►
From YouTube: 2017-MAY-25 -- Ceph Tech Talk: Ceph on ARM
Description
Members of the Ceph community working on furthering Ceph on ARM gather together to discuss status and plan for the future.
http://ceph.com/ceph-tech-talks/
A
All
right
welcome
to
the
fourth
SEF
tech
talk
of
the
year.
Unfortunately,
we
had
to
cancel
last
month's
because
we're
in
the
middle
of
OpenStack
insanity,
but
this
one
will
be
a
little
bit
different
than
some
of
the
Ceph
tech
talks
of
past
we're
going
to
be
talking
about
some
of
the
Ceph
on
arm
work
and
using
this
as
a
kick
off
going
forward
to
have
a
similar
working
group
that
you
may
see
have
seen
with
the
performance
group.
The
Ceph
performance
that
meets
every
now.
A
Thursday
mornings
used
to
be
Wednesdays,
but
sage
is
going
to
get
a
give
a
bit
of
an
overview
of
the
current
upstream
build
and
infrastructure
status,
and
then
we've
got
a
number
of
people
from
the
community
that
are
doing
stuff
on
arm
work.
It'll,
give
a
5
to
10
minute
overview
about
kind
of
where
they're
at
what
they're
doing
and
where
they're
headed
we've
got
people
from
arm
cavium,
Susa,
DT
and
we'll
talk.
Also
a
little
bit
about
red
hats
downstream
work,
the
commercial
offering
that
will
be
coming,
hopefully
in
the
near
future.
B
I'm
sure
so
I
was
just
going
to
start
with
a
little
bit
of
the
status
upstream.
So
there's
I
guess
a
little
bit
of
history.
We
did
a
bunch
of
work,
maybe
four
years
ago,
with
the
late
Cal,
theta
2
support,
arm
32,
and
so
there
were
a
handful
of
bug
fixes
and
then
mostly
just
some
optimization
work
that
made
sense
for
those
sort
of
wouldn't
be
very
two
bit.
B
B
There
is
sort
of
a
trickle
of
patches
upstream
addressing
various
arm
64
issues,
but
not
very
much
I,
don't
think
that
there
really
any
outstanding
problems
or
bugs
I
think
for
the
most
part,
things
just
work.
Fine.
There
is
some
architecture,
specific
code
incest.
When
the
eraser
coating
libraries
has
the
JRE
sure
one
has
Intel
up
the
mize
and
arm
optimized
code,
both
in
there,
so
it
supports
both
I
had
spike
compatible
whatever.
So
you
can
have
a
mixed
cluster
with
that
plugin.
B
The
is
a
plugin
which
is
based
on
the
Intel
annexed
or
acceleration
library
that
one
is
x86
only
currently.
So
if
you
use
that
one
then
you're
sort
of
locked
into
Intel
platforms,
there's
some
CRC
code
that
is
optimized
for
various
processors.
There
at
one
point,
was
some
arm
neon
code
to
optimize.
He
refused,
but
it
was
slower
than
that.
Just
open
code,
it
stuff
that
we
use.
So
it's
not
enabled
on
x86.
B
B
So
that's
the
code.
I
think
that
the
thing
to
keep
in
mind
is
that
that
should
all
be
taking
what
the
grain
of
salt,
because
we
don't
do,
testing
and
we
do
only
infrequent,
builds
on
arm
64,
so
something
breaks
on
arm
64.
We
won't
necessarily
know
for
a
while
and
we
don't
do
sort
of
the
large-scale
testing
all
over
the
scale.
Testing
regression
testing
kit,
all
dewlap
whatever
is
all
based
on
Intel
hardware.
B
So
we
do
have
some
arm
64
hardware
in
the
lab
Jeff
at
arm
and
the
folks
at
caveum
were
kind
enough
to
donate
a
bunch
of
boxes.
There
were
originally
ten
promised
three
have
arrived
and
have
been
installed
and
our
sort
of
up
and
running,
but
unfortunately
they're
not
quite
enough
to
keep
up
with
the
build
load
and
we
haven't
fiddled
with
Jenkins
to
make
it
sort
of
gracefully
degrade
so
that
it
doesn't
build
everything
it
just
sort
of
build
as
much
as
I
can
so
right
now.
