►
From YouTube: Dragonfly 2.0 — Bring a breakthrough Leap to the Domain of Cloud Native File... Zuozheng Hu & Jim Ma
Description
Don’t miss out! Join us at our upcoming event: KubeCon + CloudNativeCon North America 2021 in Los Angeles, CA from October 12-15. 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.
Dragonfly 2.0 — Bring a breakthrough Leap to the Domain of Cloud Native File and Image Distribution - Zuozheng Hu, Alibaba & Jim Ma, Ant Group
With the continuous development of dragonfly 1.X, more and more companies use it to solve the problems of image download and file distribution in their production environment. In this process, it also gradually exposed some deficiencies of the current system in architecture design, security and function, and faced more and more challenges. Therefore, it is necessary for us to launch the dragonfly 2.0 project and give it a clearer positioning: providing enterprise level (efficient, stable, secure, low-cost, product oriented) file distribution and management services, becoming the cloud native best practice and standard solution in this field. The new Dragonfly 2.0 will definitely bring a breakthrough in the field of file and image distribution.
A
A
First,
let's
took
it:
let's
take
a
look
at
the
current
situation
of
a
dragonfly
v1.
At
present,
the
number
of
start
of
dragonfly
v1
has
reached
more
than
500,
and
the
number
of
adoption
has
also
exceeded
110
involving
various
industries
such
as
telecommunication,
finance,
cloud
computing,
local
life
service
and
so
on.
There
are
turn
maintenance
in
the
project
which
are
from
alibaba
cloud
and
the
group
very
dense,
ebay,
anime
too.
At
present
there
are
77
contributors
here
we
warmly
welcome
more
people
to
join
the
dragonfly
project.
A
A
Second,
currently,
it
can
only
support
http
protocol
and
the
lacks
adaption
of
for
other
types
of
storage,
such
as
hdfs
storage,
sales
form,
various
cloud
and
vendors,
maybe
yum
and
so
on.
Thus,
it
greatly
restricts
us
coverage
seniors
and
the
further
kinders
propulsion
and
values
dimension.
Dimination
in
companions
and
communities.
A
Third,
currently
it
only
supports
the
active
poor
model
and
lacks
the
active
push,
as
well
as
synchronization
skills,
which
are
the
basic
conditions
to
satisfy
for
our
product
positioning
first
dragonfly
v1
lacks
the
protective,
no
task
management
and
the
control
data,
dashboard,
multi-tenancy
permission
control,
and
so
on
fees,
inconsistencies,
internal
and
external
versions,
high
maintenance
costs
and
the
difficulties
to
signal
synchronize
new
futures
and
the
problem
fixed
with
a
short
time
with
jungle
fly.
V1
v2
system
has
brought
some
innovation
in
many
aspects
in
terms
of
architecture.
A
Firstly,
the
subsystems
are
decoupled
from
each
other
and
support
on-demand
deployment,
which
makes
the
architecture
more
flexible.
Second,
it
supports.
It
supports
the
saddle
pad
tools
to
include
dragonfly
ability.
Natively
such
as
value
as
planned
strategy
downloaded
to
us
said,
com
completes
the
we
want
architecture
v2
supports
better
how
it
turns
horizontal
scalability
it
with
the
expansion
of
obsidian
node.
The
pressure
only
mode
source
remains
basically
unchanged.
A
A
Second,
through
the
time
analysis
of
the
disorder
of
the
client,
so
as
two
dynamic
addresses
a
pdp
network
strategy
structure
to
achieve
scheduling,
optimization,
sadly
other
schedule,
other
scheduling
strategies
are
supported
by
plugins
plugin
way,
such
as
intelligent
scheduling
based
on
machine
learning,
algorithm
on
the
client
on
the
client
side
through
the
yellow
copy
streaming
downloading
and
the
kernel
of
loading
technology.
The
io
efficiency
of
the
client
is
greatly
improved
in
terms
of
protecting
system.
A
A
A
A
Structure
of
dragonfly
v2,
which
many
includers
manage
scheduler
cdn
system,
client,
sdk
and
only
rpc
framework.
The
manager
is
the
central
controller
of
the
whole
system,
which
providers,
dashboard
user
task
management
and
the
configuration
management
and
the
exposed
last
four
api
scheduler
is
mainly
responsible
for
scheduling
peer
and
acting
as
a
metadata
trainer.
A
A
Next
up,
I
will
introduce
the
way
to
architecture
form
of
forms,
the
perspective
of
interaction
between
subsystem.
Let's
first
take
a
look
at
the
solute
solution
of
dragonflywave,
as
shown
in
the
figure.
The
whole
architecture
is
divided
into
layers
include
a
supernode
providing
scheduling
and
the
cdn
capabilities
and
the
dfg
client
deployed
on
each.
A
A
The
dftman
provides
a
service
which
is
responsible
for
downloading
and
uploading
file
blocks
and
providing
proxy
capabilities.
