►
Description
Shan Yu, Alibaba Cloud
Hyperledger Fabric has a very comprehensive design and is suitable for cross-enterprise business networks. But there are still challenges when it comes to deployment in a production environment, especially from storage's and network's perspectives. In this topic, firstly, Shan will provide a storage hotspot analysis against Hyperledger Fabric, and provide a formula for storage capacity estimation. Furthermore, Shan will suggest a cloud-native storage solution for key components (Orderer, Peer, Kafka, ZooKeeper, CouchDB) of Hyperledger Fabric to achieve high availability. Secondly, Shan will summarize typical networking patterns as well as pros&cons for consortium blockchain. Then Shan will present a cloud-based solution which can securely connect organizations within VPC on public cloud and within the private data center while protecting business applications behind blockchain.
----
Hyperledger – Open Source Blockchain Technologies
Hyperledger is an open source community focused on developing a suite of stable frameworks, tools and libraries for enterprise-grade blockchain deployments.
It serves as a neutral home for various distributed ledger frameworks including Hyperledger Besu, Fabric, Sawtooth, Indy, as well as tools like Hyperledger Avalon, Cactus and libraries like Hyperledger Aries, Ursa. Learn more about Hyperledger projects: https://www.hyperledger.org/use
Case Studies: https://www.hyperledger.org/learn/case-studies
Training & Certification: https://www.hyperledger.org/learn/training
Tutorials: https://www.hyperledger.org/use/tutorials
Webinars: https://www.hyperledger.org/learn/webinars
Events: https://www.hyperledger.org/events
Vendor Directory: https://www.hyperledger.org/use/vendor-directory
Subscribe to the Hyperledger Newsletter: https://www.hyperledger.org/newsletter
Follow-us on Twitter @Hyperledger
Learn about Hyperledger Membership: https://www.hyperledger.org/about/join
#Hyperledger #HyperledgerGlobalForum2018 #Blockchain
A
A
Okay,
first
of
all,
I
want
to
introduce
a
bit
the
background,
so
I
assume
everyone
here
knows
about
what
watching
is
just
as
described
in
this
narrow
sense.
But
what
is
more,
attracting
us
is
that
in
this
process
we
think
blockchain
is
a
kind
of
new
schemer
and
collaboration
model
for
the
organizations,
individuals
and
machines
to
construct
a
distributed,
trust
network
to
connector
trusted
data
and
let
the
value
start
flowing
among
them.
So
this
will
be
a
promising
future,
not
only
for
most
industries
but
also
for
the
whole
society.
A
So
therefore,
Alibaba
cloud
started
the
journey
in
blockchain
and
we
chose
hibachi
as
one
of
our
major
directions.
Here
are
the
key
reasons.
The
first
one
is
openness.
Openness,
bring
people
the
trust
in
technology
rather
than
a
proprietary
knowledge
which
runs
live
black
box.
People
cannot
just
trust
it.
The
second
one
is
the
inference
with
this
large
inference.
We
are
able
to
build
this
global
large
technical
and
business
ecosystem,
and
the
third
one
is
that
we
found
in
high
pleasure,
especially
in
high
blood.
A
Repeat
it-
is
it
was
born
with
our
enterprise
oriented
design
that
means
from
functional
to
architectural.
It
provides
very
high
standard
of
security
stability,
availability,
scalability
slasher,
so
this
our
customer
enterprise
customer
care
most
so
Alibaba
Klaus,
is
dedicated
to
pub
you
entrusted
infrastructure
for
everyone
and
position
yourself
as
an
open
platform
and
is
willing
to
contribute
to
the
ecosystem,
and
the
Alibaba
class
mission
is
to
hell
and
price
to
do
the
digital
transformation.
A
So
what
we
have
you
done
in
past
two
years
with
high
bladder
so
early
in
2017
in
October,
we
released
the
first
blushing
solution
of
Alibaba
clock
and
in
the
service
in
that
solution
we
provide
one-click
deployment
for
distributed
fiber
network
to
reduce
the
hours
or
days
effort
interest
just
several
minutes,
and
in
December
last
year
we
upgraded
to
support
the
communities
and
also
incorporated
export.
In
July
this
year
we
launched
the
portion
as
a
service
in
China,
and
we
are
also
the
globally
the
first
blushing
product
which
combined
Intel
SGX
technology
with
hydrogen
fabric.
