►
From YouTube: Panel: Security Is Not A Uni... Jay Beale, Marlow Weston, Trupti Shiralkar, Aeva Black & Sarah Young
Description
Don’t miss out! Join us at our upcoming events: EnvoyCon Virtual on October 15 and KubeCon + CloudNativeCon North America 2020 Virtual from November 17-20. Learn more at https://kubecon.io. The conferences feature presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects.
Panel: Security Is Not A Unicorn - Jay Beale, InGuardians; Marlow Weston, Intel; Trupti Shiralkar, Illumio; Aeva Black & Sarah Young, Microsoft
Do you have to trade performance for security? Panelists will gather to discuss whether current kubernetes security best practices are at odds with the architectural requirements of heavy computational workloads such as those in machine learning or High-Performance Computing (HPC). The panelists will discuss where the bottlenecks are, what security risks are known, and what industry tools are used to address these issues. The panelists will also draw comparisons to more traditional HPC-style workloads and see what lessons can be drawn from that stalwart legacy. The panel will be moderated by Jay Beale, CTO of InGuardians, who co-leads the Kubernetes project’s third party security audit working group.
https://sched.co/Zerz
A
B
Hi
jay
thanks,
so
my
name
is
ava
black
I've
been
a
systems;
architects,
developer
admin
database,
admin
for
on
and
off
about,
20
years
now
and
then
open
source
for
about
15
of
that
about
nine
or
so
years
ago.
I
started
working
on
openstack
and
ended
up
building
and
running
the
openstack
ironic
project,
which
served
a
lot
of
hpc
a
lot
of
bare
metal
users
that
was
sort
of
the
the
initial
products
that
we
launched
at
hp.
With
that
now
it's
today
it's
using
cern.
C
Hi,
my
name
is
marla
weston.
I
have
worked
at
a
variety
of
different
companies,
including
hp,
national
san,
diego
laboratories,
redstorm
and
now
at
intel
at
the
hpc
tools
group,
and
in
that
time
I've
done
all
sorts
of
things
from
embedded
all
the
way
up
in
security,
and
I
started
looking
at
kubernetes
about
a
year
and
a
half
ago
and
how
it
interacted
with
hpc
and
ended
up
here.
D
Hey
my
name's,
sarah
young,
I'm
a
senior
program
manager
at
microsoft,
azure,
and
so
I
do
security,
engineering
stuff.
I've
been
working
in
infrastructure
engineering
for
my
entire
career.
I
really
love
kubernetes
containers
and
how
we
do
security
better
in
a
cloud
native
environment.
So
yeah
really
excited
to
be
here.
E
E
Currently,
I'm
working
as
a
principal
application
security
engineer
at
a
security
startup
known
as
elimio,
who
builds
micro
segmentation,
products
for
data
centers
and
endpoints.
I
came
to
this
panel
because
I'm
extremely
curious
about
high
performance
computing
based
applications
and
system,
their
role
in
making
the
humanity
better
and
the
unique
security
challenges.
Hpc
systems
face
while
mitigating
these
challenges
and
threats.
The
discussion
around
how
to
strike
a
balance
between
performance
and
security,
fascinates
me
and
that's
exactly
why.
I'm
here.
A
C
I
can
take
some
of
that,
so
hpc
classic
hpc,
you
have
a
front,
float
a
friend
job
scheduler
and
you
go
past
the
job
scheduler
and
that's
where
the
jobs
are
run.
So,
unlike
cloud
native,
where
everybody
has
access
to
their
pods,
you
don't
have
access
to
their
your
jobs.
While
they're
running
like
you
can
look
at
log
files,
but
you
can't
access
the
job
actively
in
cloud
native.
It's
completely
different
format
where
you
can
access
and
when
you
start
looking
at,
say,
networking
or
storage.
C
So
in
cloud
native
you
have
name
spaces
and
c
groups,
you
don't
have
those
with
classic
hpc
and
additionally,
when
you,
when
you're
looking
at,
for
instance,
your
cni,
your
network
layer
and
you
don't
have
access
to
that
either
in
hpc,
so
you
can't
just
go
snip
the
traffic
going
around
and
also
your
job
is
scheduled
for
particular
notes.
