►
From YouTube: Kubernetes UG VMware 20230601
Description
June 1, 2023 meeting of the Kubernetes VMware User Group. Announced that this will be the final meeting of this User Group - and perhaps another group with a broader mission will be formed by end of 2023. Brief coverage of recent updates to vSphere and the CSI storage plugin and Cloud Provider as these relate to hosting Kubernetes on vSphere. Discussion of the general topic of "Platform Engineering".
A
Hi
welcome
to
the
June
1st
2023
meeting
of
the
kubernetes
VMware
User
Group.
This
user
group
is
about
people
using
any
form
of
kubernetes
on
top
of
VMware
infrastructure,
typically
meaning
the
vsphere
hypervisor.
So
it
covers
not
just
vmware's
distribution
tanzu,
but
also
people
who
want
to
run
pure
Upstream,
open
source
or
red
hat
Souza
or
whatever
distributions.
As
long
as
they
meet
the
kubernetes
compliance
test.
This
group
will
support
kubernetes
activities
on
the
agenda.
Today.
A
We've
probably
got
plenty
of
leftover
time
for
people
to
engage
in
free
form,
discussion,
but
I
thought
I'd,
just
briefly
go
over
recent
activity
in
the
CSI
and
vsphere
cloud
provider,
of
which
there
isn't
much
but
I'll
cover
anyway.
Let
me
share
my
screen
just
so.
As
I
address
these
topics,
you
can
see
them.
A
It
does
not
appear
to
be
available
in
all
the
forms
like
ISO,
specifically
OEM
isos,
but
IT
addresses
a
pretty
limited
number
of
bug,
fixes
I
looked
through
this
list
and
it
looked
like
it
was
primarily
an
issue
where
people
on
certain
one
gigabit
Nix,
who
upgraded
to
the
last
release.
L
found
issues,
and
this
has
fixes
for
those
for
most
users.
A
Once
again,
no
recent
activity
since
April
the
18th
there
was
a
doc
update
on
May,
the
9th.
Typically,
those
are
minor
and
they
didn't
seem
to
have
any
call
out
of
what
they
had
changed,
but
I
glanced
through.
It
didn't
see
anything
that
caught
my
eye.
The
CSI
plug-in
no
release
updates.
In
the
past
month.
A
A
A
The
other
observation
is
that
much
of
the
typical
discussions
we've
had
during
these
and
by
the
way,
welcome
Bryson
I
just
noticed
you
joined
us,
has
related
to
non-kubernetes
things
that
are
more
in
the
cncf
project,
Camp
things
like
load,
balancers
and
grass.
You
know
the
things
that
are
in
kubernetes
I,
just
addressed
like
the
CSI
storage
plug-in
as
well
as
the
cloud
provider,
you
kind
of
need
those,
but
it
seems
that
users
often
are
faced
with
yeah
kind
of
the
bulk
of
the
issues
can
relate
to
more
things
like.
A
How
do
you
do
a
load,
balancer
on-prem?
Or
how
do
you
cover
networking
service
mesh?
So
there's
a
request
to
lift
and
shift
this
over
to
the
cncf
and
maybe
broaden
the
mission
scope
to
cover
all
aspects
of
on-prem,
so
this
could
be
on
a
hypervisor,
whether
it
be
vmware's
or
somebody,
some
other
one,
and
it
could
also
be
on
bare
metal
and
a
lot
of
these
same
issues.
A
If
you,
if
you
buy
into
this
theory,
that
some
of
the
trickier
concepts
are
things
like
your
load
balancer,
even
how
you
maintain
storage
for
kubernetes
when
you're
on-prem,
they
cross
boundaries
that
go
just
beyond
the
VMware
hypervisor.
So
this
request
actually
makes
sense
to
me,
but
things
are
moving
slow,
so
I
I
was
given
a
deadline
by
the
kubernetes
project
of
this
month.
A
So
I'll
abide
by
that
and
frankly
summer
is
usually
pretty
slow
for
attendance
anyway,
and
they
we've
had
an
effort
going
to
get
people
to
sign
up
and
start
a
group
over
under
the
cncf.