A
Then,
in
September
this
year,
together
with
Brian
burned
off
in
Hangzhou,
we
made
a
leap,
Ibaka
cloud
service
globally
available
for
the
whole
international
market.
So
today,
just
announced
in
this
morning
we
are
very
honored
to
become
official,
become
the
member
of
hi
pleasure
and
by
close
collaboration
with
hi
pleasure.
We
expect
more
achievement
in
their
2019
and
ahead.
A
Okay,
today
you
have
seen
blockchain
has
been
applied
in
many
industries
and
scenarios
and
in
ibaba
group
we
have
also
applied
touching
in
many
industry
and
signals.
As
listed
in
this
line,
and
just
give
you
a
few
example,
we
have
helped
our
customers
to
use
blockchain,
to
encourage
safe
driving
and
avoid
traffic
accidents
and
customer
managed
to
create
value
out
of
their
driving
later
contributed
by
people,
and
second
example
is
that
we
are
working
with
the
AI
team
to
build
a
trusted
digital
accountant
platform
for
the
copyright
protection
and
the
trading
of
digital
contents.
A
The
third
example
is
that
we
are
working
with
Audi
team
to
build
a
platform
for
trusted
device,
SS
and
trusted
date
for
the
industrial
internet.
So,
in
our
vision
with,
we
are
not
only
creating
the
incremental
value
for
enterprise,
which
is
improving
the
efficiency
and
reducing
the
cost.
We
are
also
keen
on
creating
the
transformation
value
for
enterprise,
that
is
the
new
business
model.
A
The
new
signals
and
new
economic
wish
will
not
exist
if
without
blushing,
ok,
so
when
we
apply
blushing
in
advising
hours,
we
have
encountered
a
series
of
challenges
and
problems
and
we
managed
to
solve
them.
For
example,
in
security,
we
we
use
STX
to
protect
our
favorite
private
keys,
and
we
also
have
a
lot
of
secure
designed
to
address
the
security
risk
and
performance
after
optimization
for
single
channel
with
four
pianos
using
carpet
as
a
benchmark.
A
We
managed
to
achieve
1,600
TBS
for
writing
and
7,000
TBS
for
query,
but
today,
I'm
going
to
share
our
practice
around
the
storage
and
network.
So
first
is
storage.
Why
storage
is
important,
because
the
core
of
the
blockchain
is
the
ledger
and
the
larger
data
resides
in
the
storage.
So
this
is
the
reason,
so
we
encountered
these
two
major
problems
in
the
storage
when,
in
our
customer
engagement,
the
first
one
is
how
to
do.
A
Storage
capacity,
sizing
or
planning-
actually,
we
have
not
found
any
guidance
or
knowledge
in
about
this
question
before
and
but
for
our
customer.
They
need
to
do
the
storage
capacity
planning
and
worst
stage,
especially
in
the
private
cloud
deployment
or
local
center
deployment.
So
this
is
one
key
problem.
The
second
one
is
how
to
address
a
star
challenge.
I'll
explain
it
later.
A
A
It
is
a
Kafka
broker
which
also
stored
the
transaction
data
in
the
topic
kill,
but
we
can
set
the
retention
period
to
mitigate
the
pressure
and
the
other
part
slide
is
lady
B.
It
also
stored
the
value
which
is
related
with
the
transaction,
but
it
will
not
always
go
so.
We
explain
later.
Okay,
so,
with
this
trend
analysis
result
and
then
we
have
done
a
series
of
logical
analysis
and
sorry
testing
and
say
we
can
come
up
with
this
version.
A
2
of
the
formula
which
is
for
estimating
the
hydrogen,
favorite
storage
consumption,
size,
okay
and
you
can
also
use
the
Excel
format
to
customize
input
parameters
according
to
a
sinners
and
your
and
your
deployment.
So
next
I
will
explain
how
we
come
up
with
this
formula
and
the
magic
number
and
variety
in
formula
as
highlighted
in
the
yellow
color.
The
troop
on
9q
bytes
is
the
nest
size
of
a
transaction
in
fabric.
So
this
is
based
on
the
source
code,
analysis
and
the
test
results
and
1.9
kilobytes
is
necessary
back
and
the
three
point.
A
Three
four
kilobytes
is
a
rough
size
for
the
transaction
plus
the
footprint
in
the
block
index
DB
and
history
index
DB.
