<s>
The	O
Write	B-Application
Anywhere	I-Application
File	I-Application
Layout	I-Application
(	O
WAFL	O
)	O
is	O
a	O
proprietary	O
file	B-Application
system	I-Application
that	O
supports	O
large	O
,	O
high-performance	O
RAID	B-Architecture
arrays	I-Architecture
,	O
quick	O
restarts	O
without	O
lengthy	O
consistency	O
checks	O
in	O
the	O
event	O
of	O
a	O
crash	B-General_Concept
or	O
power	O
failure	O
,	O
and	O
growing	O
the	O
filesystems	B-Application
size	O
quickly	O
.	O
</s>
<s>
It	O
was	O
designed	O
by	O
NetApp	O
for	O
use	O
in	O
its	O
storage	O
appliances	O
like	O
NetApp	B-Operating_System
FAS	I-Operating_System
,	I-Operating_System
AFF	I-Operating_System
,	O
Cloud	O
Volumes	O
ONTAP	B-Operating_System
and	O
ONTAP	B-Operating_System
Select	O
.	O
</s>
<s>
Its	O
author	O
claims	O
that	O
WAFL	O
is	O
not	O
a	O
file	B-Application
system	I-Application
,	O
although	O
it	O
includes	O
one	O
.	O
</s>
<s>
It	O
tracks	O
changes	O
similarly	O
to	O
journaling	B-Application
file	I-Application
systems	I-Application
as	O
logs	O
(	O
known	O
as	O
NVLOGs	O
)	O
in	O
dedicated	O
memory	O
storage	O
device	O
non-volatile	B-General_Concept
random	I-General_Concept
access	I-General_Concept
memory	I-General_Concept
,	O
referred	O
to	O
as	O
NVRAM	B-General_Concept
or	O
NVMEM	O
.	O
</s>
<s>
WAFL	O
provides	O
mechanisms	O
that	O
enable	O
a	O
variety	O
of	O
file	B-Application
systems	I-Application
and	O
technologies	O
that	O
want	O
to	O
access	O
disk	B-General_Concept
blocks	I-General_Concept
.	O
</s>
<s>
WAFL	O
stores	O
metadata	O
,	O
as	O
well	O
as	O
data	O
,	O
in	O
files	O
;	O
metadata	O
,	O
such	O
as	O
inodes	B-Application
and	O
block	O
maps	O
indicating	O
which	O
blocks	O
in	O
the	O
volume	O
are	O
allocated	O
,	O
are	O
not	O
stored	O
in	O
fixed	O
locations	O
in	O
the	O
file	B-Application
system	I-Application
.	O
</s>
<s>
The	O
top-level	O
file	O
in	O
a	O
volume	O
is	O
the	O
inode	B-Application
file	O
,	O
which	O
contains	O
the	O
inodes	B-Application
for	O
all	O
other	O
files	O
;	O
the	O
inode	B-Application
for	O
the	O
inode	B-Application
file	O
itself	O
,	O
called	O
the	O
root	O
inode	B-Application
,	O
is	O
stored	O
in	O
a	O
block	O
with	O
a	O
fixed	O
location	O
.	O
</s>
<s>
An	O
inode	B-Application
for	O
a	O
sufficiently	O
small	O
file	O
contains	O
the	O
file	O
's	O
contents	O
;	O
otherwise	O
,	O
it	O
contains	O
a	O
list	O
of	O
pointers	O
to	O
file	O
data	O
blocks	O
or	O
a	O
list	O
of	O
pointers	O
to	O
indirect	O
blocks	O
containing	O
lists	O
of	O
pointers	O
to	O
file	O
data	O
blocks	O
,	O
and	O
so	O
forth	O
,	O
with	O
as	O
many	O
layers	O
of	O
indirect	O
blocks	O
as	O
are	O
necessary	O
,	O
forming	O
a	O
tree	O
of	O
blocks	O
.	O
</s>
<s>
All	O
data	O
and	O
metadata	O
blocks	O
in	O