That
would
be
something
along
the
lines
of
an
on-prem
user
group,
and
in
order
to
do
that,
they
don't
want
it
to
be
just
vendors.
In
fact,
I
think
the
cncf
would
prefer
that
it'd
be
almost
no
vendors.
A
However,
the
reality
is
that
it's
typically
vendors
who
have
are
willing
to
pay
somebody's
salary
to
put
in
the
work
to
keep
a
group
like
this
going.
You
know-
and
there
is
I-
can
tell
you
having
operated
this
for
years.
It
is
real
work,
I
mean
you
know
if
you're,
if
you're,
if
you've
got
a
day
job
and
you
have
to
have
a
Cadence
of
monthly
meetings,
you
are
going
to
spend
some
hours
keeping
that
alive.
A
So
I've
talked
to
a
number
of
people
who
are
interested
in
these
topics
and
certainly
when
we
go
to
cute
constant
things,
there's
often
a
lot
of
camaraderie,
even
in
physical
meetings
of
users
and
I,
would
I.
Think
I
personally
would
like
to
see
this
come
up
and
reappear
over
in
the
cncf,
but
it's
going
to
take
some
work,
and
maybe
some
of
you
would
like
to
help
contribute
to
that
and
I.
Don't
think
I
think
the
earliest.
A
A
Well,
historically,
you
you
may
be
able
to
get
passes
for
kubecon
events
and
some
kind
of
support
for
running
them,
but
you
also
I
think
get
a
gateway
to
introductions
with
a
lot
of
people
as
you're
trying
to
bring
speakers
here
and
I
know.
I've
learned
a
lot
running
this
group.
You
know
if
we
were
to
run
an
on-prem
group.
We
could
host
presentations
by
people
from
load.
A
Balancers
Ingress
is
now
sort
of
deprecated
moving
to
Gateway
API,
but
I
think
that
that
act
of
getting
involved
actively
with
a
group
like
that
isn't
just
work
there.
There
can
potentially
be
a
lot
of
learning
opportunity
associated
with
it
too.
So
anyway,
that's
my
announcement.
I'll
open
it
up.
If
anybody
has
any
thoughts
on
this.
A
B
A
B
I
think
the
I
I
think
the
problem
is
kind
of
what
you
said.
Steve
is
that
in
that,
in
reality
many
people
are
they're
running
a
you
know:
a
managed,
kubernetes
distribution
and
if
it's
in
the
data
center
on
vsphere
good
chance
at
stanzu,
and
if
it's
not
tanzu,
it's
openshift
or
Rancher,
how
you
know
how
large
is
the
community
that
is
doing.
B
A
Think,
actually,
there
may
be
large
numbers
of
people
doing
it
on
their
own,
but
when
you
get
up
to
larger
things,
like
organizations
and
Enterprises
I
suspect
it's
way
less
than
half
and
a
lot
of
them
initially
might
aspire
to
doing
it
on
their
own.
But
I
think
that
they
discover
that
the
expertise
involved
there
and
the
support
requirements
are
such
that
maybe
it
is
a
business
benefit
to
kind
of
Outsource.
Some
of
this
to
others,
particularly
when
it
comes
to
maintaining
security,
because
you
know
there.
A
One
of
the
issues
you'll
have
with
regard
to
CVS
and
security
patches
is
that
by
Design
they
aren't
entirely
open
with
the
cpes,
as
they
are
first
reported,
and
the
the
only
way
to
get
hooked
into
that
path
is
really
to
be
a
vendor.
There
might
be
some
other
one,
but
I'm
not
aware
of
it
and
I
I.
Think
for
good
reasons.
Just
so
that
hackers
can
self-identify.
You
know
as
a
user
and
get
early
looks
at
zero
days
in
order
to
exploit
them.
A
A
I
think
coming
to
a
user
meeting
like
this,
isn't
a
great
place
to
go,
resolve
your
security
issues,
other
than
kind
of
generic
principles
of
what
you
need
to
be
worried
about.
You
know,
I,
think
that
the
typical
topics
you
could
discuss
in
a
forum
like
this
would
be
things
kind
of
general
purpose.