You're,
not
going
to
be
there
listening
to
other
people's
jobs.
That
may
be
on
the
same
note.
C
A
C
A
B
That
one
yeah
openstack
went
through
a
phase
as
it
started
to
mature
and
a
lot
of
existing
companies
or
data
science
places
started,
trying
to
use
it
for
large
workloads.
I
had
one
time
sort
of
the
early
days
of
that
I
was
at
one
of
our
conferences
and
a
grad
student.
You
know
some
someone
ran
up,
looked
like
a
developer
and
said
I'm
a
grad
student
at.
I
think
it
was
university
of
madison
and
our
entire.
B
I
don't
know
a
thousand
two
thousand
machines.
It
was
a
big
cluster,
really
impressive
numbers
and
we
talked
a
little
bit
about
the
workload
and
not
a
typical
workload
that
we'd
see
on
a
cloud
back
then
or
today.
Jobs
were
scheduled
differently
and
I
learned
a
fair
bit
then
about
you
know
how
that
job
scheduling
is
is
quite
different
than
in
a
normal
cloud.
A
C
So
you
have
you
have
a
few
different
types
of
jobs
right,
so
some
of
them
are
very
much
what's
called
embarrassingly
parallel
and
when
you
say
embarrassingly
parallel,
you
can
split
it
up
into
a
whole
bunch
of
little
jobs
and
they
kind
of
run
independently.
They
don't
have
to
talk
to
each
other
and
then
there's
ones
where
you
need
communication
in
between
the
different
components
to
do
each
step.
C
So
a
lot
with
machine
learning
as
an
example
for
training.
They
have
to
talk
to
each
other.
It's
a
distributed
computation
when
you're
talking
inference
it's
not
and
with
astrophysics
there's
some
jobs,
that'll,
be
you
know
one
way
and
some
that'll
be
another
with
the
hpc.
The
ones
that
become
challenging
are
not
the
ones
that
are
embarrassingly
parallel.
They're.
A
A
B
A
A
There
are
lots
of
sessions
at
this
conference
talking
just
about
security,
we're
all
you
know
we're
all
having
at
times
really
strong
assumptions
about
how
much
security's
already
baked
in
defaults
and
then
running
into
that
hard
is
h,
does
does
hpc
make
these
security
call.
You
know
the
security
goals
and
using
those
controls
more
difficult.
D
I'm
gonna
have
to
jump
in
here
as
soon
as
I've
been
quiet
so
far,
so,
okay,
so
you're
right,
there
are
tons,
and
you
know
throughout
the
whole,
this
conference
there's
going
to
be
heaps
of
talks
on
how
you
do
kubernetes
security.
Well-
and
I
think
my
comment
here
rather
than
specifically
on
hbc-
is
to
remind
people
that
I
don't
think
we
know
how
to
do
kubernetes
security
super
well,
yet
I
don't
think
anyone
does
you
probably
know
yourself,
jay
from
pen,
testing,
kubernetes
and
things
that
people
don't
always
set
it
up?
D
Up
our
kubernetes
well
and
then
we're
going
to
run
an
hpc
environment
on
top
in.
In
my
opinion,
it
can
only
add
additional
complexity,
but
you
know
I
don't
think
I
just
think.
My
point
is
that
you
shouldn't
we
shouldn't
just
when
we're
talking
about
this
question.
Don't
forget
about
what's
underneath,
because
I
don't
think
we've
really
got
100
there
with
what's
underpinning
this
hpc
and
then
we're
going
to
have
to
put
some
more
on
top
I'll.
E
I
would
like
to
jump
in
here.
First
of
all,
I
would
like
to
say
that
just
focusing
on
kubernetes
security
is
not
sufficient
to
protect
entire
end-to-end
stack
of
hpc,
while
cooper
kubernetes
is
super
important
and
we
need
to
secure
it
because
kubernetes
is
kind
of
the
default
deployment
mechanism
in
native
cloud,
as
well
as
it's
the
part
of
infrastructure.
