<s>
RAID	B-Architecture
(	O
;	O
"	O
redundant	B-Architecture
array	I-Architecture
of	I-Architecture
inexpensive	I-Architecture
disks	I-Architecture
"	O
or	O
"	O
redundant	B-Architecture
array	I-Architecture
of	I-Architecture
independent	I-Architecture
disks	I-Architecture
"	O
)	O
is	O
a	O
data	O
storage	B-Application
virtualization	I-Application
technology	O
that	O
combines	O
multiple	O
physical	O
disk	B-Device
drive	I-Device
components	O
into	O
one	O
or	O
more	O
logical	O
units	O
for	O
the	O
purposes	O
of	O
data	B-Application
redundancy	I-Application
,	O
performance	O
improvement	O
,	O
or	O
both	O
.	O
</s>
<s>
This	O
is	O
in	O
contrast	O
to	O
the	O
previous	O
concept	O
of	O
highly	O
reliable	O
mainframe	B-Architecture
disk	B-Device
drives	I-Device
referred	O
to	O
as	O
"	O
single	O
large	O
expensive	O
disk	B-Device
"	O
(	O
SLED	O
)	O
.	O
</s>
<s>
Data	O
is	O
distributed	O
across	O
the	O
drives	O
in	O
one	O
of	O
several	O
ways	O
,	O
referred	O
to	O
as	O
RAID	B-Architecture
levels	O
,	O
depending	O
on	O
the	O
required	O
level	O
of	O
redundancy	B-Application
and	O
performance	O
.	O
</s>
<s>
The	O
different	O
schemes	O
,	O
or	O
data	O
distribution	O
layouts	O
,	O
are	O
named	O
by	O
the	O
word	O
"	O
RAID	B-Architecture
"	O
followed	O
by	O
a	O
number	O
,	O
for	O
example	O
RAID0	O
or	O
RAID1	O
.	O
</s>
<s>
Each	O
scheme	O
,	O
or	O
RAID	B-Architecture
level	O
,	O
provides	O
a	O
different	O
balance	O
among	O
the	O
key	O
goals	O
:	O
reliability	O
,	O
availability	O
,	O
performance	O
,	O
and	O
capacity	O
.	O
</s>
<s>
RAID	B-Architecture
levels	O
greater	O
than	O
RAID0	O
provide	O
protection	O
against	O
unrecoverable	O
sector	B-Application
read	O
errors	O
,	O
as	O
well	O
as	O
against	O
failures	O
of	O
whole	O
physical	O
drives	O
.	O
</s>
<s>
The	O
term	O
"	O
RAID	B-Architecture
"	O
was	O
invented	O
by	O
David	O
Patterson	O
,	O
Garth	O
A	O
.	O
Gibson	O
,	O
and	O
Randy	O
Katz	O
at	O
the	O
University	O
of	O
California	O
,	O
Berkeley	O
in	O
1987	O
.	O
</s>
<s>
In	O
their	O
June	O
1988	O
paper	O
"	O
A	O
Case	O
for	O
Redundant	B-Architecture
Arrays	I-Architecture
of	I-Architecture
Inexpensive	I-Architecture
Disks	I-Architecture
(	O
RAID	B-Architecture
)	O
"	O
,	O
presented	O
at	O
the	O
SIGMOD	O
Conference	O
,	O
they	O
argued	O
that	O
the	O
top-performing	O
mainframe	B-Architecture
disk	B-Device
drives	I-Device
of	O
the	O
time	O
could	O
be	O
beaten	O
on	O
performance	O
by	O
an	O
array	O
of	O
the	O
inexpensive	O
drives	O
that	O
had	O
been	O
developed	O
for	O
the	O
growing	O
personal	B-Device
computer	I-Device
market	O
.	O
</s>
<s>
Although	O
failures	O
would	O
rise	O
in	O
proportion	O
to	O
the	O
number	O
of	O
drives	O
,	O
by	O
configuring	O
for	O
redundancy	B-Application
,	O
the	O
reliability	O
of	O
an	O
array	O
could	O
far	O
exceed	O
that	O
of	O
any	O
large	O
single	O
drive	O
.	O
</s>
<s>
Although	O
not	O
yet	O
using	O
that	O
terminology	O
,	O
the	O
technologies	O
of	O
the	O
five	O
levels	O
of	O
RAID	B-Architecture
named	O
in	O
the	O
June	O
1988	O
paper	O
were	O
used	O
in	O
various	O
products	O
prior	O
to	O
the	O
paper	O
's	O
publication	O
,	O
including	O
the	O
following	O
:	O
</s>
<s>
Mirroring	B-General_Concept
(	O
RAID	B-Architecture
1	O
)	O
was	O
well	O
established	O
in	O
the	O
1970s	O
including	O
,	O
for	O
example	O
,	O
Tandem	O
NonStop	B-Architecture
Systems	I-Architecture
.	O
</s>
<s>
Around	O
1983	O
,	O
DEC	O
began	O
shipping	O
subsystem	O
mirrored	O
RA8X	O
disk	B-Device
drives	I-Device
(	O
now	O
known	O
as	O
RAID1	O
)	O
as	O
part	O
of	O
its	O
HSC50	O
subsystem	O
.	O
</s>
<s>
Around	O
1988	O
,	O
the	O
Thinking	O
Machines	O
 '	O
DataVault	B-Device
used	O
error	O
correction	O
codes	O
(	O
now	O
known	O
as	O
RAID2	O
)	O
in	O
an	O
array	O
of	O
disk	B-Device
drives	I-Device
.	O
</s>
<s>
Industry	O
manufacturers	O
later	O
redefined	O
the	O
RAID	B-Architecture
acronym	O
to	O
stand	O
for	O
"	O
redundant	B-Architecture
array	I-Architecture
of	I-Architecture
independent	I-Architecture
disks	I-Architecture
"	O
.	O
</s>
<s>
Many	O
RAID	B-Architecture
levels	O
employ	O
an	O
error	O
protection	O
scheme	O
called	O
"	O
parity	O
"	O
,	O
a	O
widely	O
used	O
method	O
in	O
information	O
technology	O
to	O
provide	O
fault	B-General_Concept
tolerance	I-General_Concept
in	O
a	O
given	O
set	O
of	O
data	O
.	O
</s>
<s>
Most	O
use	O
simple	O
XOR	O
,	O
but	O
RAID6	O
uses	O
two	O
separate	O
parities	O
based	O
respectively	O
on	O
addition	O
and	O
