►
From YouTube: Accelerating Data Processing and Consumption for Building Modern Apps... Marilyn Basanta & Ed Byford
Description
Don’t miss out! Join us at our upcoming event: KubeCon + CloudNativeCon Europe 2023 in Amsterdam, The Netherlands from April 17-21. Learn more at https://kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects.
Accelerating Data Processing and Consumption for Building Modern Apps at the Edge - Marilyn Basanta & Ed Byford, VMware
There is an explosion of smart devices at the edge. The upstream bandwidth and real-time decision-making needs are driving the demand for on-premise edge compute. You can easily connect off the shelf open source components to manage AI/ML and real-time applications. By leveraging developer frameworks, message brokers, and in-memory caches, edge native applications can make real-time decisions while operating within the resource constraints at the edge. A common data platform accelerates development of applications at the edge by eliminating repetitive data management tasks. In this talk, we’ll highlight emerging edge application patterns in retail and manufacturing, showcase design considerations for AI/ML and real-time application deployments. Join this talk with the desire to learn more about data at the Edge and leave with knowledge of how this field is evolving.
A
Hi
everyone
thanks
very
much
for
attending
it's
good
to
see
so
many
familiar
faces
here
and
yeah
probably
saw
some
of
you
in
the
EU
session.
My
name
is
Ed
Byford
I'm,
a
product
manager,
I
work
on
messaging
and
streaming
primarily
working
with
some
of
the
rabbit
mq
teams
that
work
at
VMware
and
helping
customers
and
Prospects
understand
how
to
use
event-driven
architectures
how
to
figure
out
all
the
hard
things
about
distributed
systems
messaging
streaming.
B
B
Sort
of
what
I
was
alluding
to
this
morning
was,
you
know,
and
the
funny
part
about
this
morning
is
the
reason
me
giving
that
keynote
was
so
much
harder
for
me
is
because
I
really
didn't
want
to
give
away
all
of
the
content
from
this
presentation-
and
this
is
the
part
of
actually
the
most
excited
to
talk
about,
and
so
at
the
edge
I
know.
I
talked
a
little
bit
earlier
about.
B
You
know
the
manufacturing
vertical,
but
I
was
talking
a
lot
about
the
instant
experiences
and
we
all
have
you
know,
since
we
never
let
go
of
our
phones.
Everyone
expects
those
instant
experiences
now
so
from
the
customers.
They
want
to
be
delighted
now
and
then
from
a
business
perspective.
I
want
to
show
efficiency.
I
want
to
show
cost
savings.
I
want
to
show
the
justification
for
putting
these
new
experiences
in
the
store
right
because
of
the
consumer.
B
So
it's
all
the
different
use
cases
what's
interesting
and
what's
challenging
I,
think
from
us
or
from
and
I
assume
it
from
other
Enterprises
of
selling
into
Enterprises
to
the
edge
is
getting
to
the
right
target
audience.
So
here
we
are
at
kubecon
and
our
audience
are
developers
and
you
know
developers
are
excited
about
the
tech
they're
excited
about
how
I'm
going
to
enable
them
to
get
these
new
type
of
workflows
put
in
the
edge.
But
it's
interesting
when
you
go
talk
to
the
Enterprise,
at
least
for
VMware.
B
B
Okay.
So
this
is
a
I
realize
now
that
this
is
probably
a
similar
slide
and
I
should
have
just
kept
one.
But
the
point
I
wanted
to
make
here
is
that
you
know
Edge
has
some
of
the
you
know
the
the
biggest
spending
that's
happening
at
the
edge
and
in
particular,
some
research
produced
out
of
the
beginning
of
the
year
from
IDC?
Is
you
know
on
what
verticals
and
those
verticals
really
are
retail
in
manufacturing?
B
So
it's
no
surprise
that
so
VMware
we're
focusing
on
retail
and
Manufacturing,
which
is
you
know,
fits
to
the
adjacent
Market,
such
as
automotive
power
utility
fulfillment
centers.
So
these
are
all
the
areas
that
we
can
really
help:
revolutionize
with
our
Tech
at
the
edge
on
on
what's
necessary,
so
going
back
to
kind
of
retail,
because
that'll
be
kind
of
what
it'll
lead
us
into.
