►
Description
Practices of Ceph Object Storage in Public Cloud Services - Yu Liyang, China Mobile
We talk about the practices of China mobile on ceph object storage, the Multi-site datacenter architecture across 3 cities, the Billing System, new Features(bucket notification, request callback, object soft link) and The difficulties we encountered.
About Yu Liyang
Software Engineer, China Mobile
yuliyang is a Object Storage software engineer, work for China Mobile since 2016.
A
A
In
the
cloud
and
the
cloud
storage,
you
must
have
received
features,
users
will
come,
compare
your
products
and
with
other
product,
and
they
were
choose
the
one
which
is
the
requirements
so
based
on
the
staff.
Well,
we
developed
some
features,
some
of
them
in
the
s3
feature,
which
is
not
implemented
in
safe
now,
like
packet
and
notification,
packet
logging
and
the
storage
class
storage
classes
implemented
in
the
master
branch.
A
Now
this
is
used
to
migrate
did
her
between
different
reports,
so
we
can
make
a
poor
with
a
rabbit.
Kicker
and
poor
has
used
a
ratio
poor.
So
we
can
you
look
start
class
to
migrated
data
between
two
ports
and
a
part
of
s3
feature.
We
also
develop
some
customized,
the
feature
which
is
required
by
others,
or
some
projects
like
request,
a
callback
and
pocketed
origin
at
MP
upload.
The
support
and
the
symbol
link.
A
A
The
target
resource
is
limited,
so
we
are,
the
resources
is
consumed,
exhaust
version.
We
must
expanded
the
cluster.
Well,
why?
No
one
solution
to
add
a
new
port
to
the
cluster,
so
the
data
rebalancing
we
ought
to
not
a
worker
in
the
force,
but
at
a
new
poll
we
needed
users
to
changes
your
package
name
and
in
public
cloud
and
the
user
wanna
change
the
package
names,
so
we
can't
add
a
new
port
with
the
pastor,
so
we
added
servers
to
the
existing
force.
A
You
really
need
some
more
development
and
we
open
especially,
are
actually
W,
so
we
open
started
I'd
never
do
when
we
updated
the
cloud
clusters,
but
some
modifications
on
our
W
is
to
be
done
in
see
us
at
the
W
modular.
So
when
we
modify
the
Cirrus
modular,
we
must
stack
all
our
STIs
and
in
our
cast
cast
her.
A
The
Oh
STIs
number
is
really
big,
so
we
start
off
with
OTS
is
a
very
time-consuming
work
and,
as
the
last
is
the
pressure
on
the
master
song
of
some
group,
this
is
this
can
be
happening
in
our
production
environment
and
we
can
see
over
ten
thousand
of
concurrent
requests
or
the
Masters
on
group
and
Izzy.
Rtw
is
really
really
slow
and
modification
arm
metadata
is
the
latency
is
really
big.
So
thanks.