We
also
need
to
take
a
look
at
the
application
security
side
of
hpc.
E
E
Are
we
actually
protecting
our
hpc
against
those
threat
agents,
and
I
would
like
to
mention
here
recently:
we
just
saw
how
foreign
state
actors
are
targeting
u.s
biotech
firms
who
are
performing
research
on
kobit
19
vaccines.
So
are
you
actually
protecting
your
hbc
systems
against
those
relevant
threats?
If
not,
there
is
some
soul
searching.
We
need
to
do
there
besides
threat
modeling.
I
also
want
to
mention
about
the
code
security.
Are
you
making
sure
the
code
that
goes
as
part
of
hpc?
E
E
Lastly,
I
would
like
to
say
that
get
your
hpc
systems
externally
or
pen
tested
using
expert
third-party
vendors,
such
as
yours,
jay,
that's
also
extremely
important,
because
you
know
you
never
know
what
kind
of
findings
these
creative
pen
testers
will
come
up
that
blue
team
may
not
able
to
identify
in
timely
fashion.
So,
while
for
traditional
non-computing
hpc
system,
setting
up
is
setting
up
a
typical
application,
security
program
is
quite
a
standard
practice.
That's
not
the
case.
I
have
seen
with
classic
hpc,
so
we
might
want
to
pay
attention
there.
A
C
One
thing
I'd
like
to
jump
in
with
is
when
you
start
talking
about
libraries
being
compromised.
You
have
to
be
careful
with
hpc,
because
these
are
real
scientific
numbers
coming
out
and
you
don't
want
to
change
your
numbers.
So
if
someone
upgrades
a
library
and
then
while
they're
at
it,
you
know
fixes
some
sort
of
math
function.
Suddenly
your
numbers
are
different.
E
So
in
this
case,
I
can
think
about
the
security
trade-off
like
instead
of
upgrading
the
libraries
to
the
latest
greatest
security
version,
which
can
which
can
introduce
a
diversion
or
performance
hint.
We
can
think
about
access
control
mechanism
so
that
we
can
limit
the
exploitability
factor
to
that
vulnerable
library.
In
this
way,
we
do
a
security
tradeoff
to
make
sure
hpc
system
is
secure
and
intact.
B
A
Yeah
absolutely-
and
I
know
eva
you've-
done
a
ton
of
work
in
software
supply
chain
marlo.
What
you've
said
scares
me
a
little
bit
like
just
realizing,
realizing
that
when,
when
you're
looking
when
you've
got
that
standard,
when
you're
doing
science,
when
you're
doing
science
and
clinical
work,
you've
got
to
get
the
same
darn
numbers
every
time
and
so
library
upgrades
library
upgrades
can
be
dangerous.
But
often
that's
the
only
way.
A
patch
happens,
wow,
okay,
so
I
I
think
we
can.
A
You
know
we
can
talk
about
different
kinds
of
controls
and
I'm
I'm
eager
to
kind
of
eager
to
to
hear
what
helps
the
most.
But
I
I
keep
thinking
from
what
you've
described
marlow,
especially
when
you
talk
about
the
way
hpc
has
traditionally
been
done
and
how
things
get
a
lot
different
when
you
bring
when
you
bring
this
to
kubernetes
when
you
bring
it
to
cloud
native
that
they're
that
it's
going
to
be
a
different
cultural.
Well,
it's
going
to
be
a
different
culture.
A
C
D
I'll
just
jump
in
there
too,
because
you
know
this
is
as
ava
was
saying.
This
is
something
that
comes
up
time
and
time
and
time
again,
you
know
people
take
when
when
you
introduce
something
that's
new
or
it
has
a
different
structure
and
or
it's
brand
new
technology,
whatever
often
people
will
take
old
security
rules
and
apply
them
to
the
new
technology,
whether
or
not
that
is
valid
or
not
it's
just
what
people
tend
to
do.
They're
like?
D
Oh,
it's,
basically,
an
application
so
we'll
just
use
all
the
same
tools,
controls,
etc
and
it'll
be
fine
and
you
inevitably
end
up
with
gaps.