So
this
is
based
on
the
statistic
of
our
test
results
and
the
transaction
payload
multiplied
by
two
means
that
each
transaction
actually
contains
the
chain
code
proposal
pelo
and
the
proposal
response
payload,
so
both
of
which
contained
the
transaction
payload.
Oh
next,
one
is
the
number
of
blocks.
This
actually
is
a
number
which
is
hard
to
determine
because
a
block
will
be
cut
when
any
of
the
below
condition
will
be
satisfied.
A
For
example,
the
first
one
is
that
how
long
the
block
have
has
been
waiting
for
transaction?
The
second
is
how
many
transaction
in
a
block
has
been
as
in
assemble.
The
third
one
is
how
many
sides
of
how
many
bytes
in
size
for
transaction
in
the
block.
So
this
is
very
hard
to
determine,
but
for
specific
scenario,
we
can
still
calculate,
for
example,
in
highly
high
school
personnel
in
which
the
transaction
key
coming
in.
In
a
very
high
speed,
we
assume
the
max
message:
count
is
10.
A
That
means
for
one
10,000
transaction
will
have
1,000
blocked
generated.
So
this
is
a
one
example
of
calculation,
and
next
is
the
number
of
Kafka
topic
records.
So
this
is
not
the
number
of
Kafka
broker's
by
Ashley
is
the
number
of
Kafka
topic
records
and
by
default,
I
will
use
3
as
the
default
value
and
the
second
one
is
a
Coastie
be
compassionate
way.
A
So
actually
the
transaction
payload
store
as
a
value
in
the
stadia
and,
as
you
can
see
in
this
configuration,
the
coach
DB
by
D
for
support
a
snappy
compression
algorithm,
but
before
but
actual
compression
wave
varies
depends
on
their
payload
contents,
if
no
compression
at
all,
we
take
1
for
this
value
for
this
parabola.
But
if,
in
the
extreme
case
we
use
a
single
letter
will
repeat
1000
times,
and
in
this
case
the
compression
will
remain
nearly
to
zero.
A
So
this
is
how
we
take
value
for
this
parabola
and
the
next
one
is
the
narrow
type,
because
for
crazy
knows,
new
key
value
will
be
created
in
the
study
B.
But
for
update
sonars.
No
new
key
value
will
be
quite
in
the
state
division,
so
it
is
1
and
therefore
for
different
cases
assists
ok,
so
how
this
formula
works
was
the
effect.
A
To
be
honest
in
the
very
beginning,
our
intention
is
to
get
a
rough
estimate
about
the
level
of
starch
consumption,
for
example,
whether
it
is
several
gigabytes
or
hundreds
of
gigabytes
of
laser
data.
But
after
this
using
this
version
to
formula
and
compare
with
the
actual
storage
size,
we
found
that
you
can
see
from
this
diagram
the
orange
line
and
the
yellow
a
blue
line
already
overlap
together
or
very
close
together.
So
today
the
result
is
much
more
accurate
than
we
expected.
So
with
this
deeper
knowledge
about
the
storage
size.
A
Now
you
can
see
the
challenge
in
storage.
The
first
is
the
storage.
Consumption
keeps
growing
fast,
as
you
can
see
in
the
formula,
it
is
many
times
of
the
size
of
the
transaction
number
and
their
transaction
payload.
I
remember
there
is
a
feature
and
a
design
document
in
the
fabric
community.
The
energy
are
talking
about
some
cleanup
of
the
larger
data,
especially
get
rid
of
their
invalid
transaction,
but
actually
even
in
our
testing
or
even
all,
the
transaction.
A
A
valid
transaction,
the
rate
of
storage
grows
as
you
quite
horrible,
especially
when
some
of
customer
want
to
apply
blushing
in
the
business
which
is
throughout
the
China,
the
whole
national
skill
and
want
to
run
it
in
production
for
years.
So
this
will
be
very
challenging
problem
and
second,
challenge
is
how
to
ensure
the
stability
reliability
of
the
starch
and
minimize
impair
to
business
applications.
A
So,
thanks
to
cloud,
we
managed
to
find
a
good
solution
for
the
above
challenges,
so
we
used
which
shows
the
albergue
cloud
NASA
and
Alibaba
cloud
disk
as
our
larger
storage-
and
we
also
chose
this
too,
because
they
provide
very
good
high
reliability
as
study
here
and
also
which
we
choose.