the	O
file	B-Application
system	I-Application
,	O
other	O
than	O
the	O
block	O
containing	O
the	O
root	O
inode	B-Application
,	O
are	O
stored	O
in	O
files	O
in	O
the	O
file	B-Application
system	I-Application
.	O
</s>
<s>
The	O
root	O
inode	B-Application
can	O
thus	O
be	O
used	O
to	O
locate	O
all	O
of	O
the	O
blocks	O
of	O
all	O
files	O
other	O
than	O
the	O
inode	B-Application
file	O
.	O
</s>
<s>
Main	O
memory	O
is	O
used	O
as	O
a	O
page	B-General_Concept
cache	I-General_Concept
for	O
blocks	O
from	O
files	O
.	O
</s>
<s>
When	O
a	O
change	O
is	O
made	O
to	O
a	O
block	O
of	O
a	O
file	O
,	O
the	O
copy	O
in	O
the	O
page	B-General_Concept
cache	I-General_Concept
is	O
updated	O
and	O
marked	O
dirty	O
,	O
and	O
the	O
difference	O
is	O
logged	O
in	O
non-volatile	O
memory	O
in	O
a	O
log	O
called	O
the	O
NVLOG	O
.	O
</s>
<s>
If	O
the	O
dirty	O
block	O
in	O
the	O
page	B-General_Concept
cache	I-General_Concept
is	O
to	O
be	O
written	O
to	O
permanent	O
storage	O
,	O
it	O
is	O
not	O
rewritten	O
to	O
the	O
block	O
from	O
which	O
it	O
was	O
read	O
;	O
instead	O
,	O
a	O
new	O
block	O
is	O
allocated	O
on	O
permanent	O
storage	O
,	O
the	O
contents	O
of	O
the	O
block	O
are	O
written	O
to	O
the	O
new	O
location	O
,	O
and	O
the	O
inode	B-Application
or	O
indirect	O
block	O
that	O
pointed	O
to	O
the	O
block	O
in	O
question	O
is	O
updated	O
in	O
main	O
memory	O
.	O
</s>
<s>
If	O
the	O
block	O
containing	O
the	O
inode	B-Application
,	O
or	O
the	O
indirect	O
block	O
,	O
is	O
to	O
be	O
written	O
to	O
permanent	O
storage	O
,	O
it	O
is	O
also	O
written	O
to	O
a	O
new	O
location	O
,	O
rather	O
than	O
being	O
overwritten	O
at	O
its	O
previous	O
position	O
.	O
</s>
<s>
This	O
is	O
what	O
the	O
"	O
Write	O
Anywhere	O
"	O
in	O
"	O
Write	B-Application
Anywhere	I-Application
File	I-Application
Layout	I-Application
"	O
refers	O
to	O
.	O
</s>
<s>
As	O
all	O
blocks	O
,	O
other	O
than	O
the	O
block	O
containing	O
the	O
root	O
inode	B-Application
,	O
are	O
found	O
via	O
the	O
root	O
inode	B-Application
,	O
none	O
of	O
the	O
changes	O
written	O
to	O
permanent	O
storage	O
are	O
visible	O
on	O
permanent	O
storage	O
until	O
the	O
root	O
inode	B-Application
is	O
updated	O
.	O
</s>
<s>
The	O
root	O
inode	B-Application
is	O
updated	O
by	O
a	O
process	O
called	O
a	O
consistency	O
point	O
,	O
in	O
which	O
all	O
dirty	O
blocks	O
not	O
yet	O
written	O
to	O
permanent	O
storage	O
are	O
written	O
to	O
permanent	O
storage	O
,	O
and	O
a	O
new	O
root	O
inode	B-Application
is	O
written	O
out	O
,	O
pointing	O
to	O
the	O
blocks	O
in	O
the	O
new	O
version	O
of	O
the	O
inode	B-Application
file	O
.	O
</s>
<s>
At	O
that	O
point	O
,	O
all	O
of	O
the	O
changes	O
to	O
the	O
file	B-Application
system	I-Application
are	O