They've
got
an
sdk
requested,
df
daemon
to
generate
a
long
task
by
calling
a
pc
interface.
Finally,
the
manager
is
mainly
responsible
for
configuration
management
keeping
alive
and
controlling
the
subsea
subsystems
and
manage
is
also
optional.
A
A
Each
pure
host
represents
a
machine
instance
on
which
multiple
file
downloaded
jobs
can
be
triggered
similar
10
year
earliest,
and
each
download
job
corresponds
a
peer
task.
Each
peer
tasker
completes
a
whole
file
download
by
executing
multiple
pieces
tasks
and
the
pc
strategically
presented
a
five
block
downloading
task.
Finally,
multiple
peer
taskers
form
a
large
pdp
task,
and
the
five
blocks
will
be
transmitted
among
peers
in
the
pdp
task.
A
A
B
The
usage
of
dfget
is
fam
similar
with
the
double
get
it
can
download
the
files
with
pdp
p2p
network.
There
are
some
significant
changes
in
dfget
v2,
first
change
download
process
in
v2.
We
split
supernode
to
scheduling
the
cdn
and
the
ender
and
at
the
manager
for
multiple
cluster
configuration
supports.
B
Second
change
integration:
protocol
in
v1,
jfk
the
cost
of
node
with
http
per
protocol
now
in
v2,
all
action
done
by
grpc
building
with
http
support,
http,
2
sports
security
and
easy
integration
with
other
pro
programming
language
and
tours.
So
the
change
embedded,
df
daemon
some
words
for
df
demon,
the
dft
money
we
want
is
a
process
to
accept
http
request
and
it
will
transfer
the
request
to
the
dfg
commander
due
to
the
due
to
one
request:
one
they
have
get.
The
fdm
cannot
process
thousands
of
requests.
B
B
Not
only
change
the
protocol
and
the
module
code,
but
also
reflects
toward
high
performance
and
the
more
future
for
dfget.
Let's
discuss
the
this
enhancement
first
by
the
auction
stream
jpc
with
other
components.
Second,
upload
uploading
file
with
sending
send
file
and
the
splice
bus
benefit
from
go
long.
Trans
transfer
data
from
os
file
to
tcp
connection
is
done
by
golong.
Standard
library
is
without
memory
copy,
from
kernel
to
user
space
so
to
download
the
file
with
splice
support.
Unfortunately,
transfer
data
from
tcp
connection
to
os5
is
not
com.
B
A
B
B
B
Next
about
five
fire
digest:
normally
we
calculate
the
fat
that
is
using
standard
library
in
column.
Kernel
must
copy
all
data
from
user
base
to
all
data
to
user
space,
but
dragonfly
data
is
immutable,
immutable
in
most
cases
and
always
hit
the
page
cache
using
aflg
with
the
kernel
craft
or
api.
We
can
avoid
the
copying
data
from
kernel
to
user
space.
B
B
B
B
These
are
these.
These
are
about
com,
client
change.
Let's
take
a
look
at
our
ecosystem,
the
dragonfly
container
image
service.
We
call
it
tonight.
It
is
a
user
space
file
system
building.
On
top
of
the
container
immediate
format,
the
duty
of
status
is
to
reduce
the
cost
of
image
polling,
accelerate
the
long
speed
of
containers
enabling
the
load
container
on
demand
and
as
well
as
ensure
the
integrated
integrity
of
the
image
contender
in
the
largest
secure
data
center.
B
Like
and
group
and
alibaba
group
that
cracks
the
case
of
netherlands,
thousands
of
connect
containers
are
launched
very
second
and
the
thousands
of
images
are
hosted
by
dynasty.
At
the
same
time,
dragonfly
have
done
this
to
do.
P2P,
container
delivery
to
solve
network
bandwidth
both
neck
for
immediately
loading
read
and
reduce.
B
Reading
latency
and
the
protect
image
registry
to
avoid
huge
traffic,
this
figure
describe
decrypts
the
process,
not
just
use
dragonfly
to
deliver
immediate
container,
instead
of
making
request
directly
to
immediate
registry
nana's,
just
proxy
all
requests
using
dragonfly
demon
so
that
a
significant
portion
of
requests
are
cached
in
dragon
by
dragonfly.
It
helps
to
reduce
reading
requests
to
image
graduate
and
reduce
traffic
to
the
immediate
registry.
B
A
B
We
will
release
the
sydney
advanced
future
and
integrate
united
in
july.
We
will
support
ssl
and
encrypt
transmission
and
integrate
multi-cloud
storage.
In
august
we
will
release
account
account
authority
and
the
dashboard
tracing
surprise.
In
september
we
will
release
pressure,
sinker,
remote
copy
and
the
file
uploader
in
october.
We
will
release
multi,
tenants
and
transfer
transmission
cross
cloud.