B
It
is
well
I've
already
talked
a
little
bit
about
the
transformation
this
morning,
but
just
to
kind
of
recap
right,
so
we
know
edge
of
the
future
and
so
really
going
back
to
the
instant
Delight.
Is
we're
really
augmenting
that
Human
Experience?
So
that's
really
what
we're
kind
of
focusing
on
so
then
again,
how
do
you
get
to
those
instance
experiences
when
I
said
this
morning
that
in
many
cases
all
these
data
is
everywhere
and
different
systems
so
to
highlight
on
the
retail
example?
B
If
you
look
at
this
chart
here,
this
really
kind
of
shows,
like
you
know,
we
use
the
term
store
of
the
future,
but
I
like
every
name
that
I
call
it
the
digitization
and
modernization
of
retail
stores.
So
a
lot
of
this
stuff
in
the
stores
gets
done.
The
manager,
pencil
and
paper
in
hand
or
they've
got
some
Legacy
VMS
or
you're.
Stuck
with
your
isvs
I
wrote
your
point
of
sale.
You
know
it's.
It's
very
costly,
then,
to
write
your
own
own
point
of
sale.
They
know
they
want
to
put
in
computer
vision.
B
You
know
how
how
fraud
and
loss
is
happening,
and
then
you
can
take
computer
vision
and
apply
it
to
inventory
control,
and
then
the
other
thing
about
retail
is
that
they
also
have
fulfillment,
because
if
you
think
of
especially
during
covid
when
stores
were
closed,
they
ended
up
becoming
kind
of
their
own
mini
fulfillment
centers,
and
although
of
course
that
was
happening
manually.
How
do
you
a
lot
of
that
technology
that
had
to
come
very
quickly
over
the
last
two
years
now
the
retailers
are
looking
at.
B
How
am
I
going
to
actually
put
this
together
into
a
more
automated
digitized
way,
rather
than
having
to
quickly
do
self?
You
know
curbside,
delivery
and
self-checkout
and
all
those
bits
so
what's
interesting
in
these
use
cases
just
to
touch
a
little
bit
more.
Is
you
know
it
covers
a
variety
of
the
topics
of
what
I
was
saying.
Some
of
it
is
too
for
the
customers
engagement.
Some
of
it
is
for
the
business
operations
and
then,
of
course,
not
to
leave
out
some
of
it
is
for
safety
and
regulation
right.
B
So
it's
ensuring
we
satisfy
with
our
Solutions
both
sides
of
the
of
the
businesses.
Okay.
So
let's
talk
a
little
bit
about
the
data
problem,
so
now
you
saw
some
of
the
use
cases
that
are
being
deployed
in
retail
and,
like
I,
said
some
of
the
stuff.
These
are
things
that
happen
day
to
day,
seemingly
mundane
things,
but
let's
kind
of
walk
through
our
workflow
of
what
that
looks
like.
So
there
are
a
lot
of
use
cases,
so
you
know,
let's
say
you're
in
the
store
and
someone
Falls
over
seems
you
know.
B
The
obvious
thing
is
to
come
to
mind
of
what.
What
do
you
need
to
do?
It
also
depends,
of
course,
the
context
is
important,
depends
on
what
that
fall
was
who's
in
the
store
all
these
different
bits.
So
let's
go
kind
of
the
first
thing,
so
someone
fell
in
the
store.
The
first
thing,
USS
says:
how
serious
is
this
problem?
Do
I
need
to
call
you
know
paramedics?
What
what
do
I
need
to
do
so
then
you
have
so
now.
You
need
to
figure
out
who's
in
the
store
right
now.
Who
should
make
that
call?
B
So
then,
now
it
goes
to
who's
got
the
skills
which
of
your
employees.
Should
you
send
out
who's?
Had
the
first
aid
training,
who's
who's
got
the
right
qualifications
to
go
help
someone
urgently.
So
that's
also
typically
yet
in
another
database.
Your
training
data
or
it's
in
the
cloud
and
then
third
is:
if
is
this
a
security
incident?
B
B
B
This
one
incident:
how
does
it
kind
of
alt
together
to
kind
of
communicate
across
these
these
items?
So
you
can
kind
of
see
that
the
data
problem
across
the
different
systems
kind
of
just
explodes,
and
this
is
just
one
scenario
of
one
thing
that
can
happen
in
a
store
today-
that
typically,
these
type
of
incidents
will
happen
multiple
times
in
a
day.