There
aren't
yet.
B
As
far
as
those
builds
go,
they're
sort
of
three
three
categories
of
things
that
we
want
to
use
those
machines
for
one
is
to
build
the
releases.
That's
an
infrequent
activity,
but
we
need
to
like
know
that
it's
going
to
work
when
we
schedule
it.
So
that's
why
we
need
the
second
thing,
which
is
there's
SFC
I
get
repository,
and
every
time
we
push
a
branch
to
that
repository,
it
gets
built
and
packages
get
generated.
That's
what
we
use
to
do
all
of
our
QA
against
and
I
think
very
briefly.
B
We
turned
the
arm
bills
on,
for
that.
Just
see
what
happened
and
it
couldn't
keep
up.
So
we
turned
it
off.
That's
for
somebody
with
some
familiarity
with
Jenkins
could
sort
of
fix
things
so
that
it
at
least
just
builds
master,
and
we
can
like
see
them
go
I
on
on
the
on
the
shaman
page
there
at
the
builds
break
that
be
really
valuable
and
then
the
last
thing
is
make
check.
B
But
that
would
also
be
nice
so
that,
if
somebody
submits
a
pull
request
that
breaks
the
compile
or
just
the
unit
tests
arm,
then
we
would
know
before
we
actually
are
able
to
merge
people
request
and
then
yeah
and
we
wouldn't
watch
anything
unless
it
doesn't.
We
know
that
doesn't
break
break
the
build
and
mostly
that
so
I
think
that
the
two
sort
of
missing
pieces
there
are
just
a
little
bit
more
Hardware.
Probably
the
rest
of
this.
B
That's
that's
sort
of
where
we're
at
as
far
as
just
doing
getting
the
CI
stuff
in
place
so
that
we
can
make
sure
set,
builds
and
past
the
really
basic
tests.
And
then,
when
we
do
a
release,
we
can
publish
arm
packages
on
set
comm
sort
of
the
phase.
Whatever
the
next
phase
of
support
for
arm
builds
would
be
actually
running
our
regression
tests
on
arm
hardware,
and
that
is
a
much
bigger
investment.
B
We
have
like
a
hundred
and
fifty
hundred
ninety
SAS
machines
with
nvme
Intel
based
machines
that
we
use
for
all
of
our
testing,
our
most
of
our
testing
or
we
go
to.
We
also
use
OVH,
there's
also
a
bunch
of
machines,
virtual
machines
on
OVH,
so
you
can
run
tests
sort
of
in
either
of
those
two
environments
and
that's
all
we
use
to
run
the
tooth
ology
test
suite
which
actually
sort
of
get
all
the
code
coverage
and
actually
make
sure
it's
F,
you
know,
works
with
a
capital
W,
so
either.
B
That
means
another
sort
of
hefty
batch
of
hardware
in
the
lab.
You
know,
probably
we're
talking
like
30
machines
minimum
before
it's
sort
of
practical
to
like
run
are
like
big
test
suite
or
a
cloud
provider
that
lets
you
create
armed,
64,
VMs
and
somebody
to
pay
the
bill.
Hopefully
that
cloud
rider
can
just
donated
it.
That's
not
it
somehow
we
can.
You
can
pay
for
that.
That
would
be
good.
That's
fine!
That's
fine!
A
you
year
and
faster
path
because
it
doesn't
involve.
B
B
A
D
Yeah
hi
guys,
so
I'm
steve,
kappa,
I'm
the
enterprise
tech
leader
dom,
and
so
we
were
involved
with
staff
or
I
was
involved
with
Seth
in
about
2015,
when
I
was
working
as
an
assignee
to
Lennar
and
Lennar
is
an
organization
that
essentially
tries
to
boost
collaboration
in
the
arm
ecosystem
with
upstream
the
idea
being
to
stop
everybody
reinventing
the
wheel,
then
sending
the
same
patches
to
people.
So
linor
has
a
series
of
engineer's
that
they
employ
and
also
member
companies.
D
Venero
provide
engineering
resource
as
well,
so
I've
been
working
at
time
as
an
engineer
worked
when
RL
in
2015,
and
we
did
some
work
on
64-bit
arm
enhancements
on
SEF,
and
essentially
these
were
things
that
were
optimizations
because
I
think
a
stage
said
it's
Seth
works
reasonably
well
out
of
the
box
on
arm.
