►
From YouTube: Keynote: The Big Leap: Conquering the Legacy System and Building a Scalable Cloud... Tan Lin
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
Keynote: The Big Leap: Conquering the Legacy System and Building a Scalable Cloud... Tan Lin
As the world's leading transportation platform, traditional IT architectures can no longer meet the rapid growth of DiDi's business needs. Embracing cloud native is the way to go. However, it's still a huge challenge to ensure a smooth transition from the legacy system to the new system, while minimizing the cost of business transformation. After three years of development and hands-on experience, our custom Kubernetes-based platform, Elastic Cloud, now handles most of the core business. It manages ~50,000 pods running on over 3000 servers. We will describe the evolution of our cloud platform; the challenges with scheduling and isolation; the difficulties of containerization; and our technological breakthroughs.
https://sched.co/NzAQ
A
A
Just
now
far
away
AWS
and
a
while
our
way
and
other
group
have
talked
about
the
benefits
advantage
and
offer
kubernetes
and
the
cloud
native
I
want
to
talk
about
our
learning
and
the
other
three
parts
are
filter.
First
of
all,
it's
about
our
purple
and
it
is
about
the
scalable
cloud
and
the
some
of
our
our
practice
and
stories
and
the
other
thing
some
of
my
learnings
and
the
Maya
prosper.
Well,
scalable
syllable
clouds
as
one
of
the
business.
A
Our
containers
offers
a
lot
of
work
like
a
line
of
line
divorce
and
that
the
architecture
is
not
a
very
sophisticated
apart,
lateral,
online/offline
mixture
and
the
service
fortune
and
all
is
a
full
service
to
the
pollination
of
it.
We
have
kubernetes
and
dock
home
with
our
technological
service
and,
of
course,
we
need
ambient
assistance.
Support
I
will
not
go
into
details.
Applaud
apart
for
our
private
cloud.
We
also
customize
the
private
cloud
of
order,
customers
which
is
being
implemented
now,
I
want
to
share
with
you
some
data
of
a
scalable
cloud.
A
We
have
a
3k
physical
machine,
the
fifty
thousand
containers,
40%
peak
CPU
utilization
and
as
a
DV,
and
it
is
up
to
50
percent
covered
by
a
line
which
are
start
from
zero.
But
now
we
reach
the
level
of
one
and
I
want
to
share
with
you
my
story
from
zero
to
one
three
years
ago:
dd
faced
the
several
difficulties,
low
resource
utilization,
complex
test,
employment
law,
happy
expansion,
time
and
the
physical
machine
for
the
scalability
and
a
very
low
response
time,
and
though
we
need
to
make
some
somewhat
changes.
A
But
how
and
the
one
who
change
three
years
ago
in
2016
kubernetes
was
still
very
young
and
though
our
task
aversion
to
1.3.
But
so
we
really
focus
on
the
acceptability
of
a
kubernetes
api.
A
declarative,
API,
inaudible,
highly
scalable,
only
imitating
is
promises
and
the
1.3
and
a
single
machine
1002
a
company
like
a
DD
such
a
large
company,
which
is
not
enough.
Therefore,
internally
we
had
a
different
services.
A
That
is,
the
system.
Integration
and
a
DD
order
service
will
be
transferred
through
the
cloud.
Therefore,
in
our
designing,
we
need
to
are
compatible
with
the
existing
architectures
user
habits
and
try
to
try
to
have
more
stability,
container
physical
machine
working
in
parallel
and
for
a
long
time
they
will
exist
in
parallel
and
a
to
be
specific
of
what
we
do
in
one
word.
The
right
one
is
the
best
and
we
didn't
use
we
a
native
container.
We
try
to.
A
And
then
we
have
a
fifth
set
under
the
support,
different
versions
of
resonances,
and
we
support
the
customer
customized,
a
transcript,
which
is
very
important.
We
must
ensure
that
when
all
the
service
is
ready,
we
will
connect
them
and
with
this
in
26:17,
and
we
have
some
core
business
and
the
CEO
for
which
is
not
enough,
and-
and
we
welcome
the
second,
the
challenge
that
is
internationalization
and
in
2018
we
want
to
go
global.
A
We
want
to
have
the
new
system
and
the
zr4
onto
the
bottom
of
it,
and
the
scalable
cloud
will
provide
100%
percent
online
traffic
for
international
business.
If
you
are
responsible
for
stability,
you
need
to
have
a
partnership
of
the
traffic.
You
know
we
only
have
one
one
up:
a
room
or
one
a
server
room
in
the
path
of
the
only
plan
plan
B.
A
But
now
we
only
have
180
plan
here,
for
we
have
done
a
lot
of
a
modification
and
to
the
bottom
of
it
we
should
have
international
business,
but
according
to
the
requirements
of
PR,
we
deleted
and
the
general
speak
a
and
where's
our
scalability
and
scalable
cloud
always
works.
Never
let
us
down,
therefore,
within
DD,
and
we
have
very
good
reputation.
So
stability
is
better
than
everything
and
then
to
be
specific
what
we
have
done.
