infrastructure that takes advantage of these new technologies.
About Jorge Campello De Souza
Western Digital
Sr Director, System and Software Technologies
Jorge Campello is a Senior Director of Systems and Software Technologies at Western Digital Research. He holds a PhD degree in Electrical Engineering from Stanford University. He has 20 years of experience in the Data Infrastructure industry. His interests include distributed storage, open source, emerging NVM technology, artificial intelligence, information theory and security.
A
A
Let
you
guys
plan
set
the
fun
part
for
a
little
bit
and
move
on.
So
what
I'm
going
to
talk
today
to
you
guys
about
is
motivated
by
the
data
driven
economy,
that
or
world
as
it
were,
that
we
live
in
today.
So
most
of
our
activities
day
today
are
given
by
data
oriented
new
technologies,
artificial
intelligence,
machine
learning,
etc,
and
this
transformation
is
generating
a
lot
of
pressure
in
the
data
infrastructure
and
I
say
data
infrastructure,
because
it's
not
just
the
storage
of
the
data,
but
it's
also.
A
A
One
of
the
things
I
want
to
note
is
we're
seeing
also
this
transformation
in
storage
technologies
from
the
types
of
interfaces
being
used
and
thus
the
paradigm
that's
used
for
the
storage
devices.
In
particular,
we
see
this
growth
happening
in
SMR
technology
on
the
HDD
side.
Now,
just
out
of
curiosity
show
of
hands
who
knows
what
SMR
is?
A
Okay,
we're
at
the
30
percent
mark,
so
SMR
that
stands
for
single
magnetic
recording
is
part
of
what
we
called
zone
block
devices.
So
what
are
zone
block
devices
well
in
a
normal
sort
device?
It
presents
you
a
linear,
address
space
of
blocks,
typically
four
K
blocks
and
you
can
do.
Is
you
write
to
them
or
you
read
from
them?
A
Typically,
you
can
read
and
write
in
whatever
order
you
choose
now.
There
are
performance
implications
and
considerations
that
you
know
you'll
try
to
do
your
iOS
group
together,
because
locality
usually
improved,
improves
performance,
well
zone
block
devices.
They
have
this
characteristic
that
they
divide
your
linear
address
space
of
blocks
into
zones
within
each
zone.
You
can
only
write
sequentially,
so
you
start
writing
at
the
beginning
of
the
zone.
Sequentially,
if
you
want
to
you,
cannot
overwrite
the
data.
A
It's
very
simple
that
reads
can
happen
in
any
order
you
want,
but
the
writes
have
this
restriction
that
within
a
zone
you
can
only
write
sequentially
now.
Needless
to
say,
that
kind
of
breaks
backwards
compatibility.
So,
if
you're
going
to
take
advantage
of
these
kind
of
devices,
you
are
going
to
have
to
update
your
software
stack.
A
A
You
overlap
the
tracks
by
overlapping
you
pack,
the
data
tightly
together
now,
there's
a
consequence.
You
cannot
go
back
and
overwrite
one
of
the
blocks
in
the
middle.
So
what
we
do
is
we
group
these
tracks
into
separate
zones
and
in
zones
can
be
rewritten
independently,
but
within
a
zone
physically
you
can
only
write
sequentially.
A
Now
when
you
come
back
to
the
solid-state
drives,
we
have
had
this
restriction
of
what
are
called
erase
blocks
within
SSDs
since
their
inception.
However,
typically
within
the
device
itself,
you
have
what's
called
the
flash
transition
layer
that
hides
that
fact
from
the
outside
world.
Internally
it
does
remapping,
it
does
garbage
collection.
It
does
wear
leveling
to
make
sure
that
all
the
different
parts
of
the
the
man
arrays
that
you
have
gets
used
uniformly
to
increase
the
increase,
the
longevity
of
the
device
well
with
zone
device
architecture.
A
You
no
longer
hide
that
inside
the
device-
and
rather
you
expose
these
erase
blocks
to
be
managed
from
the
host
system.
Now,
why
would
you
do
that
for
a
number
of
reasons?
Part
of
it
is
you
have
systems
on
the
host
they're
trying
to
manage?
Basically
where
blocks
are
allocated
within
the
storage
device
and
then
internally
in
the
device
you
have
another
layer,
that's
getting
that
first
level
of
interaction
and
putting
a
neck
a
second
one
inside
it
and
reallocating
data.
So
we
have
this
log
on
log
kind
of
a
redundancy
that
generates
very
high.
A
You
strive
for
a
very
low
write
amplification
and
to
do
that,
you
want
to
expose
this
erase
block
structure
up
to
the
higher
levels
of
the
software
stack,
so
they
can
manage
that
and
get
the
best
efficiency
on
a
global
scale
as
opposed
to
a
local
scale
inside
the
device,
and
when
you
do
that,
you
reduce
a
lot
of
the
pressures
inside
the
device
itself.
It's
neat
for
DRAM,
it's
neat
for
having
redundant
blocks
so
that
it
can
move
data
around.
A
So
all
of
these
technologies
are
coming
to
the
market
so
that
you
have
more
efficient
storage
and
these
are
standard
technologies
within
the
HD
industry.
You
have
what's
called
zone
block
commands
or
is
only
88
commands,
depending
if
it's
cozy
or
SATA
and
in
the
solid
state
area
with
an
nvme,
we
have
the
zone,
namespaces,
that's
being
standardized,
the
standard
isn't
quite
out
yet
it's
working
progress.
We
hope
some
time,
maybe
in
the
fall
or
something
we
all
have.
This
standardized
cool.
A
Alright,
so
we've
talked
about
these
new
technologies
not
backwards
compatible
because
it
breaks
the
API
it's
coming
as
the
future
of
scale-out
storage
to
increase
efficiencies,
and
we've
heard
that
self
is
the
future
of
storage.
So
it
makes
sense
that
we
get
SEF
ready
to
make
use
of
these
technology.
A
The
shift
to
use
blue
stores
are
back
in
and
that
being
a
better
fit
for
us
to
integrate
zone
block
technologies
within
SEF.
Rocs
TB
has
this
LSM
tree
structure,
which
is
natural
for
kind
of
zone,
block
architectures,
and
then
we're
looking
at
incorporating
some
of
the
data
structures
for
managing
some
plot
devices
within
the
blue
store
architecture
and.
A
We
have
a
group
of
folks
that
started
working
on
a
project
on
this,
together
with
a
graduate
student
and
at
CMU
Cary,
Mellon
University,
and
some
folks
at
Western
Digital
we're
starting
to
make
contributions
to
the
safe
project
in
order
to
get
it,
you
know
ready
for
the
future
of
storage
device
technologies
and
for
those
of
you
who
want
to
learn
more
about
what
are
these
things
and
how
is
it
coming
where
and
maybe
see
some
and
touch
some
of
these
devices?
You
can
talk
to
me.