Kubernetes
things
for
on-prem.
A
You
know
like
I'll,
just
throw
out
an
example,
but
I
could
probably
come
up
with
a
list
if
I
spent
an
hour
a
dozen
of
them
or
chat
GPT
could
do
it
for
me
in
minutes,
but
take,
for
example,
putting
resource
constraints
on
your
workloads.
You
know
it's
something
that
early
adopters
to
kubernetes,
often
don't
appreciate,
because
you
come
across
training
classes
that
just
say
put
together
this
yaml
and
indeed
the
stuff
runs,
but
you
haven't
disclosed
kind
of
your
expected
CPU
memory.
A
Consumption
of
these
workloads
and
what
happens
on
on-prem
is
unlike
public
Cloud,
where
you've
got
actual
elastic
expansion.
Should
you
be
willing
to
pay
for
it?
You
know
if
you
misuse
these
things,
you
just
get
a
Big
Bill
at
the
end
of
the
month.
You
end
up
on-prem,
where
you
don't
really
have
a
pool
of
idle
servers
sitting
around,
or
at
least
most
people,
probably
don't
if
thing.
A
If
something
goes
south
and
some
workload
you're
running,
maybe
it's
even
a
bug
goes
and
has
a
memory
leak
and
starts,
gobbling
things
down
and
starts
causing
out
of
memory
exceptions,
the
things
killed,
aren't
necessarily
the
perpetrators,
and
you
know
there
are
best
practices
when
you're
running
on-prem
that
maybe
they're
they're
still
best
practices
in
a
public
Cloud,
but
the
repercussions
of
taking
shortcuts
I
think
on-prem.
There
are
hard
more
hardcore
than
they
are
if
you're
in
a
public
Cloud
where
it
would
run
up
your
bill.
A
But
you
wouldn't
go
off
the
air
and
you
know
there.
There
are
other
aspects
that
we've
covered
here
by
Design.
Most
of
the
VMware
distributions
are
not
prescriptive
on
things
like
a
load
balancer.
So
when
you're
in
a
public
Cloud
they're
going
to
have
one,
whether
you
like
it
or
not,
they're
going
to
have
a
solution,
for
you
know
routing
IPS,
even
what
IPS
you
use,
but
when
you're
on
Prem
that
stuff
by
Design
is
left
up
to
you.
A
Just
because
there's
often
interop
with
Legacy
things
that
still
need
those
same
kind
of
features
like
a
load,
balancer
or
a
firewall
and
I
think
there's
a
lot
of
opportunity
for
discussion
of
issues
that
are
unique
to
running
kubernetes
on-prem.
Even
discussions
of
whether
kubernetes
is
the
right
solution
for
being
on-prem.
Frankly,
because
you
know
that
a
common
scenario
is
that
people
have
Legacy
workloads
that
have
been
out
there
for
a
decade,
frankly,
work
fine.
A
How
do
you
make
the
decision
as
to
what
yeah,
which
of
these
makes
sense
to
port
to
kubernetes
or
not,
if
you're
in
a
public
Cloud?
Those
things
probably
were
born
in
a
public
Cloud?
Although
granted
you
have
the
option
of
using
the
vsphere
up
in
AWS
and
other
public
Cloud
providers?
Yeah,
sorry
Bryson,
you
couldn't
stick
if
you
needed
to
discuss
anything
ping
me
on
Slack.
A
B
A
Yeah
as
a
sidebar
I
got
approached
by
some
of
the
people
on
the
vsphere
team
who
are
responsible
out
of
the
storage
business
unit,
and
they
wanted
to
run
by
some.
You
know
they
wanted
to
have
some
one-on-one
conversations
with
users
to
get
wish
list
pain,
points
and
talk
about
future
road
maps.
It
probably
isn't
the
kind
of
thing
best
suited
for
doing
in
this
very
meeting,
because
the
thought
was
that
people
who
have
pain
points
don't
necessarily
want
to
be
communicating
those
in
a
recorded,
YouTube
video.
B
Yeah,
but
also
again,
a
lot
of
the
the
vsphere
CSI
and
storage
things
will
be
related
largely
or
in
part
to
one
of
the
commercial
ways.