The
SSD
tie
for
I/o
performance
and
kubernetes,
provided
with
abstraction
for
storage
resources.
So
we
use
persistent
volume
and
persistent
rolling
claim
to
request
a
mounties
start
into
the
fabric
part.
A
Without
caring
about
the
details
of
the
storage
and
in
the
sacrament,
you
can
see
how
we
apply
this
storage
with
different
fabric
notes.
I
want
to
explain
a
little
bit
about.
We
found
our
Baba
clown-ass
is
quite
fit
for
the
glossy
analyzer.
In
two
reasons.
The
first
one
is
supported,
seamless
and
dynamic
starch
expansion.
When
you
little
slice
get
reach
the
bar,
and
during
this
process
you
don't
have
to
restart
or
interrupt
your
fabric
notes
or
your
applications.
A
The
second
thing
is
that
some
customer
concerned
that
as
time
as
time
goes,
when
the
first
system
get
get
the
larger
size
will
that
this
aisle
converse
low
so
to
the
contrary,
nASA
has
the
characteristic
that
the
larger
volume
is
the
higher
stupid
is
that
is
provided
by
Alibaba
clown-ass.
So
in
this,
in
this
sense,
we
can
eliminate
this
concern
of
the
customers.
A
So
today,
in
our
actual
customer
reference
case,
there
is
some
Russian
system
have
been
running
in
the
production
environment
for
months
and
which,
in
life
millions
of
transactions
and
hundred
gigabyte
or
larger
data
on
our
platform.
Ok,
the
second
bit
permit
B
topic
is
network.
This
ashley
is
a
very
large
topic
and
today
I
only
managed
to
talk
about
some
typical
connection
patterns
based
on
our
practice.
A
A
The
first
one
is
transaction
flow
relate
which
happened
in
sequence,
as
you
can
see
in
this
for
color
of
lies,
so
the
client
in
the
beginning
are
connected,
CA
for
registration
and
enrollment
and
lankan
astral
in
Dawson,
pier
for
the
transaction
endorsement
and
then
connected
the
other
to
block
hassle
transactions
with
other
and
then
after
or
other
order,
the
transaction
assemble
into
blocks
and
will
deliver
the
blocks
to
other
pier
within
a
channel
for
validation
and
commit.
So
this
is
a
first
type
of
connection
and
the
second
types
of
connection
happens.
A
All
the
times
includes
like
each
organization's
leader,
pier
it
will
connect
with
the
other,
to
pull
the
data
from
the
other
and
then
initiate
a
dissemination
of
the
blood
data
to
a
random
number
of
peer
inside
the
organization,
and
second
one
is
each
organization's
anchor
here
will
be
connected
for
discovering
the
peers
in
different
organizations
and
among
those
are
those
peers.
They
have
a
cough
second
communication,
many
for
three
purposes,
lighter
detecting
peers
and
this
son
disseminated
all
agitator
and
for
the
new
peer
will
transfer
the
larger
tree.
A
So,
with
this
connection
model
summary,
we
have
a
conclusion
that
for
fabric
to
function
normally
we
need
to
have
a
full
mesh
network
for
it,
yes
and
then
we'll
and
then
we'll
come
up
with
some
cheap
coke
and
action
pattern
to
connect
fabric
services.
The
first
one
is
quite
simple:
if
we
deploy
a
vibrant
network
inside
a
kubernetes,
we
can
just
use
a
slave
cook.
A
community
service
named
as
a
viable
service
name
and
communities
will
resolve
the
service
type
name
to
the
post
class
IP
automatically.
A
So
behind
the
scene,
it
is
the
coop
DNS
to
resolve
service
names
of
service.
Ip
and
that
could
proceed
to
translate
the
service
class
IP
to
the
pulse
class
IP.
So
what
you
need
to
do
is
just
to
define
the
favorite
nodes
using
the
community
service
name
in
your
committee,
s
llamo
or
critical
feet
or
mo
and
any
similar
configurations.
So
it
is
a
very
easy
to
use
and
simple,
usually
Youssef
in
the
development
and
testing
or
the
private
chain
deployment,
and
second
one
is
second
Ares
ferry
services
in
different
regions.
