<s>
ZFS	B-Application
(	O
previously	O
:	O
Zettabyte	B-Application
File	I-Application
System	I-Application
)	O
is	O
a	O
file	B-Application
system	I-Application
with	O
volume	B-Application
management	I-Application
capabilities	O
.	O
</s>
<s>
It	O
began	O
as	O
part	O
of	O
the	O
Sun	O
Microsystems	O
Solaris	B-Application
operating	I-Application
system	I-Application
in	O
2001	O
.	O
</s>
<s>
Large	O
parts	O
of	O
Solaris	B-Application
–	O
including	O
ZFS	B-Application
–	O
were	O
published	O
under	O
an	O
open	B-Application
source	I-Application
license	O
as	O
OpenSolaris	B-Operating_System
for	O
around	O
5	O
years	O
from	O
2005	O
,	O
before	O
being	O
placed	O
under	O
a	O
closed	B-Application
source	I-Application
license	O
when	O
Oracle	B-Application
Corporation	I-Application
acquired	O
Sun	O
in	O
20092010	O
.	O
</s>
<s>
During	O
2005	O
to	O
2010	O
,	O
the	O
open	B-Application
source	I-Application
version	O
of	O
ZFS	B-Application
was	O
ported	O
to	O
Linux	B-Application
,	O
Mac	B-Operating_System
OS	I-Operating_System
X	I-Operating_System
(	O
continued	O
as	O
MacZFS	O
)	O
and	O
FreeBSD	B-Operating_System
.	O
</s>
<s>
In	O
2010	O
,	O
the	O
illumos	B-Application
project	O
forked	B-Application
a	O
recent	O
version	O
of	O
OpenSolaris	B-Operating_System
,	O
to	O
continue	O
its	O
development	O
as	O
an	O
open	B-Application
source	I-Application
project	O
,	O
including	O
ZFS	B-Application
.	O
</s>
<s>
In	O
2013	O
,	O
OpenZFS	B-Operating_System
was	O
founded	O
to	O
coordinate	O
the	O
development	O
of	O
open	B-Application
source	I-Application
ZFS	B-Application
.	O
</s>
<s>
OpenZFS	B-Operating_System
maintains	O
and	O
manages	O
the	O
core	O
ZFS	B-Application
code	O
,	O
while	O
organizations	O
using	O
ZFS	B-Application
maintain	O
the	O
specific	O
code	O
and	O
validation	O
processes	O
required	O
for	O
ZFS	B-Application
to	O
integrate	O
within	O
their	O
systems	O
.	O
</s>
<s>
OpenZFS	B-Operating_System
is	O
widely	O
used	O
in	O
Unix-like	B-Operating_System
systems	I-Operating_System
.	O
</s>
<s>
The	O
management	O
of	O
stored	O
data	O
generally	O
involves	O
two	O
aspects	O
:	O
the	O
physical	O
volume	B-Application
management	I-Application
of	O
one	O
or	O
more	O
block	B-Application
storage	I-Application
devices	I-Application
such	O
as	O
hard	B-Device
drives	I-Device
and	O
SD	B-Device
cards	I-Device
and	O
their	O
organization	O
into	O
logical	B-General_Concept
block	I-General_Concept
devices	I-General_Concept
as	O
seen	O
by	O
the	O
operating	B-General_Concept
system	I-General_Concept
(	O
often	O
involving	O
a	O
volume	B-Application
manager	I-Application
,	O
RAID	B-Device
controller	I-Device
,	O
array	O
manager	O
,	O
or	O
suitable	O
device	B-Application
driver	I-Application
)	O
,	O
and	O
the	O
management	O
of	O
data	O
and	O
files	O
that	O
are	O
stored	O
on	O
these	O
logical	B-General_Concept
block	I-General_Concept
devices	I-General_Concept
(	O
a	O
file	B-Application
system	I-Application
or	O
other	O
data	O
storage	O
)	O
.	O
</s>
<s>
Example	O
:	O
A	O
RAID	B-Architecture
array	I-Architecture
of	O
2	O
hard	B-Device
drives	I-Device
and	O
an	O
SSD	B-Device
caching	B-General_Concept
disk	O
is	O
controlled	O
by	O
Intel	B-Application
's	I-Application
RST	I-Application
system	I-Application
,	O
part	O
of	O
the	O
chipset	B-Device
and	O
firmware	B-Application
built	O
into	O
a	O
desktop	O
computer	O
.	O
</s>
<s>
The	O
Windows	O
user	O
sees	O
this	O
as	O
a	O
single	O
volume	O
,	O
containing	O
an	O
NTFS-formatted	O
drive	O
of	O
their	O
data	O
,	O
and	O
NTFS	B-Application
is	O
not	O
necessarily	O
aware	O
of	O
the	O
manipulations	O
that	O
may	O
be	O
required	O
(	O
such	O
as	O
reading	O
from/writing	O
to	O
the	O
cache	B-General_Concept
drive	O
or	O
rebuilding	O
the	O
RAID	B-Architecture
array	I-Architecture
if	O
a	O
disk	O
fails	O
)	O
.	O
</s>
<s>
ZFS	B-Application
is	O
unusual	O
because	O
,	O
unlike	O
most	O
other	O
storage	O
systems	O
,	O
it	O
unifies	O
both	O
of	O
these	O
roles	O
and	O
acts	O
as	O
both	O
the	O
volume	B-Application
manager	I-Application
and	O
the	O
file	B-Application
system	I-Application
.	O
</s>
<s>
ZFS	B-Application
is	O
designed	O
to	O
ensure	O
(	O
subject	O
to	O
suitable	O
hardware	B-Architecture
)	O
that	O
data	O
stored	O
on	O
disks	O
cannot	O
be	O
lost	O
due	O
to	O
physical	O
errors	O
or	O
misprocessing	O
by	O
the	O
hardware	B-Architecture
or	O
operating	B-General_Concept
system	I-General_Concept
,	O
or	O
bit	O
rot	O
events	O
and	O
data	O
corruption	O
which	O
may	O
happen	O
over	O
time	O
,	O
and	O
its	O
complete	O
control	O
of	O
the	O
storage	O
system	O
is	O
used	O
to	O
ensure	O
that	O
every	O
step	O
,	O
whether	O
related	O
to	O
file	O
management	O
or	O
disk	O
management	O
,	O
is	O
verified	O
,	O
confirmed	O
,	O
corrected	O
if	O
needed	O
,	O
and	O
optimized	O
,	O
in	O
a	O
way	O
that	O
storage	O
controller	O
cards	O
and	O
separate	O
volume	O
and	O
file	O
managers	O
cannot	O
achieve	O
.	O
</s>
<s>
ZFS	B-Application
also	O
includes	O
a	O
mechanism	O
for	O
dataset	O
and	O
pool-level	O
snapshots	B-Application
and	O
replication	B-Architecture
,	O
including	O
snapshot	B-Application
cloning	B-Application
which	O
is	O
described	O
by	O
the	O
FreeBSD	B-Operating_System
documentation	O
as	O
one	O
of	O