visible	O
on	O
permanent	O
storage	O
,	O
using	O
the	O
new	O
root	O
inode	B-Application
.	O
</s>
<s>
If	O
the	O
server	O
crashes	B-General_Concept
before	O
all	O
changes	O
to	O
a	O
file	B-Application
system	I-Application
have	O
been	O
made	O
visible	O
in	O
a	O
consistency	O
point	O
,	O
the	O
changes	O
that	O
have	O
not	O
been	O
made	O
visible	O
are	O
still	O
in	O
the	O
NVLOG	O
;	O
when	O
the	O
server	O
reboots	O
,	O
it	O
replays	O
all	O
entries	O
in	O
the	O
NVLOG	O
,	O
again	O
making	O
the	O
changes	O
recorded	O
in	O
the	O
NVLOG	O
,	O
so	O
that	O
they	O
will	O
not	O
be	O
lost	O
.	O
</s>
<s>
Instead	O
it	O
automatically	O
places	O
data	O
using	O
temporal	B-General_Concept
locality	I-General_Concept
to	O
write	O
metadata	O
alongside	O
user	O
data	O
in	O
a	O
way	O
designed	O
to	O
minimize	O
the	O
number	O
of	O
disk	O
operations	O
required	O
to	O
commit	O
data	O
to	O
stable	O
disk	O
storage	O
using	O
single	O
and	O
dual	O
parity	O
based	O
RAID	B-Architecture
.	O
</s>
<s>
Using	O
a	O
data	O
placement	O
based	O
on	O
temporal	B-General_Concept
locality	I-General_Concept
of	O
reference	O
can	O
improve	O
the	O
performance	O
of	O
reading	O
datasets	O
which	O
are	O
read	O
in	O
a	O
similar	O
way	O
to	O
the	O
way	O
they	O
were	O
written	O
(	O
e.g.	O
</s>
<s>
a	O
database	O
record	O
and	O
its	O
associated	O
index	O
entry	O
)	O
,	O
however	O
it	O
can	O
also	O
cause	O
fragmentation	B-Architecture
from	O
the	O
perspective	O
of	O
spatial	O
locality	B-General_Concept
of	I-General_Concept
reference	I-General_Concept
.	O
</s>
<s>
On	O
spinning	O
HDDs	B-Device
this	O
does	O
not	O
adversely	O
affect	O
files	O
that	O
are	O
sequentially	O
written	O
,	O
randomly	O
read	O
,	O
or	O
are	O
subsequently	O
read	O
using	O
the	O
same	O
temporal	O
pattern	O
,	O
but	O
does	O
affect	O
sequential	O
read	O
after	O
random	O
write	O
spatial	O
data	O
access	O
patterns	O
because	O
of	O
magnetic	O
head	O
could	O
be	O
only	O
in	O
one	O
position	O
at	O
a	O
time	O
to	O
read	O
data	O
from	O
platter	O
while	O
fragmentation	B-Architecture
does	O
no	O
effect	O
on	O
SSD	B-Device
drives	I-Device
.	O
</s>
<s>
Releases	O
of	O
ONTAP	B-Operating_System
since	O
7.3.1	O
have	O
included	O
a	O
number	O
of	O
techniques	O
to	O
optimize	O
spatial	O
data	O
layout	O
such	O
as	O
the	O
reallocate	O
command	O
to	O
perform	O
scheduled	O
and	O
manual	O
defragmentation	B-Application
,	O
and	O
the	O
Write	O
after	O
Reading	O
volume	O
option	O
which	O
detects	O
and	O
automatically	O
corrects	O
suboptimal	O
data	O
access	O
patterns	O
caused	O
by	O
spatial	O
fragmentation	B-Architecture
.	O
</s>
<s>
Releases	O
of	O
ONTAP	B-Operating_System
8.1.1	O
include	O
other	O
techniques	O
to	O
automatically	O
optimize	O
contiguous	O
free-space	O