The
CSI
ends
up
most
often
being
used,
which
is
again
tanzu,
so
so
it's
yeah
it's
always
this
is
this,
has
all
been
a
problem
from
the
very
beginning
is
how
you
know
that
they're,
so
these
Technologies
are
closely
aligned
with
each
other.
How
do
you
clearly
differentiate
between,
although.
A
That's
rather
interesting
is
how
do
you
do
backups
you
know
in
over
in
CSI
there
is
a
lot
of
work
in
progress
on
doing
snapshots
to
better
support.
Backups
like
you've
had
that
technology
for
a
decade
over
on
VMS,
but
it's
new
to
kubernetes,
hosted
persistent
storage
and
that's
kind
of
a
generic
issue
that
goes
across
commercial
distributions.
Yes,
you
know
the
distribution
is
tied
to
the
actual
storage
being
used,
but
the
the
general
concept
of
even
for
newbies.
What
is
a
snapshot?
A
What
benefit
is
there
and
from
a
I
think
it's
fair
to
say
that
most
of
the
kubernetes
distributions
do
not
include
a
prescriptive
backup
solution.
You
know
you
they
might
include
a
pure
Open,
Source
One,
like
Valero
or
in
vsphere
you.
There
is
a
technique
to
back
up
VMS
the
old
vadp
Appliance,
but
it
really
is
not
probably
the
best
way
to
back
up
these
kubernetes
stateful
things.
It
might
arguably
be
a
really
bad
way
because
it's
backing
up
whole
VMS,
not
storage
volumes
and
they're.
A
Often
these
distributions
can
lead
you
to
a
plain
vanilla
thing
like
Valero
that
maybe
can
get
the
job
done
if
your
needs
are
really
simple,
but
there
are
other
things
out
there
like
casting
that
add
a
fair
amount
of
value
in
a
commercial
release
of
VM.
You
know
of
kubernetes
storage
volume,
backups,
just
backup
used
to
be
a
field.
I
was
strong
on.
A
In
fact,
that's
how
I
joined
VMware
when
they
bought
my
startup
avamar,
but
an
example
of
this
is
that
a
simple
backup
is
that
you
do
a
backup
of
a
volume
yesterday
and
restore
it
today,
and
it
pretty
much
will
work.
But
if
you
have
legal
requirements
to
keep
these
things
around
for
five
years
and
have
access
controls
on
these
things,
you
can
easily
find
that
this
backup
this
storage
had
permissions
associated
with
it.
A
That
granted
admin
privileges
to
people
who
left
your
organization
five
years
ago,
and
you
bring
them
back
and
no
one
is
allowed
to
read
them
and
the
better
Solutions.
Take
it
a
step
further
than
just
the
plain
old:
let's
save
it
and
let's
restore
it
and
are
capable
of
going
in
there
and
taking
inventories
and
even
patching
things
up
and
the
patch
UPS
could
be
security
permissions
with
policy
controls,
because
you
do
not
generally,
it's
usually
a
really
bad
day.
A
In
my
experience
with
backup,
if
the
backup
vendor
restores
it
and
gives
blanket
permission
to
the
whole
world,
that
often
even
violates
legal
rules,
but
even
if
it
does
work,
you've
also
got
issues
of
doing
restores
across
but
release
boundaries
of
your
software
so
picture
that
this
backup
backed
up
a
volume
in
kubernetes
10
and
now
you're
in
kubernetes
version.
20,
that's
years
later,
that
often
you
know
if
it's
restoring
things
associated
with
these.
A
You
know
like
kubernetes
labels
and
things,
so
the
universe
might
have
changed
between
those
two
eras
where
things
just
flat
out
get
broken
and
they
are
not
expecting
kind
of
ancient
crds
or
ancient
labeling
to
be
present
and
you
could,
potentially
it
could
potentially
be
as
ugly
as
destabilizing
your
kubernetes
cluster.
If
you
were
to
bring
these
in
or
bringing
something
in
that
duplicates
a
current
label,
you
know
somebody
labeled
something
XYZ
deleted
it.