So
I
was
working
on
optimizations
for
memory
system
to
take
into
account
the
fact
that
an
arm
we
can
have
page
sizes.
That's
a
four
yesterday
at
64
K
instead
of
4
K
and
also
works
with
a
colleague
to
him.
D
Since
then,
we
now
have
an
engineering
team
in
Shanghai
and
enterprise
software
engineering
team,
and
that
team
is
looking
at
surf
and
I'm
a
member
of
the
team
and
I'm
representing
them
now,
because
time
zone
differences
are
quite
severe,
so
I
can
give
a
very
quick
summary
of
what's
going
on
it,
the
arm
engineering
teams.
So
recently
the
team
has
been
looking
at
OpenStack
Swift
with
radars
gateway
integration,
and
that
was
termed
young.
You
sent
some
patches
and
pull
requests,
try
and
improve
the
compatibility
it
just
slight.
D
A
P
I
changes
that
go
through
in
the
networking
layer.
Look
just
the
protocol
shall
I
say,
and
this
is
still
being
worked
on
the
other
patches
to
get
merged.
There
is
work
underway
within
arms
attune
set
so
rather
than
just
code
enhancements
on
arm
platforms.
There's
a
tuning
exercises
going
on
to
see
how
one
can
vary
the
kernel
parameters
and
set
parameters
to
improve
I/o
through
pots
and
for
some
tests
being
run
internally.
D
This
has
yielded
some
good
results
in
the
benchmarks
in
certainly,
but
more
time
will
be
spent
working
on
this
before
results
and
configurations
can
be
disseminated.
The
next
steps
for
the
team
are
to
one
of
the
items
is
to
investigate,
sir,
of
course,
the
kubernetes.
It's
one
item
and
also
there's
going
to
be
another
round
a
set
performance,
profiling
unarmed
as
well,
to
see,
if
there's
anything
that
we
can
do
to
beef
up
the
performance
and
arm
systems.
A
F
I
are
the
spawn
coach
and
we
have
Vikram
I,
think
the
current
is
going
to
start
off
the
be
a
bit,
but
just
to
introduce
myself
armor
system
architecture
at
caveum,
mom
I
was
actually
at
Kells
ADA
as
well.
Only
during
the
last
few
months
of
it
and
I
have
had
exposure,
you
know
from
cell
phone
arm
for
the
last
few
years
and
sort
of
the
transitions
and
all
the
stuff
that
has
been
fixed.
That
Steve
was
talking
about
and
lived
through.
F
All
of
that-
and
you
know
it's
a
delight
that
these
days,
all
you
have
to
do
is
just
tap
get
download
and
Seth
works
an
arm
without
any
major
hassles,
and
you
know
we
do
have
a
few
things
that
we
may
want
to
look
into
from
a
performance
point
of
view.
In
a
specific
period,
recording
I
would
say,
but
otherwise
you
know
things
are
generally
looking
good
vikraanta
you
online
now.
Can
you
talk
with
our
update.
E
C
E
E
E
People
are
aware
of
packet
or
net;
they
have
a
bad
little
instance
offering
that
can
support
canonical
and
CentOS
today
and
I.
Think
that
would
debian
version
coming
out,
but
that
thing
does
the
easiest
way
for
developers
to
kind
of
kind
of
sign
up
and
see
how
steps
runs
on
arms.
They
are
interested
and
do
some
optimizations
and
code
implements
our
all
you
could
reach
out
to
our
channel
partners.
You
can
drop
me.
A
note.
Gigabyte
has
developed
black
forms
for
using
our
SOC,
so
feel
free
to
reach
out
to
me.
E
I
can
connect
you
with.
You
know,
Phoenix,
and
we
can
definitely
help
the
organization
get
the
tears
that
is
needed
for
getting
set.
Vi
deployments
are
tingling,
so
that
was
it
on.
The
updates.
I'm
definitely
excited
to
see
this
particular
initiative,
and
we
hope
that
the
community
can
leverage
the
arm,
SOC
style
architecture,
to
improve,
set
performance
in
specific
areas
and
see
the
benefit
of
it.