within	O
the	O
filesystem	B-Application
which	O
also	O
helps	O
to	O
maintain	O
optimal	O
data	O
layouts	O
for	O
most	O
data	O
access	O
patterns	O
.	O
</s>
<s>
Releases	O
of	O
ONTAP	B-Operating_System
since	O
9.1	O
have	O
included	O
a	O
number	O
of	O
techniques	O
to	O
optimize	O
SSD	B-Device
usage	O
such	O
as	O
Inline	O
Data	O
Compaction	O
(	O
in	O
9.1	O
)	O
,	O
starting	O
with	O
ONTAP	B-Operating_System
9.2	O
FabricPool	O
functionality	O
for	O
automatic	O
tiering	O
of	O
cold	O
data	O
to	O
slow	O
S3	O
storage	O
and	O
back	O
if	O
needed	O
for	O
SSD	B-Device
aggregates	O
,	O
and	O
Cross	O
Volume	O
Deduplication	O
within	O
an	O
aggregate	O
with	O
maximum	O
of	O
800TiB	O
for	O
each	O
aggregate	O
.	O
</s>
<s>
WAFL	O
supports	O
snapshots	B-Application
,	O
which	O
are	O
read-only	O
copies	O
of	O
a	O
file	B-Application
system	I-Application
.	O
</s>
<s>
Snapshots	B-Application
are	O
created	O
by	O
performing	O
the	O
same	O
operations	O
that	O
are	O
performed	O
in	O
a	O
consistency	O
point	O
,	O
but	O
,	O
instead	O
of	O
updating	O
the	O
root	O
inode	B-Application
corresponding	O
to	O
the	O
current	O
state	O
of	O
the	O
file	B-Application
system	I-Application
,	O
saving	O
a	O
copy	O
of	O
the	O
root	O
inode	B-Application
.	O
</s>
<s>
As	O
all	O
data	O
and	O
metadata	O
in	O
a	O
file	B-Application
system	I-Application
can	O
be	O
found	O
from	O
the	O
root	O
inode	B-Application
,	O
all	O
data	O
and	O
metadata	O
in	O
a	O
file	B-Application
system	I-Application
,	O
as	O
of	O
the	O
time	O
when	O
the	O
snapshot	B-Application
is	O
created	O
,	O
can	O
be	O
found	O
from	O
the	O
snapshot	B-Application
's	O
copy	O
of	O
the	O
root	O
inode	B-Application
.	O
</s>
<s>
No	O
other	O
data	O
needs	O
to	O
be	O
copied	O
to	O
create	O
a	O
snapshot	B-Application
.	O
</s>
<s>
An	O
entry	O
in	O
the	O
block	O
map	O
contains	O
a	O
bit	O
indicating	O
whether	O
the	O
block	O
is	O
in	O
use	O
in	O
the	O
current	O
version	O
of	O
the	O
file	B-Application
system	I-Application
and	O
several	O
bits	O
,	O
one	O
per	O
snapshot	B-Application
,	O
indicating	O
whether	O
the	O
block	O
is	O
in	O
use	O
in	O
the	O
snapshot	B-Application
.	O
</s>
<s>
This	O
ensures	O
that	O
data	O
in	O
a	O
snapshot	B-Application
is	O
not	O
overwritten	O
until	O
the	O
snapshot	B-Application
is	O
deleted	O
.	O
</s>
<s>
ROW	O
is	O
much	O
faster	O
on	O
rewrite	O
operations	O
compare	O
to	O
Copy-on-write	B-Language
where	O
old	O
data	O
block	O
going	O
to	O
be	O
rewritten	O
in-place	O
and	O
captured	O
in	O
a	O
snapshot	B-Application
needs	O
to	O
be	O
copied	O
first	O
to	O
space	O
allocated	O
for	O
snapshot	B-Application
reserve	O
in	O
order	O
to	O
preserve	O
original	O
data	O
,	O
this	O
generates	O
additional	O
data	O
copy	O
operations	O