So
I
don't
even
think
I
know
we're
talking
about
it
in
with
an
hpc
lens,
but
it's
something
that
we
see
time
and
time
again
if
it's
and
that's
why
security
is
so
hard
because
it's
not
just
the
technical
side
of
things
in
fact,
is
probably
easier
in
a
way
than
the
cultural
things.
D
E
I
would,
I
would
say
you
know
we
all
being
the
security
champions.
We
need
to
be
the
change
that
we
would
like
to
see
in
our
teams,
departments
or
organization,
and
I
can't
put
enough
emphasis
on
educating
all
stakeholders,
not
only
the
technical
developers
or
scientists,
but
the
business
folks
as
well
on
the
impact
of
breaches
through
you
know,
security
awareness
training
by
empowering
them
with
the
right
tools.
In
this
case
there
are
tons
of
open
source
security
tools
available
that
we
can
integrate
in
kubernetes
ci
cd
pipeline.
E
In
that
way,
security
doesn't
become
a
hindrance
for
their
innovation
all
the
time.
So
by
doing
that,
we
kind
of
giving
our
people
a
chance
to
change
and
when
they
know
the
change
is
kind
of
coming
from
a
caring
heart.
Where
we
understand
their
pain
points,
their
business
deadlines,
they
get
it.
They
don't
necessarily
feel
challenged.
B
B
A
Like
to
I'd
like
to
talk
about
tools
a
little
bit,
that
sounds,
that
sounds
really
nice.
The
secure
enclave,
just
you
know,
for
for
those
for
those
of
us,
you
know
for
whom
that's
pretty
new
stuff
secure
enclave
basically
gets
us
to
the
point
where
I
can
have
I
can
have.
I
can
have
multiple
workloads,
multiple
high,
you
know
is
this
like,
where
I
could
have
multiple
privileged
workloads
on
a
system
and
they
wouldn't
be
able
to
see
each
other
or
they
would
be
able
to
snoop
on
each
other,
as
it
were.
B
In
theory,
a
virtual
machine
boundary
does
that,
but
the
the
vm
boundary
isn't
protected
from
a
malicious
root,
user
or
malicious
firmware
on
the
host,
whereas
a
secure
enclave
is
right,
even
if
your
neighbor
someone
pops
out
of
their
shell
pops
out
of
their
vm,
gets
rude
on
the
host.
The
enclave
is
still
protected,
because
even
the
route
can't
access
it.
A
A
C
So
classic
hpc,
it's
it's
not
a
problem
because
you
don't
have
to
encrypt
the
traffic
and
transport
right
now.
If
you
look
at
the
cni's,
there
are
two
calico
and
we've
net.
I
believe
that
have
encryption,
enabled
your
performance
tanks,
so
they're
not
really
good
options
for
doing
encryption
at
this
time.
C
So
if
someone
no
one
has
yet
and
I'm
intel
uses
something
like
the
qat
cards
which
are
on
the
pcie
bus
and
uses
that
for
your
encryption
layer
through
one
of
these
cni's,
you
can
get
better
performance,
but
no
one's
done
these
things.
Yet
because
that
hasn't
been
necessarily
a
thing,
they've
been
looking
at
kubernetes
is
still
fairly
new,
so
we
don't
have
as
advanced
as
advanced
tools
as
you
do
elsewhere.
A
E
So
one
of
the
example
I
could
think
about
is
when
you're
processing,
terabytes
or
petabytes
of
sensitive
information
performance
is
extremely
important
for
latency
sensitive
application.
One
of
the
example
I
could
think
about
is,
let's
say
a
simple
credit
card
search
operation
where
you
need
to
decrypt
billions
of
encrypted
credit
card
information,
match
the
right
credit
card
record
and
written
that
information.
In
this
particular
scenario,
just
the
simple
operation
of
decrypting
billions
of
credit
card,
it's
give
a
hit
from
performance
perspective.
Now
we
have
couple
of
ways
to
deal
with
this
situation.