multiplication	O
in	O
a	O
particular	O
Galois	O
field	O
or	O
Reed	B-Error_Name
–	I-Error_Name
Solomon	I-Error_Name
error	I-Error_Name
correction	I-Error_Name
.	O
</s>
<s>
RAID	B-Architecture
can	O
also	O
provide	O
data	O
security	O
with	O
solid-state	B-Device
drives	I-Device
(	O
SSDs	B-Device
)	O
without	O
the	O
expense	O
of	O
an	O
all-SSD	O
system	O
.	O
</s>
<s>
For	O
example	O
,	O
a	O
fast	O
SSD	B-Device
can	O
be	O
mirrored	O
with	O
a	O
mechanical	O
drive	O
.	O
</s>
<s>
For	O
this	O
configuration	O
to	O
provide	O
a	O
significant	O
speed	O
advantage	O
,	O
an	O
appropriate	O
controller	O
is	O
needed	O
that	O
uses	O
the	O
fast	O
SSD	B-Device
for	O
all	O
read	O
operations	O
.	O
</s>
<s>
Adaptec	O
calls	O
this	O
"	O
hybrid	B-Architecture
RAID	I-Architecture
"	O
.	O
</s>
<s>
Originally	O
,	O
there	O
were	O
five	O
standard	O
levels	O
of	O
RAID	B-Architecture
,	O
but	O
many	O
variations	O
have	O
evolved	O
,	O
including	O
several	O
nested	B-General_Concept
levels	I-General_Concept
and	O
many	O
non-standard	B-General_Concept
levels	I-General_Concept
(	O
mostly	O
proprietary	B-Application
)	O
.	O
</s>
<s>
RAID	B-Architecture
levels	O
and	O
their	O
associated	O
data	O
formats	O
are	O
standardized	O
by	O
the	O
Storage	O
Networking	O
Industry	O
Association	O
(	O
SNIA	O
)	O
in	O
the	O
Common	O
RAID	B-Architecture
Disk	B-Device
Drive	I-Device
Format	O
(	O
DDF	O
)	O
standard	O
:	O
</s>
<s>
RAID	B-Architecture
0	O
consists	O
of	O
striping	B-General_Concept
,	O
but	O
no	O
mirroring	B-General_Concept
or	O
parity	O
.	O
</s>
<s>
But	O
because	O
striping	B-General_Concept
distributes	O
the	O
contents	O
of	O
each	O
file	O
among	O
all	O
drives	O
in	O
the	O
set	O
,	O
the	O
failure	O
of	O
any	O
drive	O
causes	O
the	O
entire	O
RAID0	O
volume	O
and	O
all	O
files	O
to	O
be	O
lost	O
.	O
</s>
<s>
The	O
benefit	O
of	O
RAID0	O
is	O
that	O
the	O
throughput	O
of	O
read	O
and	O
write	O
operations	O
to	O
any	O
file	O
is	O
multiplied	O
by	O
the	O
number	O
of	O
drives	O
because	O
,	O
unlike	O
spanned	O
volumes	O
,	O
reads	O
and	O
writes	O
are	O
done	O
concurrently	B-Operating_System
.	O
</s>
<s>
RAID	B-Architecture
1	O
consists	O
of	O
data	B-General_Concept
mirroring	I-General_Concept
,	O
without	O
parity	O
or	O
striping	B-General_Concept
.	O
</s>
<s>
Sustained	O
read	O
throughput	O
,	O
if	O
the	O
controller	O
or	O
software	B-Application
is	O
optimized	O
for	O
it	O
,	O
approaches	O
the	O
sum	O
of	O
throughputs	O
of	O
every	O
drive	O
in	O
the	O
set	O
,	O
just	O
as	O
for	O
RAID0	O
.	O
</s>
<s>
RAID	B-Architecture
2	O
consists	O
of	O
bit-level	O
striping	B-General_Concept
with	O
dedicated	O
Hamming-code	B-Error_Name
parity	O
.	O
</s>
<s>
All	O
disk	B-Device
spindle	O
rotation	O
is	O
synchronized	O
and	O
data	O
is	O
striped	B-General_Concept
such	O
that	O
each	O
sequential	O
bit	O
is	O
on	O
a	O
different	O
drive	O
.	O
</s>
<s>
Hamming-code	B-Error_Name
parity	O
is	O
calculated	O
across	O
corresponding	O
bits	O
and	O
stored	O
on	O
at	O
least	O
one	O
parity	O
drive	O
.	O
</s>
<s>
RAID	B-Architecture
3	O
consists	O
of	O
byte-level	O
striping	B-General_Concept
with	O
dedicated	O
parity	O
.	O
</s>
<s>
All	O
disk	B-Device
spindle	O
rotation	O
is	O
synchronized	O
and	O
data	O
is	O
striped	B-General_Concept
such	O
that	O
each	O
sequential	O
byte	B-Application
is	O
on	O
a	O
different	O
drive	O
.	O
</s>
<s>
Parity	O
is	O
calculated	O
across	O
corresponding	O
bytes	B-Application
and	O
stored	O
on	O
a	O
dedicated	O
parity	O
drive	O
.	O
</s>
<s>
RAID	B-Architecture
4	O
consists	O
of	O
block-level	O
striping	B-General_Concept
with	O
dedicated	O
parity	O
.	O
</s>
<s>
This	O
level	O
was	O
previously	O
used	O
by	O
NetApp	O
,	O
but	O
has	O
now	O
been	O
largely	O
replaced	O
by	O
a	O
proprietary	B-Application
implementation	O
of	O
RAID4	O
with	O
two	O
parity	O
disks	B-Device
,	O
called	O
RAID-DP	O
.	O
</s>
<s>
RAID	B-Architecture
5	O
consists	O
of	O
block-level	O
striping	B-General_Concept
with	O
distributed	O
parity	O
.	O
</s>
<s>
RAID5	O
requires	O
at	O
least	O
three	O
disks	B-Device
.	O
</s>
<s>
Rebuilding	O
an	O
array	O
requires	O
reading	O
all	O
data	O
from	O
all	O
disks	B-Device
,	O
opening	O
a	O
chance	O
for	O
a	O
second	O
drive	O
failure	O
and	O
the	O
loss	O
of	O
the	O
entire	O
array	O
.	O
</s>
<s>
RAID	B-Architecture
6	O
consists	O
of	O
block-level	O
striping	B-General_Concept
with	O
double	O
distributed	O
parity	O
.	O
</s>
<s>
Double	O
parity	O
provides	O