once	O
system	O
get	O
rewrites	O
to	O
that	O
block	O
.	O
</s>
<s>
Snapshots	B-Application
provide	O
online	O
backups	O
that	O
can	O
be	O
accessed	O
quickly	O
,	O
through	O
special	O
hidden	O
directories	O
in	O
the	O
file	B-Application
system	I-Application
,	O
allowing	O
users	O
to	O
recover	O
files	O
that	O
have	O
been	O
accidentally	O
deleted	O
or	O
modified	O
.	O
</s>
<s>
NetApp	O
's	O
Data	B-Operating_System
ONTAP	I-Operating_System
Release	O
7G	O
operating	O
system	O
supports	O
a	O
read-write	O
snapshot	B-Application
called	O
FlexClone	O
.	O
</s>
<s>
Snapshots	B-Application
are	O
basis	O
for	O
technologies	O
like	O
SnapMirror	O
,	O
SnapVault	O
and	O
Online	O
Volume	O
Move	O
while	O
features	O
like	O
FlexClone	O
,	O
SnapLock	O
,	O
SnapRestore	O
are	O
snapshot-like	O
technologies	O
leverage	O
on	O
WAFL	O
capabilities	O
and	O
properties	O
like	O
manipulations	O
with	O
inodes	B-Application
.	O
</s>
<s>
Starting	O
with	O
ONTAP	B-Operating_System
9.4	O
maximum	O
number	O
of	O
snapshots	B-Application
supported	O
for	O
each	O
FlexVol	O
is	O
1024	O
,	O
while	O
for	O
previous	O
versions	O
max	O
limit	O
was	O
255	O
.	O
</s>
<s>
Starting	O
with	O
ONTAP	B-Operating_System
9.5	O
snapshot	B-Application
sharing	O
functionality	O
were	O
added	O
to	O
run	O
deduplication	O
scan	O
across	O
Active	O
file	B-Application
system	I-Application
and	O
snapshots	B-Application
,	O
and	O
deduplication	O
savings	O
is	O
a	O
magnitude	O
of	O
number	O
of	O
snapshots	B-Application
.	O
</s>
<s>
Before	O
9.5	O
not	O
deduplicated	O
data	O
locked	O
in	O
a	O
snapshot	B-Application
could	O
n’t	O
be	O
used	O
by	O
deduplication	O
process	O
and	O
runs	O
only	O
in	O
active	O
file	B-Application
system	I-Application
.	O
</s>
<s>
An	O
important	O
feature	O
of	O
WAFL	O
is	O
its	O
support	O
for	O
both	O
a	O
Unix-style	O
file	O
and	O
directory	O
model	O
for	O
NFS	B-Protocol
clients	O
and	O
a	O
Microsoft	O
Windows-style	O
file	O
and	O
directory	O
model	O
for	O
SMB	B-Operating_System
clients	O
.	O
</s>
<s>
Unix	B-Application
can	O
use	O
either	O
access	O
control	O
lists	O
(	O
ACL	O
)	O
or	O
a	O
simple	O
bitmask	O
,	O
whereas	O
the	O
more	O
recent	O
Windows	O
model	O
is	O
based	O
on	O
access	O
control	O
lists	O
.	O
</s>
<s>
These	O
two	O
features	O
make	O
it	O
possible	O
to	O
write	O
a	O
file	O
to	O
an	O
SMB	B-Operating_System
type	O
of	O
networked	O
filesystem	B-Application
and	O
access	O
it	O
later	O
via	O
NFS	B-Protocol
from	O
a	O
Unix	B-Application
workstation	O
.	O
</s>
<s>
Alongside	O
ordinary	O
files	O
,	O
WAFL	O
can	O
contain	O
file-containers	O
called	O
LUNs	O
with	O
required	O
special	O
attributes	O
like	O
LUN	O
serial	O
number	O
for	O
block	O
devices	O
,	O
which	O
could	O
be	O
accessed	O
using	O