its	O
"	O
most	O
powerful	O
features	O
"	O
,	O
having	O
features	O
that	O
"	O
even	O
other	O
file	B-Application
systems	I-Application
with	O
snapshot	B-Application
functionality	O
lack	O
"	O
.	O
</s>
<s>
Very	O
large	O
numbers	O
of	O
snapshots	B-Application
can	O
be	O
taken	O
,	O
without	O
degrading	O
performance	O
,	O
allowing	O
snapshots	B-Application
to	O
be	O
used	O
prior	O
to	O
risky	O
system	O
operations	O
and	O
software	O
changes	O
,	O
or	O
an	O
entire	O
production	O
(	O
"	O
live	O
"	O
)	O
file	B-Application
system	I-Application
to	O
be	O
fully	O
snapshotted	O
several	O
times	O
an	O
hour	O
,	O
in	O
order	O
to	O
mitigate	O
data	O
loss	O
due	O
to	O
user	O
error	O
or	O
malicious	O
activity	O
.	O
</s>
<s>
Snapshots	B-Application
can	O
be	O
rolled	O
back	O
"	O
live	O
"	O
or	O
previous	O
file	B-Application
system	I-Application
states	O
can	O
be	O
viewed	O
,	O
even	O
on	O
very	O
large	O
file	B-Application
systems	I-Application
,	O
leading	O
to	O
savings	O
in	O
comparison	O
to	O
formal	O
backup	O
and	O
restore	O
processes	O
.	O
</s>
<s>
Snapshots	B-Application
can	O
also	O
be	O
cloned	O
to	O
form	O
new	O
independent	O
file	B-Application
systems	I-Application
.	O
</s>
<s>
A	O
pool	O
level	O
snapshot	B-Application
(	O
known	O
as	O
a	O
"	O
checkpoint	O
"	O
)	O
is	O
available	O
which	O
allows	O
rollback	O
of	O
operations	O
that	O
may	O
affect	O
the	O
entire	O
pool	O
's	O
structure	O
,	O
or	O
which	O
add	O
or	O
remove	O
entire	O
datasets	O
.	O
</s>
<s>
In	O
1987	O
,	O
AT&T	O
Corporation	O
and	O
Sun	O
announced	O
that	O
they	O
were	O
collaborating	O
on	O
a	O
project	O
to	O
merge	O
the	O
most	O
popular	O
Unix	B-Application
variants	O
on	O
the	O
market	O
at	O
that	O
time	O
:	O
Berkeley	B-Operating_System
Software	I-Operating_System
Distribution	I-Operating_System
,	O
UNIX	B-Operating_System
System	I-Operating_System
V	I-Operating_System
,	O
and	O
Xenix	B-Application
.	O
</s>
<s>
This	O
became	O
Unix	B-Operating_System
System	I-Operating_System
V	I-Operating_System
Release	O
4	O
(	O
SVR4	O
)	O
.	O
</s>
<s>
The	O
project	O
was	O
released	O
under	O
the	O
name	O
Solaris	B-Application
,	O
which	O
became	O
the	O
successor	O
to	O
SunOS	B-Operating_System
4	O
(	O
although	O
SunOS	B-Operating_System
4.1.x	O
micro	O
releases	O
were	O
retroactively	O
named	O
Solaris	B-Application
1	O
)	O
.	O
</s>
<s>
ZFS	B-Application
was	O
designed	O
and	O
implemented	O
by	O
a	O
team	O
at	O
Sun	O
led	O
by	O
Jeff	O
Bonwick	O
,	O
Bill	O
Moore	O
and	O
Matthew	O
Ahrens	O
.	O
</s>
<s>
Source	O
code	O
for	O
ZFS	B-Application
was	O
integrated	O
into	O
the	O
main	O
trunk	O
of	O
Solaris	B-Application
development	O
on	O
October	O
31	O
,	O
2005	O
,	O
and	O
released	O
for	O
developers	O
as	O
part	O
of	O
build	O
27	O
of	O
OpenSolaris	B-Operating_System
on	O
November	O
16	O
,	O
2005	O
.	O
</s>
<s>
In	O
June	O
2006	O
,	O
Sun	O
announced	O
that	O
ZFS	B-Application
was	O
included	O
in	O
the	O
mainstream	O
6/06	O
update	O
to	O
Solaris	B-Application
10	I-Application
.	O
</s>
<s>
Solaris	B-Application
was	O
originally	O
developed	O
as	O
proprietary	B-Application
software	I-Application
,	O
but	O
Sun	O
Microsystems	O
was	O
an	O
early	O
commercial	O
proponent	O
of	O
open	B-Application
source	I-Application
software	I-Application
and	O
in	O
June	O
2005	O
released	O
most	O
of	O
the	O
Solaris	B-Application
codebase	B-General_Concept
under	O
the	O
CDDL	B-Application
license	O
and	O
founded	O
the	O
OpenSolaris	B-Operating_System
open-source	B-Application
project	O
.	O
</s>
<s>
In	O
Solaris	B-Application
10	I-Application
6/06	O
(	O
"	O
U2	O
"	O
)	O
,	O
Sun	O
added	O
the	O
ZFS	B-Application
file	B-Application
system	I-Application
and	O
during	O
the	O
next	O
5	O
years	O
frequently	O
updated	O
ZFS	B-Application
with	O
new	O
features	O
.	O
</s>
<s>
ZFS	B-Application
was	O
ported	O
to	O
Linux	B-Application
,	O
Mac	B-Operating_System
OS	I-Operating_System
X	I-Operating_System
(	O
continued	O
as	O
MacZFS	O
)	O
and	O
FreeBSD	B-Operating_System
,	O
under	O
this	O
open	B-Application
source	I-Application
license	O
.	O
</s>
<s>
The	O
name	O
at	O
one	O
point	O
was	O
said	O
to	O
stand	O
for	O
"	O
Zettabyte	B-Application
File	I-Application
System	I-Application
"	O
,	O
but	O
by	O
2006	O
,	O
the	O
name	O
was	O
no	O
longer	O
considered	O
to	O
be	O
an	O
abbreviation	O
.	O
</s>
<s>
A	O
ZFS	B-Application
file	B-Application
system	I-Application
can	O
store	O
up	O
to	O
256	O
quadrillion	O
zettabytes	O
(	O
ZB	O
)	O
.	O
</s>
<s>
In	O
September	O
2007	O
,	O
NetApp	O
sued	O
Sun	O
claiming	O
that	O
ZFS	B-Application
infringed	O
some	O
of	O
NetApp	O
's	O
patents	O
on	O
Write	B-Application
Anywhere	I-Application
File	I-Application
Layout	I-Application
.	O
</s>
<s>
Ported	O
versions	O
of	O
ZFS	B-Application
began	O
to	O
appear	O
in	O
2005	O
.	O
</s>
<s>
After	O
the	O
Sun	O
acquisition	O
by	O
Oracle	B-Application
in	O
2010	O
,	O
Oracle	B-Application
's	O
version	O
of	O
ZFS	B-Application
became	O
closed	B-Application
source	I-Application
and	O
development	O
of	O
open-source	B-Application
versions	O
proceeded	O
independently	O
,	O
coordinated	O
by	O
OpenZFS	B-Operating_System
from	O
2013	O
.	O
</s>
<s>
Examples	O
of	O
features	O
specific	O
to	O
ZFS	B-Application
include	O