It's
been
gone
for
years.
A
Somebody
comes
up
with
a
totally
different
XYZ,
and
then
somebody
restores
the
five-year-old
one
with
name
collisions
and
there's
a
lot
of
value-added
in
those
and
I.
Think
there's
an
opportunity
for
discussing
topics
like
that
that
go
across
kind
of
kubernetes
to
other
project
or
product
Realms
that
are
pretty
generic
and
not
related
to
commercial
distributions.
B
B
B
A
B
B
B
B
Will
it
come
back
in
one
piece
when,
when
your
splits
is
restored,
so
there's
all
kinds
of
things
there
and
the
interesting
thing
with
this
with
this
particular
customer
was
they
they
they've
been
running
kubernetes
for
about
three
years
on
stretch
cluster,
so
their
application
teams,
the
application
owners
they
all
have
been.
You
know,
started
to
learn
this
Cloud
native
World
Under
This
assumption
that
infrastructure
would,
as
they
have
in
the
last
10
15
years,
save
their
bacon.
B
So
now
they
have
to
move
to
a
kubernetes
distribution
where
the
vendor
says
you
know
you
know,
you
cannot
run
this
on
on
this
this
on
VMS,
first
class,
it's
simply
not
supported,
don't
do
it
so
now
they're
forced
to
for
the
first
time
in
three
years
now
to
really
think
about
their
availability
and
I'm,
helping
them
doing
that.
But
it's
it's
a
absolutely
fascinating,
because
you're
forced
to
then
really
create
this.
This
aligned
model
between
all
these
different
technology
layers
and
there's
no
single
vendor.
B
A
Think
that's
a
superset
and
maybe
even
a
bigger
problem,
I
mentioned
when
it
comes
to
availability,
backup,
recovery,
the
problems
of
recording
that
doing
a
backup
and
restore
across
a
Time
boundary
where
maybe
the
kubernetes
release
has
changed,
but
potentially
you've
even
switched
to
a
different
vendor
or
switched
from
one
public
Cloud
to
another,
from
on-prem
to
public
cloud
or
the
other
direction.
And
a
lot
of
these
can
have
tough
repercussions.
I,
don't
even
know
that
anybody
out
of
the
box
promises
to
fix
all
of
that
I'd
be
super
impressed.
A
If
somebody
if
somebody
made
that
claim,
might
be
really
skeptical
and
there's
kind
of
these
things
over
a
chime
Chasm
as
well
as
one
that's
intended
for
disaster
avoidance
or
site
recovery
kind
of
things.
You
know
sort
of
like
what
you
used
to
be
and
still
could
accomplish
in
be
sphere
with
site
recovery
manager,
where
you
could
evacuate
in
a
region.
A
B
Proven
to
work
reliably,
so
there
is
actually
movement
in
this
space.
It's
one
of
the
things
I
looked
out
for
a
kubecon
in
Amsterdam
was
exactly
this
area
because
it's
you
know
it
interests
me
the
same.
We
have
similar
backgrounds
when
it
comes
to
our
our
old
loves.
You
know:
I'm
more
of
a
storage
guy
I'm
more
of
a
backup
guy
right,
but
the
there
is
some
movement
there,
you're
seeing
so
vendors
like
Dell.
B
They
are
moving
in
the
direction
where
they
will
basically
using
obviously
they're
using
technology,
that's
backed
by
their
Storage
Solutions
right,
but
where
they
will
go
as
far
as
to
give
you
a
reference
design
for
a
community
storage
backend
using
that
technology,
but
that's
and
that's
pretty
new.
By
the
way
there
are
in
the
open
source
world.
You
have
things
like
well,
you're
working
safe,
you
have
Longhorn,
but
I
have
yet
to
see
any
open
source
project
or
free.
B
A
Other
thing,
even
historically
I
think
when
it
comes
to
storage,
the
guy,
who
used
to
work
for
VMware
Chad
sackage
used
to
be
my
favorite
commentator
on
storage.
When
he
pointed
he
had
an
adage
that
it
usually
takes
three
to
five
years
for
somebody
to
really
shake
out
a
storage
implementation
to
where
it's
truly
reliable.