SAN	O
protocols	O
running	O
on	O
ONTAP	B-Operating_System
OS	O
software	O
.	O
</s>
<s>
Each	O
Flexible	O
Volume	O
(	O
FlexVol	O
)	O
is	O
a	O
separate	O
WAFL	O
file	B-Application
system	I-Application
,	O
located	O
on	O
an	O
aggregate	O
and	O
distributed	O
across	O
all	O
disks	O
in	O
the	O
aggregate	O
.	O
</s>
<s>
ONTAP	B-Operating_System
during	O
data	O
optimization	O
process	O
including	O
the	O
"	O
Tetris	O
"	O
which	O
finishes	O
with	O
Consistency	O
Points	O
(	O
see	O
NVRAM	B-General_Concept
)	O
is	O
programmed	O
to	O
evenly	O
distribute	O
data	O
blocks	O
as	O
much	O
as	O
possible	O
in	O
each	O
FlexVol	O
volume	O
across	O
all	O
disks	O
in	O
aggregate	O
so	O
each	O
FlexVol	O
could	O
potentially	O
use	O
all	O
available	O
performance	O
of	O
all	O
the	O
data	O
disks	O
in	O
the	O
aggregate	O
.	O
</s>
<s>
With	O
the	O
approach	O
of	O
even	O
data	O
block	O
distribution	O
across	O
all	O
the	O
data	O
disks	O
in	O
an	O
aggregate	O
,	O
performance	O
throttling	O
for	O
a	O
FlexVol	O
could	O
be	O
done	O
dynamically	O
with	O
storage	O
QoS	O
and	O
does	O
not	O
require	O
dedicated	O
aggregates	O
or	O
RAID	B-Architecture
groups	O
for	O
each	O
FlexVol	O
to	O
guarantee	O
performance	O
and	O
provide	O
the	O
unused	O
performance	O
to	O
a	O
FlexVol	O
volume	O
which	O
requires	O
it	O
.	O
</s>
<s>
Each	O
FlexVol	O
could	O
be	O
configured	O
as	O
thick	O
or	O
thin	B-General_Concept
provisioned	I-General_Concept
space	O
and	O
later	O
could	O
be	O
changed	O
on	O
the	O
fly	O
any	O
time	O
.	O
</s>
<s>
Block	O
device	O
access	O
with	O
storage	O
area	O
network	O
(	O
SAN	O
)	O
protocols	O
such	O
as	O
iSCSI	B-Protocol
,	O
Fibre	B-Architecture
Channel	I-Architecture
(	O
FC	O
)	O
,	O
and	O
Fibre	B-Protocol
Channel	I-Protocol
over	I-Protocol
Ethernet	I-Protocol
(	O
FCoE	B-Protocol
)	O
is	O
done	O
with	O
LUN	O
emulation	O
similar	O
to	O
Loop	B-Application
device	I-Application
technique	O
on	O
top	O
of	O
a	O
FlexVol	O
volume	O
;	O
thus	O
each	O
LUN	O
on	O
WAFL	O
file	B-Application
system	I-Application
appears	O
as	O
a	O
file	O
,	O
yet	O
have	O
additional	O
properties	O
required	O
for	O
block	O
devices	O
.	O
</s>
<s>
LUNs	O
can	O
also	O
be	O
configured	O
as	O
thick	O
or	O
thin	B-General_Concept
provisioned	I-General_Concept
and	O
can	O
be	O
changed	O
later	O
on	O
the	O
fly	O
.	O
</s>
<s>
Even	O
though	O
LUN	O
size	O
with	O
data	O
on	O
it	O
could	O
be	O
decreased	O
on	O
WAFL	O
file	B-Application
system	I-Application
,	O
ONTAP	B-Operating_System
has	O
no	O
knowledge	O
about	O
upper-level	O
block	O
structure	O
due	O
to	O
SAN	O
architecture	O
so	O
it	O
could	O
truncate	O
data	O
and	O
damage	O
the	O
file	B-Application
system	I-Application
on	O
that	O
LUN	O
,	O
so	O
the	O
host	O
needs	O
to	O
migrate	O