So
then,
now
you
have
the
the
volume
of
the
different
type
of
Eventing
happening.
B
Okay,
so
now
again,
I,
say
I
cover
that
for
one
use
case,
what
happens
with
you've
got
this
mirror
mirrored
mirrored
numerous
use.
Cases
happening
more
at
one
time,
so
you've
got
a
whole
mess
of
data
systems
that
don't
talk
to
each
other
and
it
kind
of
just
becomes
a
huge
spaghetti
monster
of
a
challenge
to
solve.
A
A
little
Shuffle
all
right,
yeah,
thanks
Marilyn,
so
we've
talked
about
one
specific
case
and
one
problem
you
can
imagine
that's
multiplied
as
Marilyn
said
over
numerous
use
cases,
but
really
what
we
wanted
to
return
to
now
is
thinking
about
what
is
the
goal
and
we
think
of
the
goal
as
being
the
ability
to
unlock
the
value
that
the
developer
can
provide
to
a
customer
and
really
you
know,
we
know
that
business
value
doesn't
solely
rely
on
infrastructure.
It
also
relies
on
really
being
focused
on
what
is
that
customer
need?
A
What's
that
internal
tool
need
and
essentially
anything
that
gets
in
the
way
of
a
developer,
just
delivering
value
accessing
data
provisioning
infrastructure,
even
just
thinking
about
what
are
the
Myriad
of
different
devices
that
exist
at
the
edge
location?
All
of
that
is
noise
that
gets
in
the
way
of
just
producing
some
kind
of
application
or
some
kind
of
tool.
A
That
brings
you
business
value,
so
we
were
looking
at
some
of
the
you
know:
customers
we
speak
to
over
the
course
of
our
jobs,
and
we
really
recognized
five
main
data
related
challenges
that
get
in
the
way
of
developers
delivering
that
value.
One
of
those
challenges
is
limited
bandwidth.
Now
this
was
you
know,
is
a
problem
across
multiple
Edge
locations
and
really
what
happens
is
the
cost
per
location
can
really
add
up?
You've
got
hundreds
of
sites.
A
So,
as
Mario
mentioned,
there's
a
lot
of
different
types
of
edge
applications
and
people
think
we're
all
already
doing.
Edge,
that's
true,
but
those
things
end
up
in
silos
because
there's
a
number
of
different
ways
that
you
can
implement
it.
Oftentimes
Enterprises,
aren't
thinking
about
a
joined
up
way
to
do
this
across
multiple
different
use
cases
they
would
have
delivered.
A
You
know
a
computer
vision,
solution
or
a
point
of
sale
solution,
but
not
thought
about
ways
to
make
efficiencies
across
those
and
because
of
that
there's
a
range
of
different
vendors
which
might
necessitate
different
kinds
of
workflows
to
access
that
data.
So
again,
your
developers
are
having
to
learn
different
ways
to
access
data
and
connect
things
together.
A
A
A
Well,
the
main
patterns
we're
seeing
people
try
and
Implement-
are
these
three
and
of
course,
they're
a
myriad
of
other
ones,
but
these
are
the
main
patterns
which
seem
to
help
with
these
problems,
so
you've
got
Ai
and
ML
inferencing
and,
as
we've
already
discussed
several
times,
that
aim
inferencing
should
be
happening
at
the
edge
for
expediency
and,
as
we
know,
the
majority
of
a
and
ml
inferencing
is,
or
training
at
least
is
gathering
data
in
order
to
to
make
those
models
work.
So
that's
why
data
is
very
important
here.
A
They
don't
need
somewhere
to
store
that
data,
and
what
we
see
is
people
implementing
and
wanting
to
implement
a
Consolidated
event
stream.
You
might
have
streams
of
data
in
different
locations,
but
sometimes
it's
the
context
between
and
as
we'll
see
in
the
demo
is.
The
contact
between
what's
happened
in
one
part
of
the
store,
with
what's
happening
in
another
part
of
the
store
that
you
really
need
to
tie
those
things
together.
A
So
you
need
something
that
consolidates
that,
and
then
you
also
need
to
be
able
to
do
analytics
on
that
Stream
So,
analyzing
the
event,
but
also
analyzing,
the
sequence
of
events
to
iteratively,
improve
on
those
processes,
and
if
we
drill
down
a
bit
further,
what
we
can
do
is
take
a
set
of
Open
Source
components,
and
really
these
are
the
heroes
of
our
story.