A
So
in
this
scenario,
for
example,
we
have
a
P
organization
in
one
regions
and
the
other
service
in
another
region,
so
we
still
use
communities
as
an
example,
because
today
it
is
already
a
standard
for
application
infrastructure.
So
here
we
can
use
a
external
service
important
to
define
this
order,
so
we
can
divide
the
external
address
and
the
port
of
the
target
order
and
define
this
inside
this
region
a
so
that
pier
can
connect
with
this
external
surface
endpoint
and
finally,
which
this
order.
So
in
this.
A
A
Okay,
the
next
one
is
more
flexible.
Is
that
still
in
Snell
of
different
regions,
but
here
we
can
use
a
global
host
name
for
every
fabric
services,
and
then
we
just
heard
the
host
name
in
a
global
DNS,
and
also
we
assigned
public
IP
and
the
pirate
port
for
each
services,
so
that
when
we
visit
this
host
name,
the
DNS
can
translated
to
the
pipe
idea
of
each
services.
So
this
way
can
satisfy
most
of
their
basic
customer
requirements
and
very
flexible
and
also
provide
a
very
good
abstraction,
especially
for
external
blushing
applications.
A
So
today
this
is
a
default
option
of
Alibaba
Klopas
and
the
next
one's
quite
a
bit
more
complex
is
for
us
know
that
the
cost
different
customers
they
want
to
deploy
the
very
nose
in
different
VPC
of
our
private
network
on
the
cloud,
and
they
also
have
very
now
knows
our
services
in
the
local
data
center
and
they
want
to
establish
a
consortium
network
to
connect
all
the
customers.
Note
in
this
case
also
thanks
to
cloud,
we
have
a
CNS
called
cloud
enterprise
network.
A
It
can
connect
all
the
VPC
into
a
ring
network
so
that
this
VPC
can
connect
to
each
other,
using
just
internal
IP
and
without
exposing
to
the
public
network,
and
also,
we
also
want
to
leverage
the
flexibility
of
using
the
hostname
to
be
visiting
services.
So
we
also
use
a
DNS
provision
product
which
can
ensure
the
the
resolution
of
service
name
to
IP,
only
effective
with
vinius
v,
pcs,
so
also
for
those
customers
who
are
in
the
local
datacenter.
A
We
can
use
either
VPN
or
the
leased
line
to
connect
with
any
of
the
V
PC
and,
furthermore,
connect
to
the
whole
ring
network.
So
this
is
already
supported
by
arabic,
help
us
today,
and
I
want
to
emphasize
that
it
is
not
only
applicable
to
highlight
the
fabric,
but
also
applicable
to
any
other
brushing
technology.
A
So
in
the
end
I
want
to
do
a
quick
summary.
So
today
we
think
the
reissue
relationship
between
cloud
and
blockchain
is
a
kind
of
perfect
match
and
by
collaboration
with
a
pleasure,
a
we
won.
We
wish
the
strength
of
both
sides
can
be
reinforced
and
also
we
welcome
any
partners
and
friends
around
the
world
to
join
us
so
that
we
can
accelerate
the
development
of
the
Russian
system
together
to
a
greater
success
in
future.
Ok,
that's
all
for
my
sharing
for
today
and
thank
you
for
listening
and
any
question.
A
B
A
Okay,
this
is
a
good
question,
so
actually
we
have
done
the
first
round
of
performance
testing
and
also
we
have
done
a
series
of
mors
performance
in
nouse.
So
currently
we
have
a
tested
lighter,
according
to
some
custom
scenarios
like
10
organizations,
with
two
notes
for
each,
so
it
can
run
in
their
test
environment.
So
currently
we
have
not
seen
some
physical
limitation
so
but
Esther
scenarios
of
performance
goes
or
maybe
we'll
have
more
summary
about
this
yeah.
B
A
A
Okay,
so
yeah.
This
is
a
very
interesting
question
about
auto
scaling
about
the
Diploma
on
cluster.
So
actually
we
have
done
some
analysis
about
the
application
pattern
for
blush.
It
contains
two
part.
The
first
one
is
the
blossom
number
order.
System
excel
is
a
kind
of
data
entry
of
complex
system
so
forth
such
a
system
like
the
database.
We
are
not
treated
as
a
stylist
or
can
be
scale
easily,
so
it's
kind
of
like
maybe
you
use
some
nose
lateral
to
deploy
combined
with
their
local
volume
or
your
user.