fault	B-General_Concept
tolerance	I-General_Concept
up	O
to	O
two	O
failed	O
drives	O
.	O
</s>
<s>
This	O
makes	O
larger	O
RAID	B-Architecture
groups	O
more	O
practical	O
,	O
especially	O
for	O
high-availability	O
systems	O
,	O
as	O
large-capacity	O
drives	O
take	O
longer	O
to	O
restore	O
.	O
</s>
<s>
RAID6	O
requires	O
a	O
minimum	O
of	O
four	O
disks	B-Device
.	O
</s>
<s>
In	O
what	O
was	O
originally	O
termed	O
hybrid	B-Architecture
RAID	I-Architecture
,	O
many	O
storage	O
controllers	O
allow	O
RAID	B-Architecture
levels	O
to	O
be	O
nested	O
.	O
</s>
<s>
The	O
elements	O
of	O
a	O
RAID	B-Architecture
may	O
be	O
either	O
individual	O
drives	O
or	O
arrays	O
themselves	O
.	O
</s>
<s>
When	O
the	O
top	O
array	O
is	O
RAID0	O
(	O
such	O
as	O
in	O
RAID1+0	O
and	O
RAID5+0	O
)	O
,	O
most	O
vendors	O
omit	O
the	O
"	O
+	O
"	O
(	O
yielding	O
RAID	B-Architecture
10	O
and	O
RAID50	O
,	O
respectively	O
)	O
.	O
</s>
<s>
RAID	B-Architecture
0+1	O
:	O
creates	O
two	O
stripes	O
and	O
mirrors	O
them	O
.	O
</s>
<s>
If	O
a	O
single	O
drive	O
failure	O
occurs	O
then	O
one	O
of	O
the	O
mirrors	O
has	O
failed	O
,	O
at	O
this	O
point	O
it	O
is	O
running	O
effectively	O
as	O
RAID	B-Architecture
0	O
with	O
no	O
redundancy	B-Application
.	O
</s>
<s>
Significantly	O
higher	O
risk	O
is	O
introduced	O
during	O
a	O
rebuild	O
than	O
RAID	B-Architecture
1+0	O
as	O
all	O
the	O
data	O
from	O
all	O
the	O
drives	O
in	O
the	O
remaining	O
stripe	O
has	O
to	O
be	O
read	O
rather	O
than	O
just	O
from	O
one	O
drive	O
,	O
increasing	O
the	O
chance	O
of	O
an	O
unrecoverable	O
read	O
error	O
(	O
URE	O
)	O
and	O
significantly	O
extending	O
the	O
rebuild	O
window	O
.	O
</s>
<s>
RAID	B-Architecture
1+0	O
:	O
(	O
see	O
:	O
RAID	B-Architecture
10	O
)	O
creates	O
a	O
striped	B-General_Concept
set	O
from	O
a	O
series	O
of	O
mirrored	O
drives	O
.	O
</s>
<s>
JBOD	O
RAID	B-Architecture
N+N	O
:	O
With	O
JBOD	O
(	O
just	O
a	O
bunch	O
of	O
disks	B-Device
)	O
,	O
it	O
is	O
possible	O
to	O
concatenate	O
disks	B-Device
,	O
but	O
also	O
volumes	O
such	O
as	O
RAID	B-Architecture
sets	O
.	O
</s>
<s>
With	O
larger	O
drive	O
capacities	O
,	O
write	O
delay	O
and	O
rebuilding	O
time	O
increase	O
dramatically	O
(	O
especially	O
,	O
as	O
described	O
above	O
,	O
with	O
RAID	B-Architecture
5	O
and	O
RAID	B-Architecture
6	O
)	O
.	O
</s>
<s>
By	O
splitting	O
a	O
larger	O
RAID	B-Architecture
N	O
set	O
into	O
smaller	O
subsets	O
and	O
concatenating	O
them	O
with	O
linear	O
JBOD	O
,	O
write	O
and	O
rebuilding	O
time	O
will	O
be	O
reduced	O
.	O
</s>
<s>
If	O
a	O
hardware	B-Architecture
RAID	B-Architecture
controller	O
is	O
not	O
capable	O
of	O
nesting	O
linear	O
JBOD	O
with	O
RAID	B-Architecture
N	O
,	O
then	O
linear	O
JBOD	O
can	O
be	O
achieved	O
with	O
OS-level	O
software	B-Application
RAID	B-Architecture
in	O
combination	O
with	O
separate	O
RAID	B-Architecture
N	O
subset	O
volumes	O
created	O
within	O
one	O
,	O
or	O
more	O
,	O
hardware	B-Architecture
RAID	B-Architecture
controller(s )	O
.	O
</s>
<s>
Besides	O
a	O
drastic	O
speed	O
increase	O
,	O
this	O
also	O
provides	O
a	O
substantial	O
advantage	O
:	O
the	O
possibility	O
to	O
start	O
a	O
linear	O
JBOD	O
with	O
a	O
small	O
set	O
of	O
disks	B-Device
and	O
to	O
be	O
able	O
to	O
expand	O
the	O
total	O
set	O
with	O
disks	B-Device
of	O
different	O
size	O
,	O
later	O
on	O
(	O
in	O
time	O
,	O
disks	B-Device
of	O
bigger	O
size	O
become	O
available	O
on	O
the	O
market	O
)	O
.	O
</s>
<s>
There	O
is	O
another	O
advantage	O
in	O
the	O
form	O
of	O
disaster	O
recovery	O
(	O
if	O
a	O
RAID	B-Architecture
N	O
subset	O
happens	O
to	O
fail	O
,	O
then	O
the	O
data	O
on	O
the	O
other	O
RAID	B-Architecture
N	O
subsets	O
is	O
not	O
lost	O
,	O
reducing	O
restore	O
time	O
)	O
.	O
</s>
<s>
Many	O
configurations	O
other	O
than	O
the	O
basic	O
numbered	O
RAID	B-Architecture
levels	O
are	O
possible	O
,	O
and	O
many	O
companies	O
,	O
organizations	O
,	O
and	O
groups	O
have	O
created	O
their	O
own	O
non-standard	O
configurations	O
,	O
in	O
many	O
cases	O
designed	O
to	O
meet	O
the	O
specialized	O
needs	O
of	O
a	O
small	O
niche	O
group	O
.	O
</s>
<s>
Linux	B-Application
MD	I-Application
RAID	I-Application
10	O
provides	O
a	O
general	O
RAID	B-Architecture
driver	O
that	O
in	O
its	O
"	O
near	O
"	O
layout	O
defaults	O
to	O
a	O
standard	O
RAID1	O
with	O
two	O
drives	O
,	O
and	O
a	O
standard	O
RAID1+0	O
with	O
four	O
drives	O
;	O
however	O
,	O
it	O
can	O
include	O
any	O
number	O