:	O
</s>
<s>
Checksums	B-Algorithm
are	O
stored	O
with	O
a	O
block	B-General_Concept
's	O
parent	O
block	B-General_Concept
,	O
rather	O
than	O
with	O
the	O
block	B-General_Concept
itself	O
.	O
</s>
<s>
This	O
contrasts	O
with	O
many	O
file	B-Application
systems	I-Application
where	O
checksums	B-Algorithm
(	O
if	O
held	O
)	O
are	O
stored	O
with	O
the	O
data	O
so	O
that	O
if	O
the	O
data	O
is	O
lost	O
or	O
corrupt	O
,	O
the	O
checksum	B-Algorithm
is	O
also	O
likely	O
to	O
be	O
lost	O
or	O
incorrect	O
.	O
</s>
<s>
Automatic	O
rollback	O
of	O
recent	O
changes	O
to	O
the	O
file	B-Application
system	I-Application
and	O
data	O
,	O
in	O
some	O
circumstances	O
,	O
in	O
the	O
event	O
of	O
an	O
error	O
or	O
inconsistency	O
.	O
</s>
<s>
Data	O
can	O
be	O
reconstructed	O
using	O
all	O
of	O
the	O
following	O
:	O
error	B-Error_Name
detection	I-Error_Name
and	I-Error_Name
correction	I-Error_Name
checksums	B-Algorithm
stored	O
in	O
each	O
block	B-General_Concept
's	O
parent	O
block	B-General_Concept
;	O
multiple	O
copies	O
of	O
data	O
(	O
including	O
checksums	B-Algorithm
)	O
held	O
on	O
the	O
disk	O
;	O
write	O
intentions	O
logged	O
on	O
the	O
SLOG	O
(	O
ZIL	O
)	O
for	O
writes	O
that	O
should	O
have	O
occurred	O
but	O
did	O
not	O
occur	O
(	O
after	O
a	O
power	O
failure	O
)	O
;	O
parity	B-Error_Name
data	O
from	O
RAID/RAID	O
-Z	O
disks	O
and	O
volumes	O
;	O
copies	O
of	O
data	O
from	O
mirrored	O
disks	O
and	O
volumes	O
.	O
</s>
<s>
Native	O
handling	O
of	O
standard	B-General_Concept
RAID	I-General_Concept
levels	I-General_Concept
and	O
additional	O
ZFS	B-Application
RAID	B-Architecture
layouts	O
(	O
"	O
RAID-Z	O
"	O
)	O
.	O
</s>
<s>
The	O
RAID-Z	O
levels	O
stripe	O
data	O
across	O
only	O
the	O
disks	O
required	O
,	O
for	O
efficiency	O
(	O
many	O
RAID	B-Architecture
systems	O
stripe	O
indiscriminately	O
across	O
all	O
devices	O
)	O
,	O
and	O
checksumming	O
allows	O
rebuilding	O
of	O
inconsistent	O
or	O
corrupted	O
data	O
to	O
be	O
minimized	O
to	O
those	O
blocks	O
with	O
defects	O
;	O
</s>
<s>
Native	O
handling	O
of	O
tiered	O
storage	O
and	O
caching	B-General_Concept
devices	O
,	O
which	O
is	O
usually	O
a	O
volume	O
related	O
task	O
.	O
</s>
<s>
Because	O
ZFS	B-Application
also	O
understands	O
the	O
file	B-Application
system	I-Application
,	O
it	O
can	O
use	O
file-related	O
knowledge	O
to	O
inform	O
,	O
integrate	O
and	O
optimize	O
its	O
tiered	O
storage	O
handling	O
which	O
a	O
separate	O
device	O
cannot	O
;	O
</s>
<s>
Native	O
handling	O
of	O
snapshots	B-Application
and	O
backup/replication	O
which	O
can	O
be	O
made	O
efficient	O
by	O
integrating	O
the	O
volume	O
and	O
file	O
handling	O
.	O
</s>
<s>
Native	O
data	B-General_Concept
compression	I-General_Concept
and	O
deduplication	B-General_Concept
,	O
although	O
the	O
latter	O
is	O
largely	O
handled	O
in	O
RAM	B-Architecture
and	O
is	O
memory	O
hungry	O
.	O
</s>
<s>
Efficient	O
rebuilding	O
of	O
RAID	B-Architecture
arrays	I-Architecture
—	O
a	O
RAID	B-Device
controller	I-Device
often	O
has	O
to	O
rebuild	O
an	O
entire	O
disk	O
,	O
but	O
ZFS	B-Application
can	O
combine	O
disk	O
and	O
file	O
knowledge	O
to	O
limit	O
any	O
rebuilding	O
to	O
data	O
which	O
is	O
actually	O
missing	O
or	O
corrupt	O
,	O
greatly	O
speeding	O
up	O
rebuilding	O
;	O
</s>
<s>
Unaffected	O
by	O
RAID	B-Architecture
hardware	B-Architecture
changes	O
which	O
affect	O
many	O
other	O
systems	O
.	O
</s>
<s>
On	O
many	O
systems	O
,	O
if	O
self-contained	O
RAID	B-Architecture
hardware	B-Architecture
such	O
as	O
a	O
RAID	B-Architecture
card	O
fails	O
,	O
or	O
the	O
data	O
is	O
moved	O
to	O
another	O
RAID	B-Architecture
system	O
,	O
the	O
file	B-Application
system	I-Application
will	O
lack	O
information	O
that	O
was	O
on	O
the	O
original	O
RAID	B-Architecture
hardware	B-Architecture
,	O
which	O
is	O
needed	O
to	O
manage	O
data	O
on	O
the	O
RAID	B-Architecture
array	I-Architecture
.	O
</s>
<s>
This	O
can	O
lead	O
to	O
a	O
total	O
loss	O
of	O
data	O
unless	O
near-identical	O
hardware	B-Architecture
can	O
be	O
acquired	O
and	O
used	O
as	O
a	O
"	O
stepping	O
stone	O
"	O
.	O
</s>
<s>
Since	O
ZFS	B-Application
manages	O
RAID	B-Architecture
itself	O
,	O
a	O
ZFS	B-Application
pool	O
can	O
be	O
migrated	O
to	O
other	O
hardware	B-Architecture
,	O
or	O
the	O
operating	B-General_Concept
system	I-General_Concept
can	O
be	O
reinstalled	O
,	O
and	O
the	O
RAID-Z	O
structures	O
and	O
data	O
will	O
be	O
recognized	O
and	O
immediately	O
accessible	O
by	O
ZFS	B-Application
again	O
.	O
</s>
<s>
Ability	O
to	O
identify	O
data	O
that	O
would	O
have	O
been	O
found	O
in	O
a	O
cache	B-General_Concept
but	O
has	O
been	O
discarded	O
recently	O
instead	O
;	O
this	O
allows	O
ZFS	B-Application
to	O
reassess	O
its	O
caching	B-General_Concept
decisions	O
in	O
light	O
of	O
later	O
use	O
and	O
facilitates	O
very	O
high	O
cache-hit	O
levels	O
(	O
ZFS	B-Application
cache	B-General_Concept
hit	O
rates	O
are	O
typically	O
over	O
80%	O
)	O
;	O
</s>
<s>
Alternative	O
caching	B-General_Concept
strategies	O
can	O
be	O
used	O
for	O
data	O
that	O
would	O
otherwise	O
cause	O
delays	O
in	O
data	O
handling	O
.	O
</s>
<s>
For	O
example	O
,	O
synchronous	O
writes	O
which	O
are	O
capable	O
of	O
slowing	O
down	O
the	O
storage	O