E
Another
thing
I
can
think
about
is
you
can
go
a
more
security,
innovative
side
where,
along
with
storing
the
encrypted
credit
card
information,
you
also
store
a
simple
hmac,
sha-256
tokenized
value
and
perform
your
searching
against
the
tokenized
value,
which
is
faster
and
will
not
give
you
any
performance
penalty.
In
this
way,
you
don't
have
to
worry
about
any
security
trade-off
and
you
still
your
performance
as
well
as
security
requirements.
So
in
hpc
world
it's
really
going
to
vary
from
application
to
application
case,
depending
on
the
architecture
and
business
use
case.
D
Yeah,
and
just
just
to
add
to
that
I
mean
when
we
talk
about
trade-offs,
we
always
talk
about
trade-offs
and
security.
I
mean
all
the
time
so
and
I
know
that
particularly
technical
people
will
roll
their
eyes
a
little
bit.
When
I
start
talking
about
saying
you
have
to
assess
the
risk
and
it's
a
risk-based
approach,
but
it
is
the
reality.
So
you
know
whenever
we
make
technical
decisions
in
security,
there's
often
a
trade-off.
D
Certainly
you
want
someone
who
gets
paid
enough
to
take
responsibility
for
that
decision,
which
probably
isn't
a
technical
person,
but
what
you
could
do
as
an
engineer
or
an
architect.
Someone
helping
with
that
is
help
the
risk.
People
and
the
decision
makers
understand
the
pros
and
cons
of
each
decision.
D
You
know
if
we
do
this,
you
know
it
might
be
more
likely
that
this
attack
would
happen,
but
but
we're
saving,
10
million
or
whatever
it
is,
and
and
companies
make
those
trade-offs
all
the
time
in
security,
and
I
don't
think
it's
a
bad
thing.
It's
just
life,
but
as
long
as
all
the
decision
makers
are
well
informed
and
and
know
the
facts
of
you
know,
what's
the
likelihood
of
this
happening?
D
What's
the
impact,
I
think
that's
the
best
we
can
do,
and
I
do
think
that
I'm
not
saying
anyone
here
go
away
and
learn
risk
management,
please
don't,
but
it
we
do
sometimes
forget
that
as
technical
people,
we
do
have
a
responsibility
to
help
non-technical
stakeholders
and
decision
makers
understand
what
what
they're
being
asked
to
choose
between.
I
do
think
we
forget
that
and
again
hpc
and
everything
else
in
security,
but
I
like
to
talk
about
it
because
it
always
comes
up.
A
That
sure
makes
sense
to
me.
It
sounds
like
there's
a
lot.
It
sounds
like
hpc
puts
us
in
a
much
more
complex
place,
as
somebody
who
you
know
somebody
who's
playing
the
attacker
almost
all
the
time,
I'm
I
I'm
lucky
enough
that
and
I
think
a
lot
of
a
lot
of
security
people
are
like
this.
I'm
lucky
enough
to
just
say
you
should
do
these.
You
should
do
these
10
things.
A
D
I
think
so
too
and,
as
I
said,
ultimately,
some
people
who
ultimately
make
the
final
call
or
or
at
least
take
responsibility
for
what
decision
is
made
there
they're
getting
paid
a
lot
more
than
I
am
so
you
know
and
that's
why.
Ultimately,
you
take
more
responsibility
as
you
go
up
the
chain
but
as
as
you
rightly
point
out
the
more
layers
of
complexity,
because
I've
just
been
doing
very
this
all
that
decisions
in
reality.
It
might
be
this
or
that
or
that
or
that
or
that,
but
there
can
be
lots
of
different
options.
A
Okay,
so
as
we
think
about
as
we
as
we
think
about
this,
as
we
talk
about
this,
I
just
keep
coming
back
to
that.
To
that
question
of
the
you
know
of
how
the
culture
is
different
and
also
how
you
know,
kubernetes
and
cloud
native
has
made
things
different
for
the
hpc
folks
and
for
all
of
us
and
and
how
much
it
changes.
A
So
what
I'd
like
to
ask
is
if
you
could
give
one
piece
of
advice
to
hpc
folks
who
are
who
are
really
looking
at
security
beyond
you
know,
beyond
their
traditional
model
for
the
first
time
or
if
you
could
talk
to
security
people
who
are
having
to
deal
with
hpc
for
the
first
time
like
what
would
you?