of	O
drives	O
,	O
including	O
odd	O
numbers	O
.	O
</s>
<s>
With	O
its	O
"	O
far	O
"	O
layout	O
,	O
MD	B-Application
RAID10	O
can	O
run	O
both	O
striped	B-General_Concept
and	O
mirrored	O
,	O
even	O
with	O
only	O
two	O
drives	O
in	O
f2	O
layout	O
;	O
this	O
runs	O
mirroring	B-General_Concept
with	O
striped	B-General_Concept
reads	O
,	O
giving	O
the	O
read	O
performance	O
of	O
RAID0	O
.	O
</s>
<s>
Regular	O
RAID1	O
,	O
as	O
provided	O
by	O
Linux	B-Application
software	I-Application
RAID	I-Application
,	O
does	O
not	O
stripe	O
reads	O
,	O
but	O
can	O
perform	O
reads	O
in	O
parallel	O
.	O
</s>
<s>
Hadoop	B-Application
has	O
a	O
RAID	B-Architecture
system	O
that	O
generates	O
a	O
parity	O
file	O
by	O
xor-ing	O
a	O
stripe	O
of	O
blocks	O
in	O
a	O
single	O
HDFS	O
file	O
.	O
</s>
<s>
BeeGFS	B-Application
,	O
the	O
parallel	O
file	B-Application
system	I-Application
,	O
has	O
internal	O
striping	B-General_Concept
(	O
comparable	O
to	O
file-based	O
RAID0	O
)	O
and	O
replication	O
(	O
comparable	O
to	O
file-based	O
RAID10	O
)	O
options	O
to	O
aggregate	O
throughput	O
and	O
capacity	O
of	O
multiple	O
servers	O
and	O
is	O
typically	O
based	O
on	O
top	O
of	O
an	O
underlying	O
RAID	B-Architecture
to	O
make	O
disk	B-Device
failures	O
transparent	O
.	O
</s>
<s>
Declustered	O
RAID	B-Architecture
scatters	O
dual	O
(	O
or	O
more	O
)	O
copies	O
of	O
the	O
data	O
across	O
all	O
disks	B-Device
(	O
possibly	O
hundreds	O
)	O
in	O
a	O
storage	O
subsystem	O
,	O
while	O
holding	O
back	O
enough	O
spare	O
capacity	O
to	O
allow	O
for	O
a	O
few	O
disks	B-Device
to	O
fail	O
.	O
</s>
<s>
When	O
one	O
or	O
more	O
disks	B-Device
fail	O
the	O
missing	O
copies	O
are	O
rebuilt	O
into	O
that	O
spare	O
capacity	O
,	O
again	O
arbitrarily	O
.	O
</s>
<s>
Because	O
the	O
rebuild	O
is	O
done	O
from	O
and	O
to	O
all	O
the	O
remaining	O
disks	B-Device
,	O
it	O
operates	O
much	O
faster	O
than	O
with	O
traditional	O
RAID	B-Architecture
,	O
reducing	O
the	O
overall	O
impact	O
on	O
clients	O
of	O
the	O
storage	O
system	O
.	O
</s>
<s>
The	O
distribution	O
of	O
data	O
across	O
multiple	O
drives	O
can	O
be	O
managed	O
either	O
by	O
dedicated	O
computer	B-Architecture
hardware	I-Architecture
or	O
by	O
software	B-Application
.	I-Application
</s>
<s>
A	O
software	B-Application
solution	O
may	O
be	O
part	O
of	O
the	O
operating	B-General_Concept
system	I-General_Concept
,	O
part	O
of	O
the	O
firmware	O
and	O
drivers	O
supplied	O
with	O
a	O
standard	O
drive	O
controller	O
(	O
so-called	O
"	O
hardware-assisted	O
software	B-Application
RAID	B-Architecture
"	O
)	O
,	O
or	O
it	O
may	O
reside	O
entirely	O
within	O
the	O
hardware	B-Architecture
RAID	B-Architecture
controller	O
.	O
</s>
<s>
Hardware	B-Architecture
RAID	B-Architecture
controllers	O
can	O
be	O
configured	O
through	O
card	O
BIOS	B-Operating_System
or	O
Option	B-Device
ROM	I-Device
before	O
an	O
operating	B-General_Concept
system	I-General_Concept
is	O
booted	O
,	O
and	O
after	O
the	O
operating	B-General_Concept
system	I-General_Concept
is	O
booted	O
,	O
proprietary	B-Application
configuration	O
utilities	O
are	O
available	O
from	O
the	O
manufacturer	O
of	O
each	O
controller	O
.	O
</s>
<s>
Unlike	O
the	O
network	B-Protocol
interface	I-Protocol
controllers	I-Protocol
for	O
Ethernet	O
,	O
which	O
can	O
usually	O
be	O
configured	O
and	O
serviced	O
entirely	O
through	O
the	O
common	O
operating	B-General_Concept
system	I-General_Concept
paradigms	O
like	O
ifconfig	B-Protocol
in	O
Unix	B-Application
,	O
without	O
a	O
need	O
for	O
any	O
third-party	O
tools	O
,	O
each	O
manufacturer	O
of	O
each	O
RAID	B-Architecture
controller	O
usually	O
provides	O
their	O
own	O
proprietary	B-Application
software	I-Application
tooling	O
for	O
each	O
operating	B-General_Concept
system	I-General_Concept
that	O
they	O
deem	O
to	O
support	O
,	O
ensuring	O
a	O
vendor	O
lock-in	O
,	O
and	O
contributing	O
to	O
reliability	O
issues	O
.	O
</s>
<s>
For	O
example	O
,	O
in	O
FreeBSD	B-Operating_System
,	O
in	O
order	O
to	O
access	O
the	O
configuration	O
of	O
Adaptec	O
RAID	B-Architecture
controllers	O
,	O
users	O
are	O
required	O
to	O
enable	O
Linux	B-Application
compatibility	O
layer	O
,	O
and	O
use	O
the	O
Linux	B-Application
tooling	O
from	O
Adaptec	O
,	O
potentially	O
compromising	O
the	O
stability	O
,	O
reliability	O
and	O
security	O
of	O
their	O
setup	O
,	O
especially	O
when	O
taking	O
the	O
long-term	O
view	O
.	O
</s>
<s>
Some	O
other	O