E
E
F
No
I
think
you
you
covered
quite
a
lot
of
it.
I
mean
you
know.
One
of
the
challenge
areas
still
remains
to
be
basically
benchmarking,
a
lot
of
different
configurations
and
sort
of
comparing
and
contrasting
it
with
others,
and
sometimes
you
know
just
figuring
out.
What's
the
right
set
of
you
know,
settings
and
then
I,
basically
categorize
the
settings
in
two
different
buckets.
F
One
would
be
you
know,
just
a
set
got
conf
settings
which
can
be
a
lot
and
can
be
a
you
know
confusing,
and
the
second
set
is
in
general
system
settings
where
you
set
various
different
parameters
on
your
system
arm.
You
know,
or
sometimes
even
the
hard
drives
themselves
to
basically
figure
out.
You
know
the
best
read
and
write
performance
combination,
so
I
think
some
work
or
pointers
to
you
know
some
community
work
or
from
you
know,
the
Ceph
team.
F
You
know
I
know
that
it's
very
hard
to
come
up
with
a
generic
list
of
settings,
but
maybe
they
could
be
broken
down
into
some
scenario
based
settings
or
guidelines,
which
would
be
helpful.
You
know
just
out
of
the
box.
People
can
start
from
that.
That
would
be
very,
very
useful.
That's
some
work
that
I
would
like
to
see
happen
and.
C
F
B
A
G
G
Being
the
key
of
that,
so
real
quick
that
we've
had
a
product
in
market
since
last
November
around
the
arm
64
platform,
it's
something
that
we've
actually
been
building
since
the
hammer
release
we've
had
we
made
that
available
to
partners
begin
working
with
and
testing
and
through
that
we've
got
our
QA
process
established
in
when
we
do
build,
we
build
across
all
the
architectures.
We
support
some,
both
x86
and
arm
today,
I'm
testing,
both
client
and
server
components
of
the
SEF
stack.
G
We
don't
build,
obviously
as
often
as
upstream,
but
it
is
getting
a
pretty
good
QA
process
working
through
Susa
and
then
again
with
our
customer.
So
you
could
actually
go
and
get
the
enterprise
storage
5
beta.
It's
a
public
beta
for
armed
64
and
work
with
that
today
and
it
works
on
any
of
the
platforms
that
we
support
for
our
Enterprise
Linux.
G
So,
cadmium
AMD
applied
micro,
pretty
much
anyone
that's
out
with
shipping
silicon
today,
one
of
the
keys
around
what
we're
trying
to
do
is
make
sure
that
it's
consumable
well
packaged
and
supportable,
and
we
also
do
build
it.
You
know
those
packages
for
openSUSE
variants
opens
just
a
leaf
and
tumbleweed,
so
those
are
available.
They
go
through
some
QA
as
well
open
to
sleep
shares
much
with
that
SUSE
Linux
Enterprise.
G
So
that
ends
up
being
good
for
the
community
in
that
space,
and
we
have
an
active
ecosystem
of
partners
that
are
testing
and
working
with
the
bits
that
we're
building
what
this
ends
up
yielding
is
finding
bugs
and
again
kicking
that
to
the
upstream
I,
wouldn't
say
that
we're
focused
on
performance,
you
know
type
of
work,
we're
more
focused
on
making
it
stable
and
supportable.
So
we're
looking
for
things
that
break
the
stability,
and
then
we
have
folks
that
are
armed,
64,
familiar
and
supporting
the
64-bit
arm
builds
of
Ceph
with
in
Susa.
G
So
that's
a
short
of
where
we
are.
We
do
have
a
fairly
sizable
amounts
of
equipment
inside
Susa
that
get
involved
in
the
Delta
and
QA
side
and
then
on
the
yhv
side.
I've
actually
got
my
own
lab
where
I
spit
up
and
test
some
of
our
milestone
builds
along
the
way
as
well.
So
that's
the
quick
and
dirty
from
the
suicide.
A
H
H
Okay,
so
what
does
Deutsche
Telekom
doing
on
seven-armed,
so
we
are
currently
working
with
vast
literature
on
their
conversion
microserver
and
doing
a
QC
to
see
if
that
is
something
that
would
work
for
us
and
something
that
we
can
move
in
a
direction
that
would
work
for
Dutch
Telecom
to
make
it
usable.