system	O
can	O
be	O
converted	O
to	O
asynchronous	O
writes	O
by	O
being	O
written	O
to	O
a	O
fast	O
separate	O
caching	B-General_Concept
device	O
,	O
known	O
as	O
the	O
SLOG	O
(	O
sometimes	O
called	O
the	O
ZIL	O
–	O
ZFS	B-Application
Intent	O
Log	O
)	O
.	O
</s>
<s>
One	O
major	O
feature	O
that	O
distinguishes	O
ZFS	B-Application
from	O
other	O
file	B-Application
systems	I-Application
is	O
that	O
it	O
is	O
designed	O
with	O
a	O
focus	O
on	O
data	O
integrity	O
by	O
protecting	O
the	O
user	O
's	O
data	O
on	O
disk	O
against	O
silent	O
data	O
corruption	O
caused	O
by	O
data	O
degradation	O
,	O
power	O
surges	O
(	O
voltage	O
spikes	O
)	O
,	O
bugs	O
in	O
disk	O
firmware	B-Application
,	O
phantom	O
writes	O
(	O
the	O
previous	O
write	O
did	O
not	O
make	O
it	O
to	O
disk	O
)	O
,	O
misdirected	O
reads/writes	O
(	O
the	O
disk	O
accesses	O
the	O
wrong	O
block	B-General_Concept
)	O
,	O
DMA	O
parity	B-Error_Name
errors	I-Error_Name
between	O
the	O
array	O
and	O
server	B-Application
memory	O
or	O
from	O
the	O
driver	O
(	O
since	O
the	O
checksum	B-Algorithm
validates	O
data	O
inside	O
the	O
array	O
)	O
,	O
driver	O
errors	O
(	O
data	O
winds	O
up	O
in	O
the	O
wrong	O
buffer	O
inside	O
the	O
kernel	O
)	O
,	O
accidental	O
overwrites	O
(	O
such	O
as	O
swapping	O
to	O
a	O
live	O
file	B-Application
system	I-Application
)	O
,	O
etc	O
.	O
</s>
<s>
A	O
1999	O
study	O
showed	O
that	O
neither	O
any	O
of	O
the	O
then-major	O
and	O
widespread	O
filesystems	B-Application
(	O
such	O
as	O
UFS	B-Operating_System
,	O
Ext	B-Application
,	O
XFS	B-Application
,	O
JFS	B-Application
,	O
or	O
NTFS	B-Application
)	O
,	O
nor	O
hardware	B-Architecture
RAID	B-Architecture
(	O
which	O
has	O
some	O
issues	O
with	O
data	O
integrity	O
)	O
provided	O
sufficient	O
protection	O
against	O
data	O
corruption	O
problems	O
.	O
</s>
<s>
Initial	O
research	O
indicates	O
that	O
ZFS	B-Application
protects	O
data	O
better	O
than	O
earlier	O
efforts	O
.	O
</s>
<s>
It	O
is	O
also	O
faster	O
than	O
UFS	B-Operating_System
and	O
can	O
be	O
seen	O
as	O
its	O
replacement	O
.	O
</s>
<s>
Within	O
ZFS	B-Application
,	O
data	O
integrity	O
is	O
achieved	O
by	O
using	O
a	O
Fletcher-based	B-Algorithm
checksum	B-Algorithm
or	O
a	O
SHA-256	B-Algorithm
hash	B-Algorithm
throughout	O
the	O
file	B-Application
system	I-Application
tree	O
.	O
</s>
<s>
Each	O
block	B-General_Concept
of	O
data	O
is	O
checksummed	O
and	O
the	O
checksum	B-Algorithm
value	O
is	O
then	O
saved	O
in	O
the	O
pointer	O
to	O
that	O
block	B-General_Concept
—	O
rather	O
than	O
at	O
the	O
actual	O
block	B-General_Concept
itself	O
.	O
</s>
<s>
Next	O
,	O
the	O
block	B-General_Concept
pointer	O
is	O
checksummed	O
,	O
with	O
the	O
value	O
being	O
saved	O
at	O
its	O
pointer	O
.	O
</s>
<s>
This	O
checksumming	O
continues	O
all	O
the	O
way	O
up	O
the	O
file	B-Application
system	I-Application
's	O
data	O
hierarchy	O
to	O
the	O
root	O
node	O
,	O
which	O
is	O
also	O
checksummed	O
,	O
thus	O
creating	O
a	O
Merkle	B-Algorithm
tree	I-Algorithm
.	O
</s>
<s>
In-flight	O
data	O
corruption	O
or	O
phantom	O
reads/writes	O
(	O
the	O
data	O
written/read	O
checksums	B-Algorithm
correctly	O
but	O
is	O
actually	O
wrong	O
)	O
are	O
undetectable	O
by	O
most	O
filesystems	B-Application
as	O
they	O
store	O
the	O
checksum	B-Algorithm
with	O
the	O
data	O
.	O
</s>
<s>
ZFS	B-Application
stores	O
the	O
checksum	B-Algorithm
of	O
each	O
block	B-General_Concept
in	O
its	O
parent	O
block	B-General_Concept
pointer	O
so	O
the	O
entire	O
pool	O
self-validates	O
.	O
</s>
<s>
When	O
a	O
block	B-General_Concept
is	O
accessed	O
,	O
regardless	O
of	O
whether	O
it	O
is	O
data	O
or	O
meta-data	O
,	O
its	O
checksum	B-Algorithm
is	O
calculated	O
and	O
compared	O
with	O
the	O
stored	O
checksum	B-Algorithm
value	O
of	O
what	O
it	O
"	O
should	O
"	O
be	O
.	O
</s>
<s>
If	O
the	O
checksums	B-Algorithm
match	O
,	O
the	O
data	O
are	O
passed	O
up	O
the	O
programming	O
stack	O
to	O
the	O
process	O
that	O
asked	O
for	O
it	O
;	O
if	O
the	O
values	O
do	O
not	O
match	O
,	O
then	O
ZFS	B-Application
can	O
heal	O
the	O
data	O
if	O
the	O
storage	O
pool	O
provides	O
data	B-Application
redundancy	I-Application
(	O
such	O
as	O
with	O
internal	O
mirroring	B-General_Concept
)	O
,	O
assuming	O
that	O
the	O
copy	O
of	O
data	O
is	O
undamaged	O
and	O
with	O
matching	O
checksums	B-Algorithm
.	O
</s>
<s>
It	O
is	O
optionally	O
possible	O
to	O
provide	O
additional	O
in-pool	O
redundancy	B-Application
by	O
specifying	O
(	O
or	O
)	O
,	O
which	O
means	O
that	O
data	O
will	O
be	O
stored	O
twice	O
(	O
or	O
three	O
times	O
)	O
on	O
the	O
disk	O
,	O
effectively	O
halving	O
(	O
or	O
,	O
for	O
,	O
reducing	O
to	O
one	O
third	O
)	O
the	O
storage	O
capacity	O
of	O
the	O
disk	O
.	O
</s>
<s>
Additionally	O
some	O
kinds	O
of	O
data	O
used	O
by	O
ZFS	B-Application
to	O
manage	O
the	O
pool	O
are	O
stored	O
multiple	O
times	O
by	O
default	O
for	O
safety	O
,	O
even	O
with	O
the	O
default	O
copies	O
=	O
1	O
setting	O
.	O
</s>
<s>
If	O
other	O
copies	O
of	O
the	O
damaged	O
data	O
exist	O
or	O
can	O
be	O
reconstructed	O
from	O
checksums	B-Algorithm
and	O
parity	B-Error_Name
data	O
,	O
ZFS	B-Application
will	O
use	O
a	O
copy	O
of	O
the	O
data	O
(	O
or	O
recreate	O
it	O
via	O
a	O