A
There
are
four
points
and
the
you
may
see:
data
from
SPL
shifted
to
the
underlay
Nashville
M
and
the
storage
from
cell
to
the
local
local
storage.
Under
these
two
solutions
sacrificed
their
flexibility,
but
so
we
remove
the
uncertainty
which
worth
it
in
order
to
cope
with
the
solutions
to
the
next.
We
customize
the
logic
under
and
though
we
have
a
a
local
startup
of
the.
After
of
the
containers
in
running
the
containers
and
the
we
restrict
the
threat
of
the
containers,
I
want
to
give
you
very
specific
of
performance,
optimization
the
sight
3500
SPF.
A
Default
a
wit
time
of
140
seconds
per
service
example,
because
the
natives
but
clowder
has
a
default
a
synchronization.
Therefore,
the
work
cannot
be
perfect
and
informal
mechanism
and
the
witness
can
ensure
that
so
SK
us
and,
though,
will
not
be
missed
out
of
our
optimization.
We
are
the
lower
99%
of
the
waiting
time
and
the
other
thing
is
the
isolation
in-house,
such
as
CPU
and
solution,
network,
isolation,
disk
I/o,
isolation,
I
believe
all
of
us
are
working
on
it.
A
And
so
you
know
that
this
is
not
enough
and
go
with
your
scalability,
and
the
new
requirements
are
raised
and
went
is
lower
than
40%
and
no
new,
more
machines,
and
you
know
that
is
really
attached
to
our
discussion
or
orchestration,
algorithm
and
ok
and
though
what
is
the
limit
request?
Ratio
request
is
very
special,
interesting
value
and
is
a
minimum
availability
which
means
that
search
on
a
horse.
What
is
the
biggest
number
on
the
containers?
And
it
is
a
one
of
the
indicators
for
competition
if
this,
the
tier
1
tier
3
level
service
each?
A
A
Racial
try
to
classify
by
the
service
we
try
to
classified
between
the
different
hosts
other
end
result
doesn't
work
very
well
because
requests
that
doesn't
count
as
the
reality
real
time
assessment
is
fine,
but
you
need
to
have
the
updates
from
the
customers
less
flexible,
more
involvement
from
the
customers
and
the
zr4,
and
we
use
a
low
cost
and
the
good
result.
The
solution-
and
we
have
the
first
seven
days-
value
improved
density
from
25
to
40%,
CPU
utilization
standard.
A
This
is
the
CP
secure
Idol,
and
we
have
a
peak
in
the
chart
on
the
left
hand
to
sister
picture
and
you
can
see
our
top
10
and
25%
and
we
have
a
300
post
with
the
idol
talent,
utilization
maximize
the
important
5%
and
the
not
don't
talk
about
a
future
outlook
and
it
is
very
helpful,
and
so
we
want
to
continue
on
it
and
in
the
future
we
want
to
optimize.
The
hybrid
deployment
native
container
is
one
last
target.
We
need
to
do
a
lot
of
things.
A
The
last
one,
some
of
my
insights
and
how
to
implement
these
technologies.
We
need
to
solve
the
problems
that
of
eating
problems.
Therefore,
we
need
to
deploy
the
most
suitable
service
and
the
technologies
try
to
minimize
across,
and
this
is
a
systematic
process
and
do
not
think
about
the
world.
Word.
That's
around
the
city
o
or
the
bosses
will
do
everything
and
we
need
to
scale
up
the
business
with
the
users.
No
matter
you
engage,
users
are
engaging
our
team
members
it's
easier
and
how
to
persistently
create
a
value
for
users.
That
is,
data
driven.
A
A
The
trust
of
the
users
is
kind
of
X
and
if
you
use
up
the
chance
that
they
will
leave,
if
your
subproblems
for
the
users
tomorrow
and
as
a
well,
be
your
ambassador
and
have
more
service
from
your
sites
and
well
I
want
to
talk
about
our
suggestions
and
the
probably-
and
it
is
a
very
recommendations
on
the
future
development
of
our
communities.
Communities
are
really
developing
fast.
One
is
to
up
the
other
ones
go
down.
A
If
we
go
up,
we
hope
to
human
communities,
can
al
cater
to
the
business
orchestration
and
a
scalability
such
as
a
VR
scheduler
ERD,
are
always
in
our
section.
If
we
go
down,
we
need
to
optimize
the
threat
for
performers.
Etcd
3.4
online
is
something
we
are
looking
forward
to.
The
other
thing
is
isolation
of
the
container
and
we
will
in
forward
projects.
Lakota
could
become
a
large-scale
case
and,
finally,
a
little
advertisement
available
cloud.
We
keep
repeating
from
1
to
100.
We
are
in
this
process.
A
We
have
many
challenges
and
the
film
a
technological
point
to
probe
into.
If
you
show
interest,
you
can
ask
an
hour.
We
chat
and
I
hope
that
the
Masters
can
talk
to
me
offline
and
our
kernel
team,
as
made
to
say
this.
We
also
recruit,
and
we
have
lot
of
things
to
make
break
group
and
you're
welcome
to
spare
my
wheelchair.
That's
the
end
of
machinery.
Thank
you.