operating	B-General_Concept
systems	I-General_Concept
have	O
implemented	O
their	O
own	O
generic	O
frameworks	O
for	O
interfacing	O
with	O
any	O
RAID	B-Architecture
controller	O
,	O
and	O
provide	O
tools	O
for	O
monitoring	O
RAID	B-Architecture
volume	O
status	O
,	O
as	O
well	O
as	O
facilitation	O
of	O
drive	O
identification	O
through	O
LED	O
blinking	O
,	O
alarm	O
management	O
and	O
hot	B-General_Concept
spare	I-General_Concept
disk	I-General_Concept
designations	O
from	O
within	O
the	O
operating	B-General_Concept
system	I-General_Concept
without	O
having	O
to	O
reboot	O
into	O
card	O
BIOS	B-Operating_System
.	O
</s>
<s>
For	O
example	O
,	O
this	O
was	O
the	O
approach	O
taken	O
by	O
OpenBSD	B-Operating_System
in	O
2005	O
with	O
its	O
bio(4 )	O
pseudo-device	O
and	O
the	O
bioctl	B-Device
utility	O
,	O
which	O
provide	O
volume	O
status	O
,	O
and	O
allow	O
LED/alarm/hotspare	O
control	O
,	O
as	O
well	O
as	O
the	O
sensors	O
(	O
including	O
the	O
drive	O
sensor	O
)	O
for	O
health	O
monitoring	O
;	O
this	O
approach	O
has	O
subsequently	O
been	O
adopted	O
and	O
extended	O
by	O
NetBSD	B-Device
in	O
2007	O
as	O
well	O
.	O
</s>
<s>
Software	B-Application
RAID	B-Architecture
implementations	O
are	O
provided	O
by	O
many	O
modern	O
operating	B-General_Concept
systems	I-General_Concept
.	O
</s>
<s>
Software	B-Application
RAID	B-Architecture
can	O
be	O
implemented	O
as	O
:	O
</s>
<s>
Some	O
advanced	O
file	B-Application
systems	I-Application
are	O
designed	O
to	O
organize	O
data	O
across	O
multiple	O
storage	O
devices	O
directly	O
,	O
without	O
needing	O
the	O
help	O
of	O
a	O
third-party	O
logical	B-Application
volume	I-Application
manager	I-Application
:	O
</s>
<s>
ZFS	B-Application
supports	O
the	O
equivalents	O
of	O
RAID0	O
,	O
RAID1	O
,	O
RAID5	O
(	O
RAID-Z1	O
)	O
single-parity	O
,	O
RAID6	O
(	O
RAID-Z2	O
)	O
double-parity	O
,	O
and	O
a	O
triple-parity	O
version	O
(	O
RAID-Z3	B-General_Concept
)	O
also	O
referred	O
to	O
as	O
RAID7	O
.	O
</s>
<s>
As	O
it	O
always	O
stripes	O
over	O
top-level	O
vdevs	O
,	O
it	O
supports	O
equivalents	O
of	O
the	O
1+0	O
,	O
5+0	O
,	O
and	O
6+0	O
nested	B-General_Concept
RAID	I-General_Concept
levels	I-General_Concept
(	O
as	O
well	O
as	O
striped	B-General_Concept
triple-parity	O
sets	O
)	O
but	O
not	O
other	O
nested	O
combinations	O
.	O
</s>
<s>
ZFS	B-Application
is	O
the	O
native	O
file	B-Application
system	I-Application
on	O
Solaris	B-Application
and	O
illumos	B-Application
,	O
and	O
is	O
also	O
available	O
on	O
FreeBSD	B-Operating_System
and	O
Linux	B-Application
.	O
</s>
<s>
Open-source	O
ZFS	B-Application
implementations	O
are	O
actively	O
developed	O
under	O
the	O
OpenZFS	B-Operating_System
umbrella	O
project	O
.	O
</s>
<s>
Spectrum	B-Application
Scale	I-Application
,	O
initially	O
developed	O
by	O
IBM	O
for	O
media	O
streaming	O
and	O
scalable	O
analytics	O
,	O
supports	O
declustered	O
RAID	B-Architecture
protection	O
schemes	O
up	O
to	O
n+3	O
.	O
</s>
<s>
A	O
particularity	O
is	O
the	O
dynamic	O
rebuilding	O
priority	O
which	O
runs	O
with	O
low	O
impact	O
in	O
the	O
background	O
until	O
a	O
data	O
chunk	O
hits	O
n+0	O
redundancy	B-Application
,	O
in	O
which	O
case	O
this	O
chunk	O
is	O
quickly	O
rebuilt	O
to	O
at	O
least	O
n+1	O
.	O
</s>
<s>
On	O
top	O
,	O
Spectrum	B-Application
Scale	I-Application
supports	O
metro-distance	O
RAID1	O
.	O
</s>
<s>
Btrfs	B-Operating_System
supports	O
RAID0	O
,	O
RAID1	O
and	O
RAID10	O
(	O
RAID5	O
and	O
6	O
are	O
under	O
development	O
)	O
.	O
</s>
<s>
XFS	B-Application
was	O
originally	O
designed	O
to	O
provide	O
an	O
integrated	O
volume	B-Application
manager	I-Application
that	O
supports	O
concatenating	O
,	O
mirroring	B-General_Concept
and	O
striping	B-General_Concept
of	O
multiple	O
physical	O
storage	O
devices	O
.	O
</s>
<s>
However	O
,	O
the	O
implementation	O
of	O
XFS	B-Application
in	O
Linux	B-Operating_System
kernel	I-Operating_System
lacks	O
the	O
integrated	O
volume	B-Application
manager	I-Application
.	O
</s>
<s>
Many	O
operating	B-General_Concept
systems	I-General_Concept
provide	O
RAID	B-Architecture
implementations	O
,	O
including	O
the	O
following	O
:	O
</s>
<s>
Hewlett-Packard	O
'	O
s	O
OpenVMS	B-Operating_System
operating	B-General_Concept
system	I-General_Concept
supports	O
RAID1	O
.	O
</s>
<s>
The	O
mirrored	O
disks	B-Device
,	O
called	O
a	O
"	O
shadow	O
set	O
"	O
,	O
can	O
be	O
in	O
different	O
locations	O
to	O
assist	O
in	O