the	O
blocks	O
containing	O
the	O
data	O
into	O
a	O
new	O
LUN	O
boundary	O
to	O
prevent	O
data	O
loss	O
.	O
</s>
<s>
Similar	O
to	O
RAID	B-Architecture
1	O
,	O
plexes	O
in	O
ONTAP	B-Operating_System
systems	O
can	O
keep	O
mirrored	O
data	O
in	O
two	O
places	O
,	O
but	O
while	O
conventional	O
RAID-1	O
must	O
exist	O
within	O
the	O
bounds	O
of	O
one	O
storage	O
system	O
,	O
two	O
plexes	O
could	O
be	O
distributed	O
between	O
two	O
storage	O
systems	O
.	O
</s>
<s>
On	O
the	O
other	O
hand	O
,	O
each	O
plex	O
includes	O
underlying	O
storage	O
space	O
from	O
one	O
or	O
more	O
NetApp	O
RAID	B-Architecture
groups	O
or	O
LUNs	O
from	O
third-party	O
storage	O
systems	O
(	O
see	O
FlexArray	O
)	O
in	O
a	O
single	O
plex	O
similarly	O
to	O
RAID	B-Architecture
0	O
.	O
</s>
<s>
If	O
an	O
aggregate	O
consists	O
of	O
two	O
plexes	O
,	O
one	O
plex	O
is	O
considered	O
a	O
master	O
and	O
second	O
as	O
a	O
slave	O
;	O
slaves	O
must	O
consist	O
of	O
exactly	O
the	O
same	O
RAID	B-Architecture
configuration	O
and	O
drives	O
.	O
</s>
<s>
For	O
example	O
,	O
if	O
we	O
have	O
an	O
aggregate	O
consisting	O
of	O
two	O
plexes	O
where	O
the	O
master	O
plex	O
consists	O
of	O
21	O
data	O
and	O
3	O
1.8	O
TB	O
SAS	O
parity	O
drives	O
in	O
RAID-TEC	O
,	O
then	O
the	O
slave	O
plex	O
must	O
consist	O
of	O
21	O
data	O
and	O
3	O
1.8	O
TB	O
SAS	O
parity	O
drives	O
in	O
RAID-TEC	O
.	O
</s>
<s>
The	O
second	O
example	O
,	O
if	O
we	O
have	O
an	O
aggregate	O
consisted	O
of	O
two	O
plexes	O
where	O
master	O
plex	O
consists	O
of	O
one	O
RAID	B-Architecture
17	O
data	O
and	O
3	O
parity	O
SAS	O
drives	O
1.8	O
TB	O
configured	O
as	O
RAID-TEC	O
and	O
second	O
RAID	B-Architecture
in	O
the	O
master	O
plex	O
is	O
RAID-DP	O
with	O
2	O
data	O
and	O
2	O
parity	O
SSD	B-Device
960	O
GB	O
.	O
</s>
<s>
The	O
second	O
plex	O
must	O
have	O
the	O
same	O
configuration	O
:	O
one	O
RAID	B-Architecture
17	O
data	O
and	O
3	O
parity	O
SAS	O
drives	O
1.8	O
TB	O
configured	O
as	O
RAID-TEC	O
,	O
and	O
the	O
second	O
RAID	B-Architecture
in	O
the	O
slave	O
plex	O
is	O
RAID-DP	O
with	O
2	O
data	O
and	O
2	O
parity	O
SSD	B-Device
960	O
GB	O
.	O
</s>
<s>
Like	O
many	O
competitors	O
,	O
NetApp	O
ONTAP	B-Operating_System
systems	O
utilizing	O
memory	O
as	O
a	O
much	O
faster	O
storage	O
medium	O
for	O
accepting	O
and	O
caching	O
data	O
from	O
hosts	O
and	O
,	O
most	O
importantly	O
,	O
for	O
data	O
optimization	O
before	O
writes	O
which	O
greatly	O
improves	O
the	O
performance	O
of	O
such	O
storage	O
systems	O
.	O
</s>
<s>
While	O
competitors	O
widely	O
using	O
non-volatile	B-General_Concept
random-access	I-General_Concept
memory	I-General_Concept
(	O
NVRAM	B-General_Concept
)	O
to	O
preserve	O
data	O
in	O
it	O
during	O
unexpected	O