You
probably
heard
about
these
drives
des
Initiative
simply
did
remove
keyboards
on
standard
3.5
inch
drives
and
put
an
ARM
processor
and
DRM
on
it.
So
and
out
of
the
back
here,
there's
a
interface.
H
It
looks
like
SATA
or
a
sauce,
but
actually
there's
coming
to
x,
1
cubic
Ethernet
out
of
it.
So
what
you
basically
have
in
this
kind
of
sources,
simply
only
a
storage
back,
Ethernet
based
plane
and
no
sort
spec
line
anymore
and
no
shared
CPUs.
So
what
they're
currently
running
is
the
second
generation
of
those
drives
our
laps.
The
third
kind
of
generation
is
the
next
upcoming
one,
so
come
on.
Tv
working
will
arm
sorry
to
put,
and
the
next
version
from
messages
will
be
64
bit.
H
H
So
to
see
you
how
he
drives
react
if
the
cluster
gets
50%
75%
full,
and
what
we
have
currently
is
something
around
1000
test
runs
and
the
data
from
that
and
the
other
part
we
are
currently
working
on
to
try
build
luminous
photos
platform.
It
seems
it
doesn't
build
out
of
the
box.
Currently,
we
currently
run
all
our
tests
visual,
but
we
are
looking
forward
to
fix
deep
goods
to
get
it
running
again.
H
What's
the
motivation
for
us
to
do
that?
One
big
top
cause
of
the
blaze
blast
radius.
So
in
this
platform
there
would
be
no
shared
view:
memory,
no
storage,
backplane
or
shadow
ass,
but
between
it
was
these
and
the
only
shared
part
has
power
and
network.
Obviously,
in
that
would
reduce
the
blast
radius
of
drives
quantify
it's
only
the
drive
failing,
but
yes
would
be
reasonable
scenarios
where
complete
chassis
would
fare,
but
the
number
of
components
that
could
fail
and
lead
to
losing
more
than
one
drivers
fairly
reduced.
H
So
they're,
our
last
single
point
of
failure
from
the
chassis
and
the
other
part
we
see
is
around
this
FS
DS
equals
d
operates
on
dedicated
hardware.
That
makes
it
easier
because
there
are
no
interference
between
one
is
struggling
or
fighting
around
resources
on
the
system,
and
that
gives
us
a
better
predictability
in
matter
of
performance
and
yeah
for
sure
one,
our
topic
for
us
as
costs,
so
the
power
consumption
that
should
be
reduce
a
lot
around
arm
and
this
platform,
and
you
have
to
remember
you-
pay
try
for
it.
H
A
Once
and
there
was
one
more
update
that
I
think
we
were,
we
were
missing
or
weren't
going
to
get
just
because
of
time.
Differences
and
I
just
wanted
to
mention
the
embedded
guys
they
have
been
they've
launched
a
cephalus,
but
the
only
reason
I
bring
them
up
is
that
Aaron
from
embedded
has
promised
to
lead
the
charge
on
taking
performance
data
from
Stefan
arm
from
their
appliance
and
start
sharing
it
to
set
brag.
And
those
of
you
that
don't
know
set
brag
has
now
gone
into
a
more
or
less
an
open
beta.
A
It's
meant
to
be
a
performance
profiling,
repository
I'll
link
it
here
in
the
chat,
it's
just
on
our
main
github
button.
If
you
go
look,
there's
an
example
submission
and
if
you
put
together
some
stuff
with
CBT
the
Ceph
benchmarking
to
land
a
few
other
things
and
submit
that
as
a
pull
request
to
the
github
it'll
show
up
in
the
new
metrics
F
comm
dashboard,
and
we
can
start
looking
at
various
use
cases
and
hardware
profiles
and
having
good
performance
data
that
we
have
all
aggregated
in
a
single
place.
A
So
just
wanted
to
call
out
that.
Hopefully,
we
will
have
some
some
good
stuff
with
Stefan
arm
performance
data
there.
Anyone
else
is
doing
performance,
profiling,
I
highly,
encourage
you
to
try
and
do
it
with
CBT
and
submit
it
to
that
repository.
But
with
that
I
guess
sage,
its
back
to
you
for
the
Red
Hat
downstream.