disaster	O
recovery	O
.	O
</s>
<s>
Apple	O
's	O
macOS	B-Application
and	O
macOS	B-Operating_System
Server	I-Operating_System
support	O
RAID0	O
,	O
RAID1	O
,	O
and	O
RAID1+0	O
.	O
</s>
<s>
FreeBSD	B-Operating_System
supports	O
RAID0	O
,	O
RAID1	O
,	O
RAID3	O
,	O
and	O
RAID5	O
,	O
and	O
all	O
nestings	O
via	O
GEOM	B-Application
modules	O
and	O
ccd	O
.	O
</s>
<s>
Linux	B-Application
's	O
md	B-Application
supports	O
RAID0	O
,	O
RAID1	O
,	O
RAID4	O
,	O
RAID5	O
,	O
RAID6	O
,	O
and	O
all	O
nestings	O
.	O
</s>
<s>
Microsoft	B-Application
Windows	I-Application
supports	O
RAID0	O
,	O
RAID1	O
,	O
and	O
RAID5	O
using	O
various	O
software	B-Application
implementations	O
.	O
</s>
<s>
Logical	O
Disk	B-Device
Manager	O
,	O
introduced	O
with	O
Windows	B-Application
2000	I-Application
,	O
allows	O
for	O
the	O
creation	O
of	O
RAID0	O
,	O
RAID1	O
,	O
and	O
RAID5	O
volumes	O
by	O
using	O
dynamic	O
disks	B-Device
,	O
but	O
this	O
was	O
limited	O
only	O
to	O
professional	O
and	O
server	O
editions	O
of	O
Windows	O
until	O
the	O
release	O
of	O
Windows	B-Application
8	I-Application
.	O
</s>
<s>
Windows	B-Application
XP	I-Application
can	O
be	O
modified	O
to	O
unlock	O
support	O
for	O
RAID0	O
,	O
1	O
,	O
and	O
5	O
.	O
</s>
<s>
Windows	B-Application
8	I-Application
and	O
Windows	B-Device
Server	I-Device
2012	I-Device
introduced	O
a	O
RAID-like	O
feature	O
known	O
as	O
Storage	O
Spaces	O
,	O
which	O
also	O
allows	O
users	O
to	O
specify	O
mirroring	B-General_Concept
,	O
parity	O
,	O
or	O
no	O
redundancy	B-Application
on	O
a	O
folder-by-folder	O
basis	O
.	O
</s>
<s>
NetBSD	B-Device
supports	O
RAID0	O
,	O
1	O
,	O
4	O
,	O
and	O
5	O
via	O
its	O
software	B-Application
implementation	O
,	O
named	O
RAIDframe	O
.	O
</s>
<s>
OpenBSD	B-Operating_System
supports	O
RAID0	O
,	O
1	O
and	O
5	O
via	O
its	O
software	B-Application
implementation	O
,	O
named	O
softraid	O
.	O
</s>
<s>
For	O
instance	O
,	O
consider	O
a	O
computer	O
whose	O
disk	B-Device
is	O
configured	O
as	O
RAID1	O
(	O
mirrored	O
drives	O
)	O
;	O
if	O
the	O
first	O
drive	O
in	O
the	O
array	O
fails	O
,	O
then	O
a	O
first-stage	B-Application
boot	I-Application
loader	I-Application
might	O
not	O
be	O
sophisticated	O
enough	O
to	O
attempt	O
loading	O
the	O
second-stage	O
boot	B-Application
loader	I-Application
from	O
the	O
second	O
drive	O
as	O
a	O
fallback	O
.	O
</s>
<s>
The	O
second-stage	O
boot	B-Application
loader	I-Application
for	O
FreeBSD	B-Operating_System
is	O
capable	O
of	O
loading	O
a	O
kernel	B-Operating_System
from	O
such	O
an	O
array	O
.	O
</s>
<s>
Software-implemented	O
RAID	B-Architecture
is	O
not	O
always	O
compatible	O
with	O
the	O
system	O
's	O
boot	O
process	O
,	O
and	O
it	O
is	O
generally	O
impractical	O
for	O
desktop	O
versions	O
of	O
Windows	O
.	O
</s>
<s>
However	O
,	O
hardware	B-Architecture
RAID	B-Architecture
controllers	O
are	O
expensive	O
and	O
proprietary	B-Application
.	O
</s>
<s>
To	O
fill	O
this	O
gap	O
,	O
inexpensive	O
"	O
RAID	B-Architecture
controllers	O
"	O
were	O
introduced	O
that	O
do	O
not	O
contain	O
a	O
dedicated	O
RAID	B-Architecture
controller	O
chip	O
,	O
but	O
simply	O
a	O
standard	O
drive	O
controller	O
chip	O
with	O
proprietary	B-Application
firmware	O
and	O
drivers	O
.	O
</s>
<s>
During	O
early	O
bootup	O
,	O
the	O
RAID	B-Architecture
is	O
implemented	O
by	O
the	O
firmware	O
and	O
,	O
once	O
the	O
operating	B-General_Concept
system	I-General_Concept
has	O
been	O
more	O
completely	O
loaded	O
,	O
the	O
drivers	O
take	O
over	O
control	O
.	O
</s>
<s>
Consequently	O
,	O
such	O
controllers	O
may	O
not	O
work	O
when	O
driver	O
support	O
is	O
not	O
available	O
for	O
the	O
host	O
operating	B-General_Concept
system	I-General_Concept
.	O
</s>
<s>
An	O
example	O
is	O
Intel	B-Application
Rapid	I-Application
Storage	I-Application
Technology	I-Application
,	O
implemented	O
on	O
many	O
consumer-level	O
motherboards	O
.	O
</s>
<s>
Because	O
some	O
minimal	O
hardware	B-Architecture
support	O
is	O
involved	O
,	O
this	O
implementation	O
is	O
also	O
called	O
"	O
hardware-assisted	O
software	B-Application
RAID	B-Architecture
"	O
,	O
"	O
hybrid	O
model	O
"	O
RAID	B-Architecture
,	O
or	O
even	O
"	O
fake	O
RAID	B-Architecture
"	O
.	O
</s>
<s>
If	O
RAID5	O
is	O
supported	O
,	O
the	O
hardware	B-Architecture
may	O
provide	O
a	O
hardware	B-Architecture
XOR	O
accelerator	O
.	O
</s>
<s>
An	O
advantage	O
of	O
this	O
model	O
over	O
the	O
pure	O
software	B-Application
RAID	B-Architecture