You
know.
Do
you
really
trust
some
kind
of
open
source
project
that
just
popped
out
on
GitHub
in
the
last
X
years,
and
you
know,
has
a
churn
of
the
developers
actually
contributing
to
it?
A
B
A
But
you
know
if
you
migrated,
even
in
a
metro
area
from
one
site
to
the
other
and
moved
your
your
persistent
volumes,
but
didn't
have
your
container
image
registry,
your
Helm
registry,
or
whatever,
is
keeping
your
packages
and
there's
a
new
trend
where
I
think
Registries
are
expanding
to
where
they
used
to
call
those
things.
Oci
could
oci
containers
where
C
stood
for
container
and
I
think
they
should
have
called
it
content,
because
that
spec
was
open-ended
enough,
that
they're
putting
all
kinds
of
stuff,
including
web
assemblies,
potentially
there's
a
growing
chat.
A
B
Yeah
and
the
so
this
is
this
is
pink
now
colloquially
called
registry
Ops
right
as
in,
and
the
funny
thing
is
that
I
mean
to
your
point
when
you're
thinking
of
an
availability
design,
it's
always
more
than
just
what's
running
in
kubernetes,
one
of
the
things
I'm
discussing
with
this
customer
is
well
which
of
your
applications
actually
need
to
be
replicated.
How
you
know
most
of
these
are
stateless.
Most
of
them
simply
need
to
be
redeployed,
but
to
redeploy
them.
You
need
your
registry,
you
need
git
and
you
maybe
need
your
pipelining
tool.
A
B
It's
a
damn
will
better
design
it
yeah
yeah
exactly
well.
Each
of
these
tools
has
will
have
their
own
availability
methodology
right,
a
harbor.
You
can
do
with
most
registries.
You
can
do
registry
replication
at
the
registry
level,
so
I
could,
you
know,
have
Two
Harbors
and
one
sinks
to
the
other
right
perfectly
perfectly
fine,
but
but
then
I
have
to
think
about
well.
I
have
Two
Harbors
I
have
two
different
URLs
for
those
registry
images.
So
how
do
I
make
sure
that
those
URLs
are
updated
on
the
other
side?
B
A
I,
don't
think
I'll
be
hosting
another
Zoom
meeting
I
am
going
to
still
be
the
channel,
isn't
going
to
go
away
and
I'm
told
that
I
I'm
a
little
lacks
on
uploading.
The
recorded
videos-
but
you
know
maybe
a
few
months
behind
and
they're
not
going
to
disappear.
The
record
of
recorded
beatings
up
on
YouTube.
B
But
one
of
the
one
of
the
funny
things
is
that
if
you
look
at
all
of
the
VMware
open
source
projects,
you
know
all
the
projects
around
the
VMware
ecosystem
that
are
in
the
open
source
community
that
are
out
there
and
you
know,
I
mean
just
within
the
CNC.
If
you've
got,
you
know,
Andrea
and
harbor,
and
cartographer
and
Contour
right
and
you've
got
the
rabbitmq
operator
and
you've
got
all
these
things
all
right
and
then
and
then
you've
got
obviously
the
kind
of
more
of
the
you
know.
B
B
So
there's
if
we
take
all
of
those
projects
together
and
you
have
a
pretty
substantial
Community,
actually
yeah.
It's
just
just
checking
the
kubernetes
slack
and
it's
funny
because
in
the
provider
vsphere
Channel
at
the
top
is
still
your
old
link
to
the
Sig
VMware
meeting
with
the
old
notes.
It's
still
up
there
as
a
pinned
as
a
pin
topic.
So
you
know
that's
going
to
be
two
generations
old
now.
A
B
B
They
don't
have
a
community
manager
attached
at
least
not
by
VMware,
and
that's
a
shame
because
you
know,
if
you
look
at
that
whole
landscape
there's
actually
quite
a
lot
of
stuff
going
on
so
I've
always
felt.
There
was
a
place
for
a
meeting
like
this.
That's
that's!
It's
all
a
little
bit
fragmented
yeah.