RAID	B-Architecture
recovery	O
mechanism	O
)	O
,	O
and	O
recalculate	O
the	O
checksum	B-Algorithm
—	O
ideally	O
resulting	O
in	O
the	O
reproduction	O
of	O
the	O
originally	O
expected	O
value	O
.	O
</s>
<s>
If	O
the	O
data	O
passes	O
this	O
integrity	O
check	O
,	O
the	O
system	O
can	O
then	O
update	O
all	O
faulty	O
copies	O
with	O
known-good	O
data	O
and	O
redundancy	B-Application
will	O
be	O
restored	O
.	O
</s>
<s>
If	O
there	O
are	O
no	O
copies	O
of	O
the	O
damaged	O
data	O
,	O
ZFS	B-Application
puts	O
the	O
pool	O
to	O
a	O
faulted	O
state	O
,	O
preventing	O
its	O
future	O
use	O
and	O
providing	O
no	O
documented	O
ways	O
to	O
recover	O
pool	O
contents	O
.	O
</s>
<s>
Consistency	O
of	O
data	O
held	O
in	O
memory	O
,	O
such	O
as	O
cached	O
data	O
in	O
the	O
ARC	O
,	O
is	O
not	O
checked	O
by	O
default	O
,	O
as	O
ZFS	B-Application
is	O
expected	O
to	O
run	O
on	O
enterprise-quality	O
hardware	B-Architecture
with	O
error	B-General_Concept
correcting	I-General_Concept
RAM	I-General_Concept
,	O
but	O
the	O
capability	O
to	O
check	O
in-memory	O
data	O
exists	O
and	O
can	O
be	O
enabled	O
using	O
"	O
debug	O
flags	O
"	O
.	O
</s>
<s>
For	O
ZFS	B-Application
to	O
be	O
able	O
to	O
guarantee	O
data	O
integrity	O
,	O
it	O
needs	O
multiple	O
copies	O
of	O
the	O
data	O
,	O
usually	O
spread	O
across	O
multiple	O
disks	O
.	O
</s>
<s>
Typically	O
this	O
is	O
achieved	O
by	O
using	O
either	O
a	O
RAID	B-Device
controller	I-Device
or	O
so-called	O
"	O
soft	O
"	O
RAID	B-Architecture
(	O
built	O
into	O
a	O
file	B-Application
system	I-Application
)	O
.	O
</s>
<s>
While	O
ZFS	B-Application
can	O
work	O
with	O
hardware	B-Architecture
RAID	B-Architecture
devices	O
,	O
ZFS	B-Application
will	O
usually	O
work	O
more	O
efficiently	O
and	O
with	O
greater	O
data	O
protection	O
if	O
it	O
has	O
raw	O
access	O
to	O
all	O
storage	O
devices	O
.	O
</s>
<s>
ZFS	B-Application
relies	O
on	O
the	O
disk	O
for	O
an	O
honest	O
view	O
to	O
determine	O
the	O
moment	O
data	O
is	O
confirmed	O
as	O
safely	O
written	O
and	O
it	O
has	O
numerous	O
algorithms	O
designed	O
to	O
optimize	O
its	O
use	O
of	O
caching	B-General_Concept
,	O
cache	B-General_Concept
flushing	O
,	O
and	O
disk	O
handling	O
.	O
</s>
<s>
Disks	O
connected	O
to	O
the	O
system	O
using	O
a	O
hardware	B-Architecture
,	O
firmware	B-Application
,	O
other	O
"	O
soft	O
"	O
RAID	B-Architecture
,	O
or	O
any	O
other	O
controller	O
that	O
modifies	O
the	O
ZFS-to-disk	O
I/O	B-General_Concept
path	O
will	O
affect	O
ZFS	B-Application
performance	O
and	O
data	O
integrity	O
.	O
</s>
<s>
If	O
a	O
third-party	O
device	O
performs	O
caching	B-General_Concept
or	O
presents	O
drives	O
to	O
ZFS	B-Application
as	O
a	O
single	O
system	O
without	O
the	O
low	O
level	O
view	O
ZFS	B-Application
relies	O
upon	O
,	O
there	O
is	O
a	O
much	O
greater	O
chance	O
that	O
the	O
system	O
will	O
perform	O
less	O
optimally	O
and	O
that	O
ZFS	B-Application
will	O
be	O
less	O
likely	O
to	O
prevent	O
failures	O
,	O
recover	O
from	O
failures	O
more	O
slowly	O
,	O
or	O
lose	O
data	O
due	O
to	O
a	O
write	O
failure	O
.	O
</s>
<s>
For	O
example	O
,	O
if	O
a	O
hardware	B-Architecture
RAID	B-Architecture
card	O
is	O
used	O
,	O
ZFS	B-Application
may	O
not	O
be	O
able	O
to	O
:	O
determine	O
the	O
condition	O
of	O
disks	O
;	O
determine	O
if	O
the	O
RAID	B-Architecture
array	I-Architecture
is	O
degraded	O
or	O
rebuilding	O
;	O
detect	O
all	O
data	O
corruption	O
;	O
place	O
data	O
optimally	O
across	O
the	O
disks	O
;	O
make	O
selective	O
repairs	O
;	O
control	O
how	O
repairs	O
are	O
balanced	O
with	O
ongoing	O
use	O
;	O
or	O
make	O
repairs	O
that	O
ZFS	B-Application
could	O
usually	O
undertake	O
.	O
</s>
<s>
The	O
hardware	B-Architecture
RAID	B-Architecture
card	O
will	O
interfere	O
with	O
ZFS	B-Application
 '	O
algorithms	O
.	O
</s>
<s>
RAID	B-Device
controllers	I-Device
also	O
usually	O
add	O
controller-dependent	O
data	O
to	O
the	O
drives	O
which	O
prevents	O
software	O
RAID	B-Architecture
from	O
accessing	O
the	O
user	O
data	O
.	O
</s>
<s>
In	O
the	O
case	O
of	O
a	O
hardware	B-Architecture
RAID	B-Device
controller	I-Device
failure	O
,	O
it	O
may	O
be	O
possible	O
to	O
read	O
the	O
data	O
with	O
another	O
compatible	O
controller	O
,	O
but	O
this	O
is	O
n't	O
always	O
possible	O
and	O
a	O
replacement	O
may	O
not	O
be	O
available	O
.	O
</s>
<s>
Alternate	O
hardware	B-Architecture
RAID	B-Device
controllers	I-Device
may	O
not	O
understand	O
the	O
original	O
manufacturer	O
's	O
custom	O
data	O
required	O
to	O
manage	O
and	O
restore	O
an	O
array	O
.	O
</s>
<s>
Unlike	O
most	O
other	O
systems	O
where	O
RAID	B-Architecture
cards	O
or	O
similar	O
hardware	B-Architecture
can	O
offload	O
resources	O
and	O
processing	O
to	O
enhance	O
performance	O
and	O
reliability	O
,	O
with	O
ZFS	B-Application
it	O
is	O
strongly	O
recommended	O
that	O
these	O
methods	O
not	O
be	O
used	O
as	O
they	O
typically	O
reduce	O
the	O
system	O
's	O
performance	O
and	O
reliability	O
.	O
</s>
<s>
If	O
disks	O
must	O
be	O
attached	O
through	O
a	O
RAID	B-Architecture
or	O
other	O
controller	O
,	O
it	O
is	O
recommended	O
to	O
minimize	O
the	O
amount	O
of	O
processing	O
done	O
in	O
the	O
controller	O
by	O
using	O
a	O
plain	O
HBA	O
(	O
host	B-Architecture