We've
got,
you
know
the
spring
application
framework.
We've
got
rabbit,
mq
that
can
do
messaging
message
queuing
and
event
streaming
and
you've
also
got
key
value
store
in
Apache
geode.
A
So
what
I
want
to
do
now
is
expand
or
move
away
from,
maybe
retail
but
think
about
logistics,
and
in
this
case
it's
a
a
package
sourcing
application
at
a
logistics
company,
and
we
want
to
you
know
this
to
be
customizable
or
composable,
but
again
we're
going
to
use
some
of
those
Heroes.
We
just
mentioned
to
think
about
how
to
implement
different
parts
of
this
of
this
application.
A
So
what's
happening
here.
Well,
imagine
a
delivery
company,
logistics
company,
that's
got
packages
coming
down
a
conveyor
belt
and
they
need
to
be
determined
which
shoot
which
delivery
shoot
to
be
pushed
into
by
a
particular
diverter
arm
and
that
that
decision
needs
to
be
made
quickly.
You
can't
be
communicating
back
up
to
the
clouds,
you
need
to
make
it
locally
and
if
it
cannot
be
made
within
a
certain
time
frame,
the
package
is
going
to
go
somewhere
else
to
the
wrong
place.
It
may
need
to
slow
down
the
belt.
A
You
need
to
Halt
sort
of
processing
and
again
that
costs
money.
So
this
is
laid
out
a
little
bit
like
a
service
blueprint.
This
diagram,
if
you're
familiar
with
that,
so
we've
got
the
kind
of
front
stage
and
at
the
front
stage
you've
got
the
developer
who's
sitting.
There
writing
application
source
code.
A
A
Then
you've
got
over
here.
This
Edge
compute
stack,
which
is
essentially
taking
the
application
building.
It
versioning
it
deploying
it
making
sure
that
it's
running
making
sure
that
dependencies
are
installed
in
those
Edge
locations
and
then
down
here.
You've
actually
got
the
physical
Edge
location,
so
you've
got
some
compute,
you've
got
running
apps,
and
so
essentially,
that
process
means
that
you've
now
got
the
source
code
up
and
running
again
via
a
set
of
common
apis.
A
Now
you've
got
your
applicated
your
package
rather
and
that's
being
scanned
and
inferenced
Via
local
AI
and
what
happens
is
the
package
is
traveling
down
the
conveyor
belt
and
as
soon
as
the
package
is
being
scanned,
then
there's
an
event
which
is
being
generated
and
that's
a
package
recognized
event.
So
we've
got
to
have
somewhere
to
put
that
and
as
previously
mentioned,
we
need
to
put
that
somewhere.
A
That
has
a
end-to-end
view
of
events,
that's
happening
in
the
business,
and
so
we
put
that
into
a
Consolidated
event
stream,
but
that
process
doesn't
stop
there
because
you've
also
got
the
whole
journey
of
the
package.
So
you've
got
the
diverter
arm,
which
is
moving
and
routing
packages
to
different
locations.
That's
again
a
decision
and
also
a
diverter
arm
movement
and
we'll
see
why
that's
important
later,
you've
then
got
once
the
package
goes
into
that
delivery
shoot
it's
going
to
go
to
a
particular
delivery
truck.
A
Now,
as
we
said,
a
lot
of
inferencing
happens
at
the
edge
and
we
want
to
also
make
sure
that
the
important
events
are
being
filtered
and
synced
back
to
headquarters
depending
on
what
network
is
available
and
what
bandwidth
you
have.
And
why
do
we
want
to
do
that?
Well,
there's
a
number
of
different
reasons,
but
you
might
want
to
have
an
audit.
A
So
here
we
have
the
icons
put
into
place
which
describe
some
of
the
open
source
components
that
we
can
use
so
you've
got
the
application
deployment
mechanism
in
k-native.
You've
then
got
also
rabbitmq,
acting
as
a
Consolidated
event
stream,
and
then
you've
got
a
number
of
different
processes
happening
which
allow
you
to
filter
the
data
using
spring
using
key
Value
Store
of
geode
and
then
also
using
the
deployment
rule
set
to
deploy
that
AI
model
back
to
the
edge,
and
why
is
that
important?
