►
Description
Join us for Kubernetes Forums Seoul, Sydney, Bengaluru and Delhi - learn more at kubecon.io
Don't miss KubeCon + CloudNativeCon 2020 events in Amsterdam March 30 - April 2, Shanghai July 28-30 and Boston November 17-20! Learn more at kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects
Serverless Kubernetes Boosts AI Business - Jian Huang, Huawei
Kubernetes is becoming more and more popular in IT systems including running the AI workloads. Currently, ML/DL Services of Huawei cloud are running over Kubernetes clusters.In order to make the AI services focus on their business without caring the underlayer infrastructure like physical machines and GPU offering. We provide the serverless Kubernetes services(CCI) in order to meet their requirements. And serverless Kubernetes is very suitable for the users to run the short-time jobs. In this session, we try to introduce the effort we have make on this area. Such as use kata container to protect the container's security so multi-tenant's workload can run on the same physical machine; support AI jobs with multiple devices like Nvidia GPU, InfiniBand; experience of running DL frameworks like tensorflow over Kubernetes in Huawei and so on.
To learn more click here: https://sched.co/FuJv
A
B
A
First,
let
me
introduce
myself:
my
name
is
Quan
Xia
I
am
the
architect
for
soiree
in
charge
of
cloud
computing.
Actually,
three
years
ago,
when
we
first
apples
in
cloud
and
container
technologies,
I
began
to
take
on
the
architect,
works
and
also
technology.
Development
I
mean
first
of
all.
Let's
look
at
what
is
several
is
computing.
Actually,
this
is
a
quote
in
Google
I/o.
What
is
surveillance?
Well?
Actually,
it
means
a
developer.
You
don't
have
to
think
about
those
servers
when
you're
doing
the
developing
in
the
future.
A
You
focus
on
the
code
on
your
business
rather
than
the
server's,
so
only
focus
on
the
code
only
focus
on
your
business.
That
would
be
the
future
ideal.
Futurology
and
cognitive
actually
also
focus
on
services
Computing's,
because
they
are
servicing
open
space.
They
are
doing
a
VM
and
after
cognitive,
so
we
are
focusing
on
a
VP
when
the
customer
is
deploying
their
ap
PE,
they
don't
have
to
care
about
the
service.
That
would
be
the
future
for
computing,
and
so
this
is
several
as
kubernetes.
A
So,
in
the
future,
a
lot
of
cloud
service,
a
lot
of
industries
well
based
on
kubernetes
or
based
on
cloud,
so
they
want
to
choose
what
kind
of
technology
should
be
the
structure
for
several
ISM
services.
So
when
it
comes
to
criminals,
they
have
function
service,
they
have
other
functions
and
now
the
definition
is
no
proud
example.
The
package
circles
use
KPI
to
call
up
cloud
service
or
you
could
have
a
middleware
service,
which
is
also
service,
and,
as
you
can
see
here,
this
is
designed
by
myself.
A
This
is
service
ecosystems
based
on
kubernetes,
and
it
was
centered
kubernetes
centered
on
several
s,
computer.
This
is
a
service
platform
and
there's
a
lot
of
a
very
hot
dog
applications,
for
example,
connected
ok
native
is
for
you
to
have
the
application
on
super
necklace.
It
will
support
you
and
also
it
will
have
or
scalings,
and
it
is
very
elastic
and
also
kubernetes.
K
native
also
provides
you
with
other
services
to
help
you
to
transfer
your
service
and
also
I
have
talked
film
and
others-
machine
learning,
scribes.
A
All
of
those
can
be
deployed
on
the
cloud
platforms.
You
don't
have
to
care
about
how
much
resources
they
occupied
and
also
you
have
traditional
function-
service,
for
example,
cookies
or
fishing.
These
are
the
function
service
focused
on
communities
and
also
yes.
Riff
is
also
focus
on
terminators,
so
it
is
a
application.
It
is
not
just
a
function.
We
also
have
butyl
riffs.
This
are
also
these
accumulated
apartment
function
service.
We
have
batch
service
so,
for
example,
HPC,
high-performance
service
and
also
in
urban
edges
servers.
A
You
have
other
incubating
service
that
can
help
you
with
batch
service
and
in
AI
service.
We
have
cute
flow,
so
cute
flow
is
a
very
hot
topic
in
this
market
on
a
lot
of
people's
such
after
this,
and
also
when
it
comes
to
big
data
service,
a
father
is
also
doing
this
kind
of
big
data
service,
for
example,
spark
and
also
you
have
fit
link.
F
link,
have
a
kubernetes
deployed
solutions
and
also
spark
so
these
are
also
managed
by
Cuba
Nettie's
and
also
spark
is
based
on
chica
missiles.
A
So
when
it
comes
to
sever
this
middleware
servers,
you
have
witness
or
TV.
There
are
database
when
it
comes
to
messages
serving
this
middleware
service.
You
have
cloud
events
and
mass
and
also
called
events,
so
these
are
the
standardization,
and
so
when
it
comes
to
que
native,
it
has
already
realized
the
standardization
of
the
industries
have
his
own
architect.
You
can
also
edit
open
source.
You
can
also
use
those
Canadian
architects
when
it
comes
to
storage
of
the
Sylvanas
middleware
servers
to
have
cells,
to
help
you
to
storage
and
to
combine
storage
with
computing.