adapter	I-Architecture
)	O
,	O
a	O
simple	O
fanout	O
card	O
,	O
or	O
configure	O
the	O
card	O
in	O
JBOD	O
mode	O
(	O
i.e.	O
</s>
<s>
turn	O
off	O
RAID	B-Architecture
and	O
caching	B-General_Concept
functions	O
)	O
,	O
to	O
allow	O
devices	O
to	O
be	O
attached	O
with	O
minimal	O
changes	O
in	O
the	O
ZFS-to-disk	O
I/O	B-General_Concept
pathway	O
.	O
</s>
<s>
A	O
RAID	B-Architecture
card	O
in	O
JBOD	O
mode	O
may	O
still	O
interfere	O
if	O
it	O
has	O
a	O
cache	B-General_Concept
or	O
,	O
depending	O
upon	O
its	O
design	O
,	O
may	O
detach	O
drives	O
that	O
do	O
not	O
respond	O
in	O
time	O
(	O
as	O
has	O
been	O
seen	O
with	O
many	O
energy-efficient	O
consumer-grade	O
hard	B-Device
drives	I-Device
)	O
,	O
and	O
as	O
such	O
,	O
may	O
require	O
Time-Limited	B-Device
Error	I-Device
Recovery	I-Device
(	O
TLER	B-Device
)	O
/CCTL/ERC	O
-enabled	O
drives	O
to	O
prevent	O
drive	O
dropouts	O
,	O
so	O
not	O
all	O
cards	O
are	O
suitable	O
even	O
with	O
RAID	B-Architecture
functions	O
disabled	O
.	O
</s>
<s>
Instead	O
of	O
hardware	B-Architecture
RAID	B-Architecture
,	O
ZFS	B-Application
employs	O
"	O
soft	O
"	O
RAID	B-Architecture
,	O
offering	O
RAID-Z	O
(	O
parity	B-Error_Name
based	O
like	O
RAID	B-Architecture
5	O
and	O
similar	O
)	O
and	O
disk	B-General_Concept
mirroring	I-General_Concept
(	O
similar	O
to	O
RAID	B-Architecture
1	O
)	O
.	O
</s>
<s>
RAID-Z	O
is	O
a	O
data/parity	O
distribution	O
scheme	O
like	O
RAID-5	B-General_Concept
,	O
but	O
uses	O
dynamic	O
stripe	O
width	O
:	O
every	O
block	B-General_Concept
is	O
its	O
own	O
RAID	B-Architecture
stripe	O
,	O
regardless	O
of	O
blocksize	O
,	O
resulting	O
in	O
every	O
RAID-Z	O
write	O
being	O
a	O
full-stripe	O
write	O
.	O
</s>
<s>
This	O
,	O
when	O
combined	O
with	O
the	O
copy-on-write	B-Language
transactional	B-General_Concept
semantics	O
of	O
ZFS	B-Application
,	O
eliminates	O
the	O
write	O
hole	O
error	O
.	O
</s>
<s>
RAID-Z	O
is	O
also	O
faster	O
than	O
traditional	O
RAID5	O
because	O
it	O
does	O
not	O
need	O
to	O
perform	O
the	O
usual	O
read-modify-write	B-Operating_System
sequence	O
.	O
</s>
<s>
As	O
all	O
stripes	O
are	O
of	O
different	O
sizes	O
,	O
RAID-Z	O
reconstruction	O
has	O
to	O
traverse	O
the	O
filesystem	B-Application
metadata	O
to	O
determine	O
the	O
actual	O
RAID-Z	O
geometry	O
.	O
</s>
<s>
This	O
would	O
be	O
impossible	O
if	O
the	O
filesystem	B-Application
and	O
the	O
RAID	B-Architecture
array	I-Architecture
were	O
separate	O
products	O
,	O
whereas	O
it	O
becomes	O
feasible	O
when	O
there	O
is	O
an	O
integrated	O
view	O
of	O
the	O
logical	O
and	O
physical	O
structure	O
of	O
the	O
data	O
.	O
</s>
<s>
Going	O
through	O
the	O
metadata	O
means	O
that	O
ZFS	B-Application
can	O
validate	O
every	O
block	B-General_Concept
against	O
its	O
256-bit	O
checksum	B-Algorithm
as	O
it	O
goes	O
,	O
whereas	O
traditional	O
RAID	B-Architecture
products	O
usually	O
cannot	O
do	O
this	O
.	O
</s>
<s>
In	O
addition	O
to	O
handling	O
whole-disk	O
failures	O
,	O
RAID-Z	O
can	O
also	O
detect	O
and	O
correct	O
silent	O
data	O
corruption	O
,	O
offering	O
"	O
self-healing	O
data	O
"	O
:	O
when	O
reading	O
a	O
RAID-Z	O
block	B-General_Concept
,	O
ZFS	B-Application
compares	O
it	O
against	O
its	O
checksum	B-Algorithm
,	O
and	O
if	O
the	O
data	O
disks	O
did	O
not	O
return	O
the	O
right	O
answer	O
,	O
ZFS	B-Application
reads	O
the	O
parity	B-Error_Name
and	O
then	O
figures	O
out	O
which	O
disk	O
returned	O
bad	O
data	O
.	O
</s>
<s>
RAID-Z	O
and	O
mirroring	B-General_Concept
do	O
not	O
require	O
any	O
special	O
hardware	B-Architecture
:	O
they	O
do	O
not	O
need	O
NVRAM	O
for	O
reliability	O
,	O
and	O
they	O
do	O
not	O
need	O
write	O
buffering	O
for	O
good	O
performance	O
or	O
data	O
protection	O
.	O
</s>
<s>
With	O
RAID-Z	O
,	O
ZFS	B-Application
provides	O
fast	O
,	O
reliable	O
storage	O
using	O
cheap	O
,	O
commodity	O
disks	O
.	O
</s>
<s>
There	O
are	O
five	O
different	O
RAID-Z	O
modes	O
:	O
striping	B-General_Concept
(	O
similar	O
to	O
RAID0	O
,	O
offers	O
no	O
redundancy	B-Application
)	O
,	O
RAID-Z1	O
(	O
similar	O
to	O
RAID5	O
,	O
allows	O
one	O
disk	O
to	O
fail	O
)	O
,	O
RAID-Z2	O
(	O
similar	O
to	O
RAID6	O
,	O
allows	O
two	O
disks	O
to	O
fail	O
)	O
,	O
RAID-Z3	O
(	O
a	O
RAID7	O
configuration	O
,	O
allows	O
three	O
disks	O
to	O
fail	O
)	O
,	O
and	O
mirroring	B-General_Concept
(	O
similar	O
to	O
RAID1	O
,	O
allows	O
all	O
but	O
one	O
disk	O
to	O
fail	O
)	O
.	O
</s>
<s>
The	O
need	O
for	O
RAID-Z3	O
arose	O
in	O
the	O
early	O
2000s	O
as	O
multi-terabyte	O
capacity	O
drives	O
became	O
more	O
common	O
.	O
</s>
<s>
By	O
increasing	O
parity	B-Error_Name
,	O
RAID-Z3	O
reduces	O
the	O
chance	O
of	O
data	O
loss	O
by	O
simply	O
increasing	O
redundancy	B-Application
.	O
</s>
<s>
ZFS	B-Application
has	O
no	O
tool	O
equivalent	O
to	O
fsck	B-Device
(	O
the	O
standard	O
Unix	B-Application
and	O
Linux	B-Application
data	O
checking	O
and	O
repair	O
tool	O
for	O
file	B-Application
systems	I-Application
)	O
.	O
</s>
<s>
Instead	O
,	O
ZFS	B-Application
has	O
a	O
built-in	O
scrub	B-Error_Name
function	O
which	O
regularly	O
examines	O
all	O
data	O
and	O
repairs	O
silent	O
corruption	O
and	O
other	O
problems	O
.	O
</s>
<s>
fsck	B-Device
must	O
be	O
run	O
on	O
an	O
offline	O