What
would
you
want
us
to
learn
and
share?
What
would
you
share.
E
You
know
it
takes
a
village
to
build
a
good
security
program
and
it
takes
time.
Change
is
hard,
so
we
need
to
be
patient
and
if
I
have
to
answer
your
question,
what
is
that?
One
thing
that
I
would
like
to
implement
to
make
sure
hpc
systems
are
secure,
is
shift
your
security
towards
left
as
much
as
possible
to
build
an
agile,
scalable
holistic
security
program.
D
And
just
building
on
that
to
me
like
no
one
does
like
it's
really
don't
beat
yourself
up.
If
you
don't
know
security,
especially
if
you're
just
coming
into
it
for
the
first
time,
because
everything
the
scale
the
rate
of
change
in
cloud
native
and
other
technologies
is
so
immense
that
it's
actually
normal
not
to
know
everything
and
that's
the
same
for
security
as
well.
D
So
I
think
it's
really
important
that
you
know
people
don't
beat
themselves
off
if
they
don't
know
everything,
because
I
do
think
that
we
have
a
little
bit
of
a
culture
problem,
sometimes
in
technical
and
technical
spheres,
where
it's
like,
oh
yeah.
I
know
this.
I've
never
done
it
before,
but
you
know
to
admit
that
I
don't
know
this
is
seen
as
like
throwing
shade
on
my
technical
ability.
So
I
won't
say
I
don't
know
it
and
I'll
just
try
and
muddle
through
security.
D
Isn't
one
of
the
things
where
you
want
to
do
that,
and
I
actually
wouldn't
like
recommend
that
approach
for
any
kind
of
technology
by
the
way.
But
you
know,
if
you
don't
know
something:
that's
okay,
go
and
ask
for
help,
reach
out
to
the
community
reach
out
to
specialists,
because
there's
always
someone
else
who
will
know
it
at
least
better
than
you.
B
I
know
I
work
at
a
large
cloud
provider
and
I
have
worked
at
other
ones
before,
but
I'd
still
say
with
the
state
of
kubernetes
right
now.
If
you
can
it's
more
secure
to
run
it
on
your
own
hardware
and
your
own
data
center
in
a
non-multi-tenant
environment,
because
I
don't,
I
don't
know
that
we
really
have
good,
secure
multi-tenancy.
Yet
in
kubernetes
you
do
need
to
use
a
public
cloud
or
many
public
clouds
still
run.
B
C
And
to
build
on
that
we're
in
a
really
exciting
time,
where
hpc
really
isn't
developed
yet
to
work
on
cloud
native,
so
there's
a
whole
bunch
of
work
that
can
be
done,
there's
a
whole
bunch
of
tools
that
can
be
done.
We
can
imagine
what
we
want
it
to
look
like
and
as
an
example,
we
could
start
using
the
qat
cards
as
a
accelerator
for
tls
so
that
you
have
faster
crypto
traffic,
there's
a
whole
bunch
of
things.
C
A
Yeah,
I
love
that
I
love
I
mean
it
sounds
so
that
sounds
so
much
like
the
spirit
of
open
source.
It's
it's
not
it's
not
quite
ready
yet
and
that's
great
that's
the
challenge
we
get
to
build
and
sarah,
I
feel
like
what
you've
said
about.
You
know:
don't
try
to
fake.
Like
you
know
it
all.
Don't
think
that
you
know
don't
worry
that
you're
expected
to
whether
you're
you
know
whether
you're
super
senior
in
your
field
or
whether
you're
new
to
this
and
that's,
I
feel
like
that's
really
really.
A
That's
really
really
good
advice
that
the
great
thing
about
kubernetes
is
nobody's.
Expecting
you
to
know
everything
and
nail
it
all.
This
is
moving
so
fast.
We're
changing
we're
changing
so
much
and
heck
nobody's
in
the
nobody's
been
in
this
for
six
years,
so
there's
a
yeah.
I
just
I
feel
like
this
is
really
great.