A
Just
now,
the
nasa
ecologist
for
the
data
persistence,
but
the
principle
is
that
you
not
easily
scale
out
one
peer
nodes
easily.
This
is
not
the
bonnet
by
the
native
design,
so
we
mul
I
use
auto
screen
for
our
plus
an
application,
because
partial
application-
maybe
Australis,
maybe
web
application
or
any
other
application.
So
that's
a
good
example
to
apply
the
auto
scaling.
A
So
that's
in
our
project,
so
for
blushing
notes
you
can
easily
do
a
DDT,
auto
scaling
for
the
expression
keynote
yeah,
but
in
another
other
hand
the
piano
itself
is
also
natively
provided
higher
very
bit
here,
because
they
they
have
the
same.
Lather
can
replicate
each
other.
They
can
provide
the
services
when
one
goes
offline,
yeah,
okay,
not
sure.
If
I
have
answered
your
question.
B
A
Yeah,
so
it's
quite
coincident
that
I
also
confirm
their
continued
service
department
of
Ababa
cloud.
So
we
have
already
supported
this
note
as
far
as
the
dynamically
skill
out
to
more
virtual
machine
under
the
kubernetes
and
also,
we
can
also
scale
the
pause
of
the
communities
according
to
the
autoscaler.
Some
criteria
set
yeah
and
also,
let
me
make
some
yeah.
So
this
is
already
an
supported
feature
by
Arabic.
How
contain
your
service.
B
B
A
With
that,
okay,
oh,
this
is
a
very
good
question.
I
love
very
much
because,
just
several
weeks
ago,
I
presented
a
speech
at
the
Linux
called
a
cloud
native
called
talking
about
this
topic.
So
this
is
a
long
existing
challenge
and
because,
in
the
lifecycle
of
container
at
the
pier
will
involve
a
docker
demon
to
first
create
a
fabric
city
base
called
chain,
chang-ho
builder
environment,
to
build
a
channel
image,
and
then
I
called
the
doctor
daemon
to
start
a
single
image.
A
single
container.
A
This
is
all
out
of
the
communities
lifecycle
your
hard
to
do
the
management
throughout
the
chain
code
lifecycle.
So
we
have
already
in
fabric
JIRA,
have
a
feature
about
taking
the
chain
code
management
into
the
communities
like
pop,
but
still
not
implement
it,
but
we
think
that's
the
ideal
way
to
solve
the
problem.
So
we
have
also
experienced
experienced
they're
talking
docker
solution,
but
we
found
there
are
more
disadvantaged
than
advantage
of
this.
A
This
approach,
because
especially
the
speed,
will
become
very
slow
because
I
need
to
pull
the
1.3
gigabytes
of
every
season,
TV
based
image
each
time
you
start
off
here
and
also
the
dock.
Andhaka
approach
is
more
used
in
the
CIC
D
and
that
Bob's
not
for
production,
because
the
stability
changes
quite
obvious.
So
today,
in
Arabic
loud
bass,
we
use
an
other
way
is
true,
solve
the
key
problem.
A
First,
we
ensure
by
using
the
configuration
and
some
IP
table
in
our
policy,
we
ensure
the
communication
between
P
and
Jenko
is
okay,
and
then
we
use
the
now
policy
to
restrict
their
impound
and
outbound
connection
for
the
channel
container.
So
this
still
I
think
our
trade-off
may
be
in
future.
Maybe
we
or
you
can
together
contributors
the
final
solution,
the
source
code
level,
to
this
problem.
Okay,
thank
you.
B
Can
you
probably
share
your
experience
on
how
putting
notes
and
peers
in
different
regions
affect
the
actual
boxing
performance
like?
Does
it
actually
influence
it
somehow
or
like
when
you
have
an
old
one
in
I,
don't
know
one
in
Singapore
and
in
the
second
one
in
us
and
the
third
one
in
China
like
how
does
it
affect
the
blockchain
phones,
okay,.
A
So
this
is
a
very
good
scenario.
Actually
to
be
honest,
I
have
not
done
excited
seems
in
there,
but
we
can
I
can
talk
about
based
on
the
performance
experiences.
So
first
principle
is
I.
Don't
know
whether
the
child
English
term,
for
this
is
like
the
bucket
theory.
The
shortest
straw'
shortest
part
will
determine
how
many
water
can
contain
in
the
bucket
right.