is	O
that	O
—	O
if	O
using	O
a	O
redundancy	B-Application
mode	O
—	O
the	O
boot	O
drive	O
is	O
protected	O
from	O
failure	O
(	O
due	O
to	O
the	O
firmware	O
)	O
during	O
the	O
boot	O
process	O
even	O
before	O
the	O
operating	B-General_Concept
system	I-General_Concept
's	O
drivers	O
take	O
over	O
.	O
</s>
<s>
Data	B-Error_Name
scrubbing	I-Error_Name
(	O
referred	O
to	O
in	O
some	O
environments	O
as	O
patrol	O
read	O
)	O
involves	O
periodic	O
reading	O
and	O
checking	O
by	O
the	O
RAID	B-Architecture
controller	O
of	O
all	O
the	O
blocks	O
in	O
an	O
array	O
,	O
including	O
those	O
not	O
otherwise	O
accessed	O
.	O
</s>
<s>
Data	B-Error_Name
scrubbing	I-Error_Name
checks	O
for	O
bad	O
blocks	O
on	O
each	O
storage	O
device	O
in	O
an	O
array	O
,	O
but	O
also	O
uses	O
the	O
redundancy	B-Application
of	O
the	O
array	O
to	O
recover	O
bad	O
blocks	O
on	O
a	O
single	O
drive	O
and	O
to	O
reassign	O
the	O
recovered	O
data	O
to	O
spare	O
blocks	O
elsewhere	O
on	O
the	O
drive	O
.	O
</s>
<s>
Frequently	O
,	O
a	O
RAID	B-Architecture
controller	O
is	O
configured	O
to	O
"	O
drop	O
"	O
a	O
component	O
drive	O
(	O
that	O
is	O
,	O
to	O
assume	O
a	O
component	O
drive	O
has	O
failed	O
)	O
if	O
the	O
drive	O
has	O
been	O
unresponsive	O
for	O
eight	O
seconds	O
or	O
so	O
;	O
this	O
might	O
cause	O
the	O
array	O
controller	O
to	O
drop	O
a	O
good	O
drive	O
because	O
that	O
drive	O
has	O
not	O
been	O
given	O
enough	O
time	O
to	O
complete	O
its	O
internal	O
error	O
recovery	O
procedure	O
.	O
</s>
<s>
Consequently	O
,	O
using	O
consumer-marketed	O
drives	O
with	O
RAID	B-Architecture
can	O
be	O
risky	O
,	O
and	O
so-called	O
"	O
enterprise	O
class	O
"	O
drives	O
limit	O
this	O
error	O
recovery	O
time	O
to	O
reduce	O
risk	O
.	O
</s>
<s>
The	O
utility	O
enabled	O
TLER	B-Device
(	O
time	O
limited	O
error	O
recovery	O
)	O
,	O
which	O
limits	O
the	O
error	O
recovery	O
time	O
to	O
seven	O
seconds	O
.	O
</s>
<s>
Around	O
September	O
2009	O
,	O
Western	O
Digital	O
disabled	O
this	O
feature	O
in	O
their	O
desktop	O
drives	O
(	O
such	O
as	O
the	O
Caviar	O
Black	O
line	O
)	O
,	O
making	O
such	O
drives	O
unsuitable	O
for	O
use	O
in	O
RAID	B-Architecture
configurations	O
.	O
</s>
<s>
However	O
,	O
Western	O
Digital	O
enterprise	O
class	O
drives	O
are	O
shipped	O
from	O
the	O
factory	O
with	O
TLER	B-Device
enabled	O
.	O
</s>
<s>
For	O
non-RAID	O
usage	O
,	O
an	O
enterprise	O
class	O
drive	O
with	O
a	O
short	O
error	O
recovery	O
timeout	O
that	O
cannot	O
be	O
changed	O
is	O
therefore	O
less	O
suitable	O
than	O
a	O
desktop	O
drive	O
.	O
</s>
<s>
In	O
late	O
2010	O
,	O
the	O
Smartmontools	B-Language
program	O
began	O
supporting	O
the	O
configuration	O
of	O
ATA	O
Error	B-Device
Recovery	I-Device
Control	I-Device
,	O
allowing	O
the	O
tool	O
to	O
configure	O
many	O
desktop	O
class	O
hard	O
drives	O
for	O
use	O
in	O
RAID	B-Architecture
setups	O
.	O
</s>
<s>
While	O
RAID	B-Architecture
may	O
protect	O
against	O
physical	O
drive	O
failure	O
,	O
the	O
data	O
is	O
still	O
exposed	O
to	O
operator	O
,	O
software	B-Application
,	O
hardware	B-Architecture
,	O
and	O
virus	O
destruction	O
.	O
</s>
<s>
Many	O
studies	O
cite	O
operator	O
fault	O
as	O
a	O
common	O
source	O
of	O
malfunction	O
,	O
such	O
as	O
a	O
server	O
operator	O
replacing	O
the	O
incorrect	O
drive	O
in	O
a	O
faulty	O
RAID	B-Architecture
,	O
and	O
disabling	O
the	O
system	O
(	O
even	O
temporarily	O
)	O
in	O
the	O
process	O
.	O
</s>
<s>
Unrecoverable	O
read	O
errors	O
(	O
URE	O
)	O
present	O
as	O
sector	B-Application
read	O
failures	O
,	O
also	O
known	O
as	O
latent	O
sector	B-Application
errors	O
(	O
LSE	O
)	O
.	O
</s>
<s>
The	O
associated	O
media	O
assessment	O
measure	O
,	O
unrecoverable	O
bit	O
error	O
(	O
UBE	O
)	O
rate	O
,	O
is	O
typically	O
guaranteed	O
to	O
be	O
less	O
than	O
one	O
bit	O
in	O
1015	O
for	O
enterprise-class	O
drives	O
(	O
SCSI	B-Architecture
,	O
FC	B-Architecture
,	O
SAS	O
or	O
SATA	O
)	O
,	O
and	O
less	O
than	O
one	O
bit	O
in	O
1014	O
for	O
desktop-class	O
drives	O
(	O
IDE/ATA/PATA	O
or	O
SATA	O
)	O
.	O
</s>
<s>
Increasing	O
drive	O
capacities	O
and	O
large	O
RAID5	O
instances	O
have	O
led	O
to	O
the	O
maximum	O
error	O
rates	O
being	O
insufficient	O
to	O
guarantee	O
a	O
successful	O
recovery	O
,	O
due	O
to	O
the	O
high	O
likelihood	O
of	O
such	O
an	O
error	O
occurring	O
on	O
one	O
or	O
more	O
remaining	O
drives	O
during	O
a	O
RAID	B-Architecture
set	O
rebuild	O
.	O
</s>
<s>
When	O