A
Why
are
some
of
those
future
analytics
use
cases
important,
well
a
number
of
different
things
you
can
do
with
that
data
and
the
fact
that
you
have
a
common
API
which
allows
you
to
do
that
greatly
reduces
the
time
to
develop
some
of
those
new
applications.
So
you
know
we
can
take
some
examples
here
where
you've
got
training
staff
can
be
easier
because
you've
actually
built
up
a
digital,
twin
type
model
of
an
end-to-end
package,
workflow
and
everything
that
happens
in
that
package
workflow.
So
you
can
try
and
help
staff
understand
what
to
do.
A
You
can
also
do
things
like
predictive
maintenance.
You
never
mentioned
the
diverter
arm.
That's
moving
packages
to
different
locations.
You
can
keep
track
of
what
that
means
for
the
diverter
arm
and
what
you
need
to
do
in
terms
of
maintaining
that
machine
so
that
you
can
prevent
it
from
breaking
down
and
actually
stopping
the
flow
of
packages.
A
You've
then
also
got
things
like
you
know.
Future
automation
use
cases,
things
like
forklift
trucks,
or
you
know,
lifting
and
shifting
AI
controlled
trolleys
that
could
move
packages
to
different
locations
and
you've
even
got.
You
know,
dynamic
scheduling
of
maybe
even
automated
and
self-driving
Vehicles
delivery
vehicles
in
the
future.
But
in
order
to
do
all
those
things
in
order
to
train
that,
you
need
to
start
keeping
track
of
the
events
that
are
happening
today,
so
that
you
can
work
out
or
even
model
before
you
start
some
of
these
expensive
projects.
A
A
Okay,
so
what
we've
got
here
is
we've
got
a
dashboard
and
the
idea
of
this
dashboard
is:
it
shows,
in
a
retail
location,
a
fridge
door
being
open
or
closed
and
there's
also
a
number
of
sensors
in
this
demo,
which
show
the
temperature
of
various
food
items,
and
you
can
see
there's
alerts
and
warnings.
So
the
intelligence
comes
here
not
from
just
taking
the
silos
of
data
from
each
of
the
sets
of
sensors,
but
also
processing
and
correlating
the
data
at
the
edge
location.
A
Then
filtering
it
such
that
only
the
relevant
alerts
are
communicated
back
to
headquarters.
So
we've
got
some
components
here:
Communications
being
done
locally
using
mqtt
via
rubber
mq.
We've
also
got
Apache,
geode
or
gemfire
used
to
correlate
events,
so
the
door
open
and
higher
temperatures
and
what
we're
using
here
is
spring
cloud
data
flow
to
deploy,
applications
which
are
going
to
analyze
some
of
the
sensor
data
and
they
deploy
them
as
streams.
The
stream
is
essentially
a
spring
boot
application.
A
That's
going
to
be
deployed
at
the
edge,
and
we
have
here,
as
you
can
see,
a
scripting
language
which
defines
the
rule
set,
so
we're
looking
for
temperatures
over
a
certain
value
and
the
door
being
open
or
closed.
If
the
doors
open,
we
send
a
warning
and
we're
also
logging
everything
into
a
rabbitmq
stream.
So
we
can
order
and
track
what's
happening
if
we
want
to
later,
as
you
can
see
here,
we're
creating
the
streams
spring
cloud
data
flow
is
actually
deploying
those
streams
and
deploying
those
applications.
A
And
the
important
thing
here
is
that
you
know
only
the
correlated
events
are
being
communicated
as
being
actionable,
so
we
don't
currently
see
any
alerts
or
Warnings.
Currently,
what
we
need
to
do
is
generate
some
data
from
the
sensor.
So,
as
you
can
see,
we've
got
a
temperature
sensor
here,
showing
a
value
of
11
degrees.
A
B
A
Has
a
door
open?
What
we're
seeing
is
that
warnings
are
coming
through
because
there's
a
door
open,
and
so
the
idea
is
then
that
the
employee
should
go
and
close
the
door
and,
as
we
see
here
what's
happening
neatly
is
we've
got
our
local
application
running
on
the
left
and
then
we've
got
only
the
well.
We've
got
data
being
replicated
to
another
application,
that's
running
in
the
cloud.
So
again
we
want
to
process
locally
and
then
broadcast
data
to
the
cloud
only
as
required.