A
I,
the
other
alternative
is,
you
know,
like
I,
say
the
kubernetes
group
effectively
pretty
crudely.
They
want
us
out
now
they've,
given
us
the
option
to
go
to
the
cncf
and
I
think
the
cncf
would
host
us,
but
it
won't
be
VMware
specific.
If
your
point
is,
you
want
a
VMware
specific,
maybe
we'd
almost
have
to
do
that
by
shifting
into
vmug
or
something,
and
that
might
be
another
option
to
consider
where
it
really
effectively
at
that
point
becomes
vendor
run.
B
At
the
same
time
as
I
recognize,
this
is
kubecon
is,
is
there
is
going
to
be
an
increased
attention
on
it
to
the
on-premise
situation,
we're
seeing
a
little
bit
of
cloud
repatriation
happening,
but
more
more
dramatic
than
that
is
the
fact
that
we're
six
years
six,
seven
eight
years
into
this
kubernetes
journey,
all
the
vendors,
the
big
software
vendors
right
have
now
caught
they've
caught
up
with
their
skill
sets
with
their
developers
with
their
application
methodologies
and
some
of
those
big
vendors,
I.
Think
IBM.
B
Think
sap
are
now
ready
to
bring
out
the
next
version
of
their
software
kubernetes
native
and
they
are
starting
to
talk
to
their
customers.
Saying
listen.
The
next
version
of
sap
might
just
be
container
based,
and
you
better
have
a
kubernetes
distribution
sitting
there
on
premises,
ready
to
go
for
our
whole
sap
version
to
land
on
Enterprises
by
and
large
still
are
not
ready
for
this.
So
there
is
this
new
wave
coming
that
will
focus
back
on
the
Enterprise
and
the
on-premises
space.
B
A
Know
I
might
be
biased
because,
frankly,
maybe
VMware
is
in
a
position
to
make
better
margins
in
on-prem
versus
Cloud
I.
Don't
really
know
it's
not
my
job
level,
but
I
actually
am
a
believer
that
this
big
Awakening
with
regard
to
Ai
and
ml
is
going
to
create
a
strong
impetus
to
move
back
to
at
least
Metro
areas.
A
lot
of
this
is
political
data,
privacy
regulations
and
legitimate
concerns
about
having
these
models
and
sharing
your
data.
A
That's
proprietary
in
these
public
clouds,
helping
to
train
a
shared
model
like
chat
DPT
that
is
just
I
I,
think
that
people,
maybe
with
some
early
adopters,
discovering
this
the
hard
way
that
that's
just
lunacy.
If
you're
most
organizations-
and
it
might
in
fact
even
be
illegal,
even
if
it
would
work
and
there's
no
putting
that
Genie
back
into
the
bottle,
the
the
AIML
there
are
just
such
every
week,
I
see
a
new
eye-opening
demo
of
oh,
my
God
I.
A
B
A
You
know
where
you're
going
to
care
about
this
being
a
unique
host,
there's
even
some
really
interesting
aspects
of
this,
or
even
that
you're
not
going
to
do
that.
Ai
ml
without
acceleration
and
Hardware.
You
know
and
hypervisors
like
vsphere
kind
of
own,
the
space
with
hypervising
general
purpose
CPU
hardware-
and
there
are
some
spot
solutions
that
kind
of
tried
to
do,
call
it
call
it
generation
one
or
maybe
it's
generation,
0.5,
considering
how
rapidly
things
are
moving
with
AIML
accelerators.
A
That
I
think
are
going
to
cause
huge
unleashing
of
r
d
funds
to
make
this
change
really
rapidly
kind
of
in
general
purpose.
Cpu
Moore's
Law
really
was
there
and
now
it's
a
legend
where
you
don't
really
see
that
doubling
of
transistors
every
18
months
or
if
you
did,
you
could
double
the
transistors,
but
it
won't
result
in
double
the
workload
but
in
ml.
The
stuff
is
moving
so
fast
that
it
might
be
at
a
nothing
grows
to
the
moon,
but
a
temporary
Pace.
That's
even
going
at
a
ramp
above
Moore's
law.