filesystem	B-Application
,	O
which	O
means	O
the	O
filesystem	B-Application
must	O
be	O
unmounted	O
and	O
is	O
not	O
usable	O
while	O
being	O
repaired	O
,	O
while	O
scrub	B-Error_Name
is	O
designed	O
to	O
be	O
used	O
on	O
a	O
mounted	O
,	O
live	O
filesystem	B-Application
,	O
and	O
does	O
not	O
need	O
the	O
ZFS	B-Application
filesystem	B-Application
to	O
be	O
taken	O
offline	O
.	O
</s>
<s>
fsck	B-Device
usually	O
only	O
checks	O
metadata	O
(	O
such	O
as	O
the	O
journal	O
log	O
)	O
but	O
never	O
checks	O
the	O
data	O
itself	O
.	O
</s>
<s>
This	O
means	O
,	O
after	O
an	O
fsck	B-Device
,	O
the	O
data	O
might	O
still	O
not	O
match	O
the	O
original	O
data	O
as	O
stored	O
.	O
</s>
<s>
fsck	B-Device
cannot	O
always	O
validate	O
and	O
repair	O
data	O
when	O
checksums	B-Algorithm
are	O
stored	O
with	O
data	O
(	O
often	O
the	O
case	O
in	O
many	O
file	B-Application
systems	I-Application
)	O
,	O
because	O
the	O
checksums	B-Algorithm
may	O
also	O
be	O
corrupted	O
or	O
unreadable	O
.	O
</s>
<s>
ZFS	B-Application
always	O
stores	O
checksums	B-Algorithm
separately	O
from	O
the	O
data	O
they	O
verify	O
,	O
improving	O
reliability	O
and	O
the	O
ability	O
of	O
scrub	B-Error_Name
to	O
repair	O
the	O
volume	O
.	O
</s>
<s>
ZFS	B-Application
also	O
stores	O
multiple	O
copies	O
of	O
data	O
—	O
metadata	O
,	O
in	O
particular	O
,	O
may	O
have	O
upwards	O
of	O
4	O
or	O
6	O
copies	O
(	O
multiple	O
copies	O
per	O
disk	O
and	O
multiple	O
disk	B-General_Concept
mirrors	I-General_Concept
per	O
volume	O
)	O
,	O
greatly	O
improving	O
the	O
ability	O
of	O
scrub	B-Error_Name
to	O
detect	O
and	O
repair	O
extensive	O
damage	O
to	O
the	O
volume	O
,	O
compared	O
to	O
fsck	B-Device
.	O
</s>
<s>
scrub	B-Error_Name
checks	O
everything	O
,	O
including	O
metadata	O
and	O
the	O
data	O
.	O
</s>
<s>
The	O
effect	O
can	O
be	O
observed	O
by	O
comparing	O
fsck	B-Device
to	O
scrub	B-Error_Name
times	O
—	O
sometimes	O
a	O
fsck	B-Device
on	O
a	O
large	O
RAID	B-Architecture
completes	O
in	O
a	O
few	O
minutes	O
,	O
which	O
means	O
only	O
the	O
metadata	O
was	O
checked	O
.	O
</s>
<s>
Traversing	O
all	O
metadata	O
and	O
data	O
on	O
a	O
large	O
RAID	B-Architecture
takes	O
many	O
hours	O
,	O
which	O
is	O
exactly	O
what	O
scrub	B-Error_Name
does	O
.	O
</s>
<s>
while	O
fsck	B-Device
detects	O
and	O
tries	O
to	O
fix	O
errors	O
using	O
available	O
filesystem	B-Application
data	O
,	O
scrub	B-Error_Name
relies	O
on	O
redundancy	B-Application
to	O
recover	O
from	O
issues	O
.	O
</s>
<s>
While	O
fsck	B-Device
offers	O
to	O
fix	O
the	O
file	B-Application
system	I-Application
with	O
partial	O
data	O
loss	O
,	O
scrub	B-Error_Name
puts	O
it	O
into	O
faulted	O
state	O
if	O
there	O
is	O
no	O
redundancy	B-Application
.	O
</s>
<s>
The	O
official	O
recommendation	O
from	O
Sun/Oracle	O
is	O
to	O
scrub	B-Error_Name
enterprise-level	O
disks	O
once	O
a	O
month	O
,	O
and	O
cheaper	O
commodity	O
disks	O
once	O
a	O
week	O
.	O
</s>
<s>
ZFS	B-Application
is	O
a	O
128-bit	O
file	B-Application
system	I-Application
,	O
so	O
it	O
can	O
address	O
1.84	O
×	O
1019	O
times	O
more	O
data	O
than	O
64-bit	O
systems	O
such	O
as	O
Btrfs	B-Operating_System
.	O
</s>
<s>
The	O
maximum	O
limits	O
of	O
ZFS	B-Application
are	O
designed	O
to	O
be	O
so	O
large	O
that	O
they	O
should	O
never	O
be	O
encountered	O
in	O
practice	O
.	O
</s>
<s>
For	O
instance	O
,	O
fully	O
populating	O
a	O
single	O
zpool	O
with	O
2128	O
bits	O
of	O
data	O
would	O
require	O
3×1024TB	O
hard	B-Device
disk	I-Device
drives	I-Device
.	O
</s>
<s>
Some	O
theoretical	O
limits	O
in	O
ZFS	B-Application
are	O
:	O
</s>
<s>
With	O
Oracle	B-Application
Solaris	I-Application
,	O
the	O
encryption	O
capability	O
in	O
ZFS	B-Application
is	O
embedded	O
into	O
the	O
I/O	B-General_Concept
pipeline	O
.	O
</s>
<s>
During	O
writes	O
,	O
a	O
block	B-General_Concept
may	O
be	O
compressed	O
,	O
encrypted	O
,	O
checksummed	O
and	O
then	O
deduplicated	O
,	O
in	O
that	O
order	O
.	O
</s>
<s>
The	O
policy	O
for	O
encryption	O
is	O
set	O
at	O
the	O
dataset	O
level	O
when	O
datasets	O
(	O
file	B-Application
systems	I-Application
or	O
ZVOLs	O
)	O
are	O
created	O
.	O
</s>
<s>
The	O
wrapping	O
keys	O
provided	O
by	O
the	O
user/administrator	O
can	O
be	O
changed	O
at	O
any	O
time	O
without	O
taking	O
the	O
file	B-Application
system	I-Application
offline	O
.	O
</s>
<s>
Only	O
descendant	O
datasets	O
(	O
snapshots	B-Application
and	O
clones	O
)	O
share	O
data	O
encryption	O
keys	O
.	O
</s>
<s>
the	O
encryption	O
feature	O
is	O
also	O
fully	O
integrated	O
into	O
OpenZFS	B-Operating_System
0.8.0	O
available	O
for	O
Debian	O
and	O
Ubuntu	B-Operating_System
Linux	I-Operating_System
distributions	I-Operating_System
.	O
</s>
<s>
ZFS	B-Application
will	O
automatically	O
allocate	O
data	O
storage	O
across	O
all	O
vdevs	O
in	O
a	O
pool	O
(	O
and	O
all	O
devices	O
in	O
each	O
vdev	O
)	O
in	O
a	O
way	O
that	O
generally	O
maximises	O
the	O
performance	O
of	O
the	O
pool	O
.	O
</s>
<s>
ZFS	B-Application
will	O
also	O
update	O
its	O
write	O
strategy	O
to	O
take	O
account	O
of	O
new	O
disks	O
added	O
to	O
a	O
pool	O
,	O
when	O
they	O
are	O
added	O
.	O
</s>
<s>
As	O
a	O
general	O
rule	O
,	O
ZFS	B-Application
allocates	O
writes	O
across	O
vdevs	O
based	O