events	O
like	O
a	O
reboot	O
for	O
both	O
write	O
caching	O
and	O
data	O
optimization	O
,	O
NetApp	O
ONTAP	B-Operating_System
systems	O
using	O
ordinary	O
random-access	B-Architecture
memory	I-Architecture
(	O
RAM	B-Architecture
)	O
for	O
data	O
optimization	O
and	O
dedicated	O
NVRAM	B-General_Concept
or	O
NVDIMM	O
for	O
logging	O
of	O
initial	O
data	O
in	O
an	O
unchanged	O
state	O
as	O
they	O
came	O
from	O
hosts	O
similarly	O
as	O
transaction	O
logging	O
done	O
in	O
Relational	B-Application
databases	I-Application
.	O
</s>
<s>
So	O
in	O
case	O
of	O
disaster	O
,	O
naturally	O
,	O
RAM	B-Architecture
will	O
be	O
automatically	O
cleared	O
after	O
reboot	O
,	O
and	O
data	O
stored	O
in	O
non-volatile	O
memory	O
in	O
the	O
form	O
of	O
logs	O
called	O
NVLOGs	O
will	O
survive	O
after	O
reboot	O
and	O
will	O
be	O
used	O
for	O
restore	O
consistency	O
.	O
</s>
<s>
All	O
changes	O
and	O
optimizations	O
in	O
ONTAP	B-Operating_System
systems	O
done	O
only	O
in	O
RAM	B-Architecture
,	O
which	O
helps	O
to	O
reduce	O
the	O
size	O
of	O
non-volatile	O
memory	O
for	O
ONTAP	B-Operating_System
systems	O
.	O
</s>
<s>
After	O
optimizations	O
data	O
from	O
hosts	O
structured	O
in	O
Tetris-like	O
manner	O
,	O
optimized	O
and	O
prepared	O
with	O
passing	O
few	O
stages	O
(	O
i.e.	O
,	O
WAFL	O
and	O
RAID	B-Architecture
)	O
to	O
be	O
written	O
in	O
underlying	O
disks	O
in	O
RAID	B-Architecture
groups	O
on	O
the	O
aggregate	O
where	O
data	O
are	O
going	O
to	O
be	O
stored	O
.	O
</s>
<s>
Data	O
written	O
to	O
aggregates	O
will	O
contain	O
necessary	O
WAFL	O
metadata	O
and	O
RAID	B-Architecture
parity	O
so	O
no	O
additional	O
read	O
from	O
data	O
disks	O
,	O
calculate	O
and	O
write	O
to	O
parity	O
disks	O
operations	O
will	O
occur	O
as	O
with	O
traditional	O
RAID-6	O
and	O
RAID-4	O
groups	O
.	O
</s>
<s>
CP	O
at	O
first	O
creating	O
system	O
snapshot	B-Application
on	O
an	O
aggregate	O
where	O
data	O
are	O
going	O
to	O
be	O
written	O
,	O
then	O
optimized	O
and	O
prepared	O
data	O
from	O
RAM	B-Architecture
written	O
sequentially	O
as	O
a	O
single	O
transaction	O
to	O
the	O
aggregate	O
,	O
if	O
it	O
fails	O
,	O
the	O
whole	O
transaction	O
fails	O
in	O
case	O
of	O
a	O
sudden	O
reboot	O
which	O
allows	O
WAFL	O
file	B-Application
system	I-Application
always	O
to	O
be	O
consistent	O
.	O
</s>
<s>
In	O
case	O
of	O
successful	O
CP	O
transaction	O
new	O
active	O
file	B-Application
system	I-Application
point	O
is	O
propagated	O
and	O
corresponding	O
NVLOGs	O
cleared	O
.	O
</s>
<s>
Starting	O
with	O
the	O
All-Flash	O
FAS	O
A800	O
system	O
,	O
NetApp	O
replaced	O
the	O
NVRAM	B-General_Concept
PCI	O
module	O
with	O
NVDIMMs	O
connected	O
to	O
the	O
memory	O
bus	O
,	O
increasing	O
the	O
performance	O
.	O
</s>