A
In
terms
of
what's
out
there,
I
did
a
spot
check
for
my
iot
Edge
group
of
just
trying
to
find
what
is
out
there
for
Edge,
accelerators
and
I
quit
after
I
hit
like
16
vendors,
who
have
come
out
with
some
chip
accelerators
for
Edge
on-prem
in
the
last
12
months.
Effectively.
It's
everybody
and
I
found
articles
saying
the
VCS
couldn't
write
checks
fast
enough
for
this
absolutely.
B
A
B
B
A
My
I
think
a
lot
of
this
is
that
you
don't
wait
for
the
cncf
to
do
it.
You
know
they
really
are
open
to
this
to
where
it.
If
you
want
to
make
it
happen,
you
can
generally
go
in
there
and
make
a
proposal,
and
if
you
give
us
a
sync,
compelling
argument,
you
can
make
it
happen.
What
you
do
have
to
do,
though,
is
they
start
looking
at
things
like
diversity
of
vendors,
diversity
on
other
metrics
other
than
vendors
and
Geographic
diversity,
and
they
will
generally
put
something
together.
A
I
mean
none
of
these
organizations
are
going
to
survive
long
run
unless
they
respond
to
what's
changing
out
there
in
the
world
right
I
mean
that
would
be
a
recipe
for
being
irrelevant
in
a
decade
or
so
I.
Think
even
the
leadership
of
the
cncf
knows
this.
There
are
things
coming
together,
like
I.
Do
Host
this
Edge
Group,
which
is
like
Egypt
pretty
much
is
on-prem,
but
also
combining
that
with
low
resource,
and
here
let
me
just
I
posted
a
link
in
a
chat.
A
This
is
early
stage
where
some
people
from
Cisco
and
IBM
wanted
to
put
together
a
white
paper
on
what
they're
calling
patterns
kind
of
put
together
a
scope
for
some
of
the
common
concerns.
You'll
see
and
it's
in
early
comment
stage,
where
I
think
there's
a
lot
that
could
be
legitimately
criticized
as
missing
aspects
or
something.
But
this
is
occurring
within
the
cncf
now
and
it
happens
to
be
within
iot
Edge,
but
I.
A
Think
on-prem
in
a
data
center
context
is
not
the
same
as
what's
going
on
in
Edge,
so
it
might
Merit
its
own,
and
you
could
be
right
that
maybe
the
platform
engineering
is
a
better
place
to
do
it,
but
it
should
be
done
someplace
and
with
this
group
going
down.
If
you
want
to
try
to
help
me
pop
something
up
somewhere,
I'm
willing
to
put
in
some
hours
to
try
to
make
it
happen,
and
maybe
it
isn't.
A
They
kind
of
need
to
say
that
they've
got
some
ongoing
activity
there,
which
to
me,
tells
me
that
you
know
they're
writing
checks
and
it
they
undoubtedly
don't
have
expertise
in-house
so
that,
if
you
could
compose,
you
know
an
outside
vendor
to
Outsource
that,
for
them,
there's
got
to
be
huge
opportunities.
There.
B
A
You
know
in
Amsterdam,
certainly
the
hallway
track
was
AIML,
but
actually
the
the
exhibit
hall.
It
was
not
I
think
a
lot
of
vendors,
just
like
a
lot
of
users
that
it's
almost
like
this
chat.
Gpt
gave
a
wake-up
call
around
March
of
this
year
to
the
whole
world
yep,
and
it's
new
to
everybody,
vendors
who
I
I,
almost
wonder
if
half
the
vendors
who
are
saying
they
have
it
aren't,
just
mostly,
you
know,
blow
hard
liars
and
they
cobbled
something
together
a
week
before
the
show
and.
B
A
The
only
thing
I'll
say
is
counter
to
that
is
that
I
believe
that
the
way
things
are
being
operated
today,
training
is
an
intermittent
kind
of
batch
thing
that
you
don't
need
running
24x7.
So
if
there's
a
way,
you
could
call
it
time
share
in
the
old
days
of
sharing
that
Capital
expense
with
others
that
make
might
make
a
lot
of
sense,
but
the
inference.
On
the
other
hand,
that
is
clearly
on-prem.