Unless
you
brought
somebody
else
in
your
pocket,
no.
B
No
and
honestly,
I,
don't
know
a
whole
lot,
but
I
can
give
a
very
brief
summary
so
for
I,
mean
fred
has
been
interested
in
this
for
a
while.
It's
mostly
been
hung
up
on
meeting
the
rel
to
be
released
for
arm
sixty-four,
but
that
finally
happened
with
Pharrell,
seven
point
three.
So
now
it's
sort
of
up
to
the
storage
product
teams
to
deliver
this
in
a
sort
relief
there
they're
sort
of
following
an
upstream
first
approach.
So
they
want
to
make
sure
things
are
looking
good
upstream
and
they
have
some
hardware.
B
A
Well,
I
guess
then
that
brings
us
to
the
next
steps.
So
I
guess
some
stage.
You
have
a
few
thoughts
down.
What
do
we
need
upstream,
for
you
know,
do
we
really
need
to
do
the
32
bit
builds
and
then
what
kind
of
stuff
do
we
need
from
CI
and
then
yeah?
Where
do
we
take
that
next
I
think
Jeff,
if
you
could
weigh
in
also
after
sage
on
the
the
working
group,
stuff
and
kind
of
where
we're
taking
this
a
big
picture?
Yes,.
I
B
Yes,
I
mean
I
think
the
biggest
thing
is
to
really
get
the
the
upstream
seaied
squared
away.
We
want
arm
built
to
be
part
of
every
release
and
at
least
basic
arm
testing
to
be
part
of
the
normal
release
merged
test
cycle.
So,
ideally,
we
would
have
github
triggering
a
Jenkins
build
that
tests
every
pull
request.
B
So
we
know
that
we're
not
breaking
the
build
and
then
we'll
have
regular
builds
on
master
so
that
we
know
that
we
can
generate
usable
packages
so
that,
when
release
time
comes
around,
every
release
is
published,
results
and
published
our
actives
as
well
as
Intel
packages.
So
that's
what
we'd
like
to
get
so
far.
It's
really
just
been
sort
of
two
things
holding
us
up.
B
One
is
black
of
hardware
and
those
those
systems
from
that
cavium
has
generously
donated
or
arm
I,
actually
don't
know
who
donated
them,
but
their
caveat
boxes
they've
been
very
slow
to
sort
of
make
their
way
into
our
lab.
For
various
unrelated
reasons.
We
have
three
of
them
the
seventh
day,
their
seven
are
supposedly
on
their
way
part
of
the
way,
I'm,
not
sure
exactly
how
to
interpret
the
the
most
reason.
B
A
B
A
I
We
had
a
little
disadvantage
right
now,
because
this
is
Jeffrey
from
arm,
so
we're
a
little
disadvantage
because
we
don't
have
Martin
Stadler
on
the
line
and
or
yema
from
from
Leonora
they're,
both
they're
both
on
holiday
this
week.
Currently
you
know
we
had
had
some
original
conversations
with
the
lonardo
cloud.
Being
you
know
OpenStack.
Could
we
use
that
as
an
endpoint
to
try
to
show
how
to
help
and
support
this?
I
So
I
think
that's
probably
one
of
the
follow-on
actions
you
need
to
explore
more
with
you
know
in
this
age
and
kind
of
your
team
up
what
we
can
do
there.
We
have
10
more
platforms
going.
You
know
allocated
to
support
that
within
the
Lenara
labs
as
well.
So
an
email
and
gentler
stage
was
that
I've
been
told
by
the
distributors
or
shipping
so
I.
You
know
what.
H
B
Cool
okay,
so
it
yeah
I
mean
it
if
it
if
they,
if
the
machines
I,
don't
know
that
the
machines
necessarily
have
to
be
OpenStack.
If
somebody
like
sets
up
the
Machine
and
gives
us
an
endpoint,
a
login
is
that
enough
for
us
to
just
add
it
as
a
Jenkins
worker
dan
yeah
doesn't
necessarily
like
dynamically
provisioned
and
key
provisions.
So
right.
C
Week
with
a
couple
builders,
we
have
are
actually
set
up
like
that.