So
it's
same
to
the
performance,
because
in
the
entry
m
of
android
architecture,
the
weakest
part
determines
the
whole
performance
of
the
whole
system.
A
So
in
our
performance
we
found,
if
in
a
very
high
speed
network
environment,
then
the
major
bottom
that
is
in
the
CPU
is
computing
power
because
we
found
the
fabric
is
very
calculation
intensive
kind
of
system.
So
this
will
be
a
one
major
optimization
and
the
other
one
is
like
the
configuration
about
the
just
mention
like
how
a
block
is
being
cut,
the
the
parameter
in
the
channel
config.
So
this
is
the
second
optimization,
but
in
the
scenario
just
described
in
the
cost
region,
especially
course
continental
snows.
A
Now,
what
will
become
a
very
serious
part,
I
think
in
maybe
two
maybe
two
a
spare
one
is
getting
the
endorsement,
because
the
client
need
to
cannot
show
a
different
peer.
According
to
the
endorsement
policy,
two
different
peer
organization,
peer
gather,
endorsed
many
response
and
cannot
for
the
others.
So
this
part
the
network
latency.
Maybe
it
will
become
a
problem
right,
oh
by
the
way
I
want
to
mention.
There
are
two
metrics
for
performance.
A
The
transaction
process
context
the
segment
instant
latency,
so
in
this
part,
net
were
latency
may
decide
the
whole
performance
latency
and
second,
one
is
about
the
order
to
deliver
drugs
to
other
peers
and
the
third
one
is:
if
you,
your
peer
relies
on
the
channel
avenge
will
notify
about
the
final
commitment
of
the
block.
You'll
also
wait
for
the
kind
of
event
to
arrive
at
the
class,
so
if
three
part
may
come
together
will
slow
down
the
whole
performance,
because
never
latency
is
very
significant
in
these
scenarios.
B
A
Because,
today,
fabric,
before
fabrics,
support
RAF
in
the
future
versions.
Today
the
Capcom
is
still
not
so
distributed
because
other
in
nature
is
just
a
Kafka
producer
and
consumer
client.
So
you
can.
Maybe
you
can
put
other
in
very
long
distance,
so
this
will
slow
down
the
connection
between
client
to
the
Kafka
broker.
So
this
may
be
one
solution,
but
I
I
cannot
guarantee
the
performance
will
be
better
in
this
way
yeah.
So
let
me
think
about
another
approach
for
this.
A
You
know
no
latency
here
so
because
different
regions,
the
network
connection
or
bandwidth
we're
still
different.
So
we
better
we'll
do
a
good
planning
about
the
placement
of
different
notes
in
these
regions.
So
because
we
know
the
network
speed,
especially
when
it
comes
to
cross
China
from
China,
obviously
may
be
different,
so
yeah,
you
know
it.
A
In
Scenario
just
presented,
it
is
a
different
community
cluster
in
different
regions,
but
actually
today,
our
container
service
also
support
different
kind
of
communities,
maybe
not
implosion.
But,
for
example,
is
the
communities
cluster
spanning
different
availability?
Soon,
that's
once
in
now,
and
the
other
one
is
use
community
Federation
to
to
combine
different
communities
in
different
regions
into
single
class
desert
federation.
Okay,.
A
Okay,
this
also
a
good
question
so
that
different
choices.
First,
we
identify
CA,
is
a
kind
of
low
frequency
usage
and
not
the
transaction
directly
related
services,
so
communities
for
a
very
good
recovery
or
kind
of
caused
them
by
a
change.
So
we
use
a
single
si
can
certify
most
of
their
situation,
but
higher
stand-up.
We
can
use
the
C
cluster
because
ferry
also
support
the
C
cluster
is
a
cluster
of
database
for
that
deployment.
Yeah.
A
Oh,
this
is
very
interesting
question
actually
I
replied
to
Chris
Ferris
last
week
in
the
high
black
male
group,
because
Chris
proposed
interoperability
between
different
plots,
invented
special,
have
larger
vendor
to
create
a
mobile
application
which
can
run
between
different
blushes
and
service
vendors
to
make
it
work
and
proof
for
the
interoperability
is
feasible.
So
I
think
this
is
a
good
direction
that
we
are
working
towards
it.
So
maybe
in
future
we
can
work
together
to
make
it
happen.
Okay,.