rebuilding	O
,	O
parity-based	O
schemes	O
such	O
as	O
RAID5	O
are	O
particularly	O
prone	O
to	O
the	O
effects	O
of	O
UREs	O
as	O
they	O
affect	O
not	O
only	O
the	O
sector	B-Application
where	O
they	O
occur	O
,	O
but	O
also	O
reconstructed	O
blocks	O
using	O
that	O
sector	B-Application
for	O
parity	O
computation	O
.	O
</s>
<s>
Double-protection	O
parity-based	O
schemes	O
,	O
such	O
as	O
RAID6	O
,	O
attempt	O
to	O
address	O
this	O
issue	O
by	O
providing	O
redundancy	B-Application
that	O
allows	O
double-drive	O
failures	O
;	O
as	O
a	O
downside	O
,	O
such	O
schemes	O
suffer	O
from	O
elevated	O
write	O
penalty	O
—	O
the	O
number	O
of	O
times	O
the	O
storage	O
medium	O
must	O
be	O
accessed	O
during	O
a	O
single	O
write	O
operation	O
.	O
</s>
<s>
Schemes	O
that	O
duplicate	O
(	O
mirror	O
)	O
data	O
in	O
a	O
drive-to-drive	O
manner	O
,	O
such	O
as	O
RAID1	O
and	O
RAID10	O
,	O
have	O
a	O
lower	O
risk	O
from	O
UREs	O
than	O
those	O
using	O
parity	O
computation	O
or	O
mirroring	B-General_Concept
between	O
striped	B-General_Concept
sets	O
.	O
</s>
<s>
Data	B-Error_Name
scrubbing	I-Error_Name
,	O
as	O
a	O
background	O
process	O
,	O
can	O
be	O
used	O
to	O
detect	O
and	O
recover	O
from	O
UREs	O
,	O
effectively	O
reducing	O
the	O
risk	O
of	O
them	O
happening	O
during	O
RAID	B-Architecture
rebuilds	O
and	O
causing	O
double-drive	O
failures	O
.	O
</s>
<s>
The	O
recovery	O
of	O
UREs	O
involves	O
remapping	O
of	O
affected	O
underlying	O
disk	B-Application
sectors	I-Application
,	O
utilizing	O
the	O
drive	O
's	O
sector	B-Application
remapping	O
pool	O
;	O
in	O
case	O
of	O
UREs	O
detected	O
during	O
background	O
scrubbing	O
,	O
data	B-Application
redundancy	I-Application
provided	O
by	O
a	O
fully	O
operational	O
RAID	B-Architecture
set	O
allows	O
the	O
missing	O
data	O
to	O
be	O
reconstructed	O
and	O
rewritten	O
to	O
a	O
remapped	O
sector	B-Application
.	O
</s>
<s>
Given	O
an	O
array	O
with	O
only	O
one	O
redundant	O
drive	O
(	O
which	O
applies	O
to	O
RAID	B-Architecture
levels	O
3	O
,	O
4	O
and	O
5	O
,	O
and	O
to	O
"	O
classic	O
"	O
two-drive	O
RAID1	O
)	O
,	O
a	O
second	O
drive	O
failure	O
would	O
cause	O
complete	O
failure	O
of	O
the	O
array	O
.	O
</s>
<s>
Mirroring	B-General_Concept
schemes	O
such	O
as	O
RAID10	O
have	O
a	O
bounded	O
recovery	O
time	O
as	O
they	O
require	O
the	O
copy	O
of	O
a	O
single	O
failed	O
drive	O
,	O
compared	O
with	O
parity	O
schemes	O
such	O
as	O
RAID6	O
,	O
which	O
require	O
the	O
copy	O
of	O
all	O
blocks	O
of	O
the	O
drives	O
in	O
an	O
array	O
set	O
.	O
</s>
<s>
Triple	O
parity	O
schemes	O
,	O
or	O
triple	O
mirroring	B-General_Concept
,	O
have	O
been	O
suggested	O
as	O
one	O
approach	O
to	O
improve	O
resilience	O
to	O
an	O
additional	O
drive	O
failure	O
during	O
this	O
large	O
rebuild	O
time	O
.	O
</s>
<s>
A	O
system	O
crash	O
or	O
other	O
interruption	O
of	O
a	O
write	O
operation	O
can	O
result	O
in	O
states	O
where	O
the	O
parity	O
is	O
inconsistent	O
with	O
the	O
data	O
due	O
to	O
non-atomicity	O
of	O
the	O
write	O
process	O
,	O
such	O
that	O
the	O
parity	O
cannot	O
be	O
used	O
for	O
recovery	O
in	O
the	O
case	O
of	O
a	O
disk	B-Device
failure	O
.	O
</s>
<s>
This	O
is	O
commonly	O
termed	O
the	O
write	O
hole	O
which	O
is	O
a	O
known	O
data	O
corruption	O
issue	O
in	O
older	O
and	O
low-end	O
RAIDs	B-Architecture
,	O
caused	O
by	O
interrupted	O
destaging	O
of	O
writes	O
to	O
disk	B-Device
.	O
</s>
<s>
The	O
write	O
hole	O
can	O
be	O
addressed	O
with	O
write-ahead	B-Algorithm
logging	I-Algorithm
.	O
</s>
<s>
This	O
was	O
fixed	O
in	O
mdadm	B-Application
by	O
introducing	O
a	O
dedicated	O
journaling	O
device	O
(	O
to	O
avoid	O
performance	O
penalty	O
,	O
typically	O
,	O
SSDs	B-Device
and	O
NVMs	B-General_Concept
are	O
preferred	O
)	O
for	O
that	O
purpose	O
.	O
</s>
<s>
There	O
are	O
concerns	O
about	O
write-cache	O
reliability	O
,	O
specifically	O
regarding	O
devices	O
equipped	O
with	O
a	O
write-back	O
cache	O
,	O
which	O
is	O
a	O
caching	O
system	O
that	O
reports	O
the	O
data	O
as	O
written	O
as	O
soon	O
as	O
it	O
is	O
written	O
to	O
cache	O
,	O
as	O
opposed	O
to	O
when	O
it	O
is	O
written	O
to	O
the	O
non-volatile	B-General_Concept
medium	O
.	O
</s>
<s>
If	O
the	O
system	O
experiences	O
a	O
power	O
loss	O
or	O
other	O
major	O
failure	O
,	O
the	O
data	O
may	O
be	O
irrevocably	O
lost	O
from	O
the	O
cache	O
before	O
reaching	O
the	O
non-volatile	B-General_Concept
storage	I-General_Concept
.	O
</s>