Obviously,
the
physical
machines
that
we're
using
for
trinkets
builds
now
live
on
a
public
network
in
the
community
cage,
and
so
they
talk
to
our
public
instance
of
Jenkins.
You
know
credentials
have
to
be
stored
there,
so
they
need
to
be
at
least
a
little
secure,
but
yeah.
B
C
A
All
right,
well,
let's
so
that
sounds
like
there's
some
good
tactical
stuff
that
we
can
follow
up
on,
but
Jeff
you
want
to
talk
about
the
kind
of
strategic
level
like
where
are
we
headed?
This
is
a
kickoff.
You
know
long
term.
What's
what's
the
vision
here
and
we
have
a
lot
more
people
that
are
interested
in
participating
than
are
actually
on
this
call
so
yeah.
I
I
mean
not
a
whole
lot
to
say:
they're
the
flutter
audience
I
mean,
like
you
said,
there's
there's
a
lot
of
a
lot
of
other
folks
who
are
doing
this
or
some
you
know,
additional
people
who
have
more
clients,
like
things
like
like,
embedded
and
others
who
are
wanna,
participate
in
there.
I
think
I.
Think
the
key
for
us
right
now
is
identifying
what
everyone's
working
on,
very
specifically
within
the
within
the
context
of
Linares
feed
capper,
mentioned
some
of
the
work
that
goes
on
there.
I
I
think
there's
a
level
of
coordination,
I
think
as
we
start,
including,
including
you
know,
folks,
like
Danny
from
from
duty
and
other
places.
We
understand
where
there's
some,
where
there's
some
work.
That
can
be
done.
We've
done
this
on
some
other
packages
before
other
other
upstream
stuff,
where
we
find
own.
You
know,
two
of
our
partners
are
working
on
the
exact
same
thing
and
approaching
at
different
ways,
and
that
may
it
may
not
be
the
best
use
of
the
time
when
we
all
have
a
lot
of
work
that
we
can
get
done
here.
I
So
part
of
the
part
of
the
exercise
is
just
even
an
awareness
amongst
the
folks
working
on
the
arm
bills
and
the
arm
products.
What
actually
exists
and
who's
working
at
what
and
I
think
that's
sort
of
the
first
step
and
really
for
the
for
the
next
quarter.
So
we
just
have
to
kind
of
get
the
organizational
piece
of
us
all
communicating.
What's
going
on
that
way,
you
know
longer
term
I.
Think
as
as
we
see,
other
features
are
not
being
supported
or
new
features
that
that
you
guys
are
pulling
into
their
upstream.
B
You
give
me
a
light
about
that,
but
yeah
I,
don't
I
mean
there's
not
a
hope.
There's
no
I,
don't
think
there
are
any
like
features
per
se
that
architecture-specific
we
do.
There
is
sort
of
a
lot
of
architecture,
specific
optimizations,
though
so
there's
like
a
new
compression
abstraction,
that's
being
used
by
blue
store
and
buy
the
Rios
gateway
and
there's
code.
That's
coming
in
that
uses
some
of
the
Intel
specialized
instructions
to
do
that.
So
those
are
all
opportunities
to
find
to
take
advantage
of
architecture.
B
Yeah
I
think
in
in
in
the
case
of
Cirrus
II,
for
example,
I,
don't
I,
don't
know
if
there's
an
opportunity
to
do
an
improvement
or
not,
but
I'm
also
not
investing
it.
So
this
is
a
case
where
somebody
else
on
this
call
could
do
some
some
tests
and
see.
If
the
you
know
the
neon
or
whatever
it
is
kotas
is
better
looting
them,
but
to
me
platforms
yep,
so
I
just
have
like
a
couple
follow-on
questions
here.
One
is
the
subject
of
flurries
food
bills
came
up
and
Danny
was
talking
about
that.
B
Nobody
uses
AI
386
anymore
and
there's
I.
Don't
know
that
there
are
many
armed
HF
users
out
there.
Maybe
there
are
but
I
think
I
did
get
the
question.
The
logic
question
is
like:
should
there
be
like,
should
we
be
doing
TI
on
32-bit
so
I'm
guessing
that
those
issues
are
not
actually
arm
specific
they're
just
32-bit
issues
yeah,
but
maybe
that
any
any
32-bit
machine
is
fine
or
machines
are
fine,
but
any
opinions
on
the
validators
of
doing
32-bit
builds
or
this
nobody
cares
more
I.