A
So
you
can
help
you
with
that.
So
before
the
kubernetes
you
have
middlewares,
you
have,
and
all
of
these
have
already
formed
a
ecosystems,
welcome
to
the
crowd
in
the
future,
so
a
still
possible
to
have
kubernetes
cluster
native.
Now
we
are
buying
nodes
and
deploy
it
on
the
platforms,
but
in
the
future,
is
it
possible
for
us
to
use
Amazon
use,
pave
platforms
using
kubernetes
api
of
all
the
issues
that
will
be
a
part,
probable
solutions?
A
So
we
come
up
with
a
CCI
which
is
a
cloud
container
incident
service.
It
is
a
service
service
based
on
kubernetes.
It
is
a
cloud
container
service.
It
provides
you
with
more
than
95
percent
api
compatible
except
post,
posh
cetera,
and
it
is
the
sub-q
Bonetti
service
supported
by
itvs.
As
you
can
see,
this
is
the
CCI
architects
you
can
see
here.
A
We
has
a
CCI
console,
CCI
API
servers
on
the
second
layers
and
CCI
Federation's,
and
you
can
see
that
you
see
I
can
ensure
can
take
on
all
the
different
clusters
you
and
the
clients,
the
end
users
don't
know
whether
their
API
is
running
on
some
particular
platform.
They
don't
know
that,
but
we
have
also
a
lot
of
physical
hosts
and
on
each
physical
host
values,
container
technologies.
We
use
kata
and
runway
technology
because
kata
containers
they
are
based
on
the
tradition
of
docker
technologies.
A
There
are
a
mutual
of
course
and
that
technologies
can
help
you
with
the
multi-tenant
requirements.
So
this
is
a
tenant
isolation
solutions,
so
CCI
can
help
us
with
a
very
flat
architect,
so
it
helped
us
with
I
cost
technologies
to
manage
those
physical
servers
and
the
artist
physical
servers
we
can
use
to
Banaras
and
we
provide
our
end-users
with
kubernetes
api
and
when
it
comes
to
MDR
and
host
pass
for
example,
host
pass.
We
cannot
expose
host
pass
through
these
container
because
it
is
a
security
related
issues.
So
most
of
the
APR
is
the
same.
A
Is
the
same
with
the
traditional
kpr
api
deployment.
In
the
past
we
were
thinking
about
whether
to
use
vm,
whether
we
trying
to
user
others,
but
after
service
considerations,
we
know
that
tripping
areas
would
be
a
very
basic
infrastructure
for
the
future
cloud
computing.
So
we
are
trying
to
create
this
very
flat
architect
for
our
end-users,
and
this
is
a
foundational
layer,
cube
Elijah.
You
can
see,
there's
a
cubed
letters
in
charge
of
CRI
flex
volumes
at
Siena
and
when
it
comes
to
kubernetes
it
is
as
computing
technologies,
but
we're
trying
to
manage
the
storage.
A
It
is
a
CLI
interface
C
to
CLI
interface
will
connect
with
a
public
account
technologies.
The
C
and
I
will
connect
with
huawei
cloud
V
PC
networks.
So
this
is
the
same
with
the
VM
networks
and
when
it
comes
to
storage,
we
are
using
the
fast
learner.
This
is
a
wave,
a
container
storage
manager
we
manage.
The
storage
is
also
fast
storage
and
object
storage.
A
So
you
can
see
here
when
it
comes
to
CCI
of
valet.
It
is
based
on
kubernetes
technologies
and
combine
all
the
different
cloud
computing,
creating
a
very
flat
ecosystem.
It
is
a
container
clouding
as
cloud
computing
technologies,
so
using
the
CCI
platform.
What
is
the
advantage
of
that?
Let's
look
at
it
for
next
step.
This
is
the
photo
of
our
societal.
So
if
you're
using
our
pave
CCI
portal,
the
end-users
only
have
to
use
the
cake
kubernetes
deployment,
they
could
only
deployed
these
kind
of
applications.
A
You
can
create
a
VM
space
name
space,
so
we
have
our
own
resource
pause.
You
can
have
a
particular
customer
made
a
tailor-made
namespace
to
help
you
with
that
and
when
it
comes
to
different
clouds
and
hole
in
the
knot,
it
is
not
exposed.
So
what
is
the
advantage
of
the
CCI
portal?
Be?
The
first
thing
is
that
you
don't
have
to
care
about
a
building
kubernetes
cluster
Center
nodes.
So
in
the
future
you
could
only
use.
A
A
B
C
C
C
A
C
C
B
B
C
C
C
A
B
C
B
A
A
B
C
B
C
B
A
A
C
C
B
B
A
C
A
C
C
A
A
B
B
C
A
C
B
B
A
C
B
C
B
C
A
B
A
A
A
B
A
B
B
A
A
B
A
A
A
B
A
Otherwise,
if
you
have
training
failures
again,
you
need
to
retrain
again,
I.
Think
there's
a
lot
of
questions
from
the
floor.
So
I
want
to
ask
you,
because
in
our
own
company
we
developed
a
similar
systems,
but
the
challenge
we
have
is
that
when
you
have
a
deep
learning
duck
image,
when
you
operate
deep
learning,
that
image
is
very
big,
for
example,
it
could
reach
20,
G
or
20
gigabytes.