H
Personally
think
it's
not
any
more
needed
right,
so
I
don't
expect
that
so
many
people
would
run
32-bit
code
code.
I
will
be
working
on
that
only
because
it's
we
would
like
to
test.
Do
Minos
on
those
drives
to
compare
to
the
next
generation
yeah.
Some
stitches
on
the
next
generation
are
running
on
Luna,
so
that
was
a
only
reason
currently
for
school.
Then
sometimes
on
the
advert,
it's
a
lot
of
work.
B
H
B
G
B
B
This
is
the
like.
Probably
the
easiest
thing
for
anybody
on
this
call
to
assist
the
upstream
effort.
If
you
want
upstream
builds,
they
will
happen
if
you
help
here.
So
is
there
anybody?
Who
has
you
know,
part
of
I,
don't
know
whatever
it's
not
even
like
a
full-time
job.
Obviously
it's
just
like
somebody
who
can
help
help
babysitting,
but
the
arm
builds
upstream,
because
without
that,
I
can't
promise
that
they're
going
to
happen.
I
guess
what?
If.
B
I
I
mean,
like
I,
said,
I'm
a
little
bit
of
a
disadvantage
because
Martin's
not
on
the
call
right,
so
a
lot
of
the
work.
We
do
a
lot
of
work
we
do
through
you
know
upstream,
is
done
in
conjunction
or
through
Lennar,
and
a
lot
of
those
the
resources
that
we
sort
I'm,
not
sure
the
right
phrase
we
use,
but
that
we
reallocate
to
manaro
get
get
said
assigned
via
what
Martin
and
the
team
in
the
one
our
enterprise
group
are
doing
so.
I
A
So
it
sounds
like
Lynne
ro
has
potentially
two
things
that
we
could
lean
on
them,
for,
or
at
least
they
will
know
someone,
but
you
know
likewise
if
anyone
else
on
the
call
now
or
watching
on
YouTube
later
feels
so
inclined
just
shoot
an
email
off
to
the
Deaf
DeBell
list,
and
you
know
throw
yourself
upon
the
altar
of
Stephan
arm
and
Jenkins,
but
so
I
think
beyond
that.
We
wanted
to
make
this
a
regular
call.
I
Think
we
should
I
think
we
should
start
with
probably
monthly
right
now,
as
we
get
things
ramping
there.
Patrick
right
I
mean
I
think
unless
other
people
think
they're
going
to
have
more
updates
on
a
more
weekly
basis,
I
think
we're
probably
we're
probably
in
the
we
gotta
coordinate
and
go
from
there.
Okay.
A
I
C
A
All
right,
well,
I
will
I
will
get
that
call
scheduled
and
maybe
start
mocking
up
a
arm,
page
onset
comm,
where
we
can
coordinate
this
stuff
and
then
I
guess
from
there
we'll
launch
into
other
things,
but
sounds
like
we
least
have
a
couple
of
things.
You
know
people
are
working
hard.
We
now
know
who
each
other
are
and,
and
obviously
we'll
start
looking
for
for
concrete
volunteers,
starting
with
one
arrow
and
maybe
branching
out
from
there.
C
Got
about
seven
minutes
left
just
to
be
slightly
more
welcoming
than
than
stuff
to
bill.
If
anyone
wants
to
hit
me
up
personally
on
IRC
at
some
point
with
a
quick
question,
I'll
do
what
I
can
to
help
them.
I've
got
a
limited
range
of
experience
with
what's
going
on
here,
but
at
least
I
could
find
you
the
right
answer
for
Red
Hat's
perspective.
If
I
can,
if
I
can
so
I'm,
usually
hanging
out
and
in
the
chef
de
Valera
Channel.
A
All
right
so
I
think
that
pretty
much
covers
it.
Thank
you
all
to
folks
for
giving
updates
it's
nice
to
see
the
amount
of
work
and
interest
going
into
Stephan
arm,
and
hopefully
we
can
make
things
even
even
better
and
easier
from
here
on
out.
So
we
will
see
you
again
next
month
at
the
next
SEF
Tech
Talk,
which
actually
is
an
open
topic.