on	O
the	O
free	O
space	O
in	O
each	O
vdev	O
.	O
</s>
<s>
When	O
mirroring	B-General_Concept
,	O
block	B-General_Concept
devices	O
can	O
be	O
grouped	O
according	O
to	O
physical	O
chassis	O
,	O
so	O
that	O
the	O
filesystem	B-Application
can	O
continue	O
in	O
the	O
case	O
of	O
the	O
failure	O
of	O
an	O
entire	O
chassis	O
.	O
</s>
<s>
Storage	O
pool	O
composition	O
is	O
not	O
limited	O
to	O
similar	O
devices	O
,	O
but	O
can	O
consist	O
of	O
ad-hoc	O
,	O
heterogeneous	O
collections	O
of	O
devices	O
,	O
which	O
ZFS	B-Application
seamlessly	O
pools	O
together	O
,	O
subsequently	O
doling	O
out	O
space	O
to	O
as	O
needed	O
.	O
</s>
<s>
The	O
storage	O
capacity	O
of	O
all	O
vdevs	O
is	O
available	O
to	O
all	O
of	O
the	O
file	B-Application
system	I-Application
instances	O
in	O
the	O
zpool	O
.	O
</s>
<s>
A	O
quota	B-Application
can	O
be	O
set	O
to	O
limit	O
the	O
amount	O
of	O
space	O
a	O
file	B-Application
system	I-Application
instance	O
can	O
occupy	O
,	O
and	O
a	O
reservation	O
can	O
be	O
set	O
to	O
guarantee	O
that	O
space	O
will	O
be	O
available	O
to	O
a	O
file	B-Application
system	I-Application
instance	O
.	O
</s>
<s>
ZFS	B-Application
uses	O
different	O
layers	O
of	O
disk	O
cache	B-General_Concept
to	O
speed	O
up	O
read	O
and	O
write	O
operations	O
.	O
</s>
<s>
Ideally	O
,	O
all	O
data	O
should	O
be	O
stored	O
in	O
RAM	B-Architecture
,	O
but	O
that	O
is	O
usually	O
too	O
expensive	O
.	O
</s>
<s>
Frequently	O
accessed	O
data	O
will	O
be	O
stored	O
in	O
RAM	B-Architecture
,	O
and	O
less	O
frequently	O
accessed	O
data	O
can	O
be	O
stored	O
on	O
slower	O
media	O
,	O
such	O
as	O
solid	B-Device
state	I-Device
drives	I-Device
(	O
SSDs	B-Device
)	O
.	O
</s>
<s>
Data	O
that	O
is	O
not	O
often	O
accessed	O
is	O
not	O
cached	O
and	O
left	O
on	O
the	O
slow	O
hard	B-Device
drives	I-Device
.	O
</s>
<s>
If	O
old	O
data	O
is	O
suddenly	O
read	O
a	O
lot	O
,	O
ZFS	B-Application
will	O
automatically	O
move	O
it	O
to	O
SSDs	B-Device
or	O
to	O
RAM	B-Architecture
.	O
</s>
<s>
ZFS	B-Application
caching	B-General_Concept
mechanisms	O
include	O
one	O
each	O
for	O
reads	O
and	O
writes	O
,	O
and	O
in	O
each	O
case	O
,	O
two	O
levels	O
of	O
caching	B-General_Concept
can	O
exist	O
,	O
one	O
in	O
computer	O
memory	O
(	O
RAM	B-Architecture
)	O
and	O
one	O
on	O
fast	O
storage	O
(	O
usually	O
solid	B-Device
state	I-Device
drives	I-Device
(	O
SSDs	B-Device
)	O
)	O
,	O
for	O
a	O
total	O
of	O
four	O
caches	B-General_Concept
.	O
</s>
<s>
Where	O
stored	O
Read	O
cache	B-General_Concept
Write	O
cache	B-General_Concept
First	O
level	O
cache	B-General_Concept
In	O
RAM	B-Architecture
Known	O
as	O
ARC	O
,	O
due	O
to	O
its	O
use	O
of	O
a	O
variant	O
of	O
the	O
adaptive	B-General_Concept
replacement	I-General_Concept
cache	I-General_Concept
(	O
ARC	O
)	O
algorithm	O
.	O
</s>
<s>
RAM	B-Architecture
will	O
always	O
be	O
used	O
for	O
caching	B-General_Concept
,	O
thus	O
this	O
level	O
is	O
always	O
present	O
.	O
</s>
<s>
If	O
RAM	B-Architecture
is	O
too	O
small	O
there	O
will	O
hardly	O
be	O
any	O
ARC	O
at	O
all	O
;	O
in	O
this	O
case	O
,	O
ZFS	B-Application
always	O
needs	O
to	O
access	O
the	O
underlying	O
disks	O
,	O
which	O
impacts	O
performance	O
,	O
considerably	O
.	O
</s>
<s>
This	O
allows	O
writes	O
to	O
be	O
organized	O
more	O
efficiently	O
for	O
the	O
underlying	O
disks	O
at	O
the	O
risk	O
of	O
minor	O
data	O
loss	O
of	O
the	O
most	O
recent	O
transactions	O
upon	O
power	O
interruption	O
or	O
hardware	B-Architecture
fault	O
.	O
</s>
<s>
In	O
practice	O
the	O
power	O
loss	O
risk	O
is	O
avoided	O
by	O
ZFS	B-Application
write	O
journaling	B-Application
and	O
by	O
the	O
SLOG/ZIL	O
second	O
tier	O
write	O
cache	B-General_Concept
pool	O
(	O
see	O
below	O
)	O
,	O
so	O
writes	O
will	O
only	O
be	O
lost	O
if	O
a	O
write	O
failure	O
happens	O
at	O
the	O
same	O
time	O
as	O
a	O
total	O
loss	O
of	O
the	O
second	O
tier	O
SLOG	O
pool	O
,	O
and	O
then	O
only	O
when	O
settings	O
related	O
to	O
synchronous	O
writing	O
and	O
SLOG	O
use	O
are	O
set	O
in	O
a	O
way	O
that	O
would	O
allow	O
such	O
a	O
situation	O
to	O
arise	O
.	O
</s>
<s>
Second	O
level	O
cache	B-General_Concept
&	O
Intent	O
log	O
On	O
fast	O
storage	O
devices	O
(	O
which	O
can	O
be	O
added	O
or	O
removed	O
from	O
a	O
"	O
live	O
"	O
system	O
without	O
disruption	O
in	O
current	O
versions	O
of	O
ZFS	B-Application
,	O
although	O
not	O
always	O
in	O
older	O
versions	O
)	O
Known	O
as	O
L2ARC	B-Application
(	O
"	O
Level	O
2	O
ARC	O
"	O
)	O
,	O
optional	O
.	O
</s>
<s>
ZFS	B-Application
will	O
cache	B-General_Concept
as	O
much	O
data	O
in	O
L2ARC	B-Application
as	O
it	O
can	O
,	O
which	O
can	O
be	O
tens	O
or	O
hundreds	O
of	O
gigabytes	O
in	O
many	O
cases	O
.	O
</s>
<s>
L2ARC	B-Application
will	O
also	O
considerably	O
speed	O
up	O
deduplication	B-General_Concept
if	O
the	O
entire	O
deduplication	B-General_Concept
table	O
can	O
be	O
cached	O
in	O
L2ARC	B-Application
.	O
</s>
<s>
It	O
can	O
take	O
several	O
hours	O
to	O
fully	O
populate	O
the	O
L2ARC	B-Application
from	O
empty	O
(	O
before	O
ZFS	B-Application
has	O
decided	O
which	O
data	O
are	O
"	O
hot	O
"	O
and	O
should	O
be	O
cached	O
)	O
.	O
</s>
<s>
If	O
the	O
L2ARC	B-Application
device	O
is	O
lost	O
,	O
all	O
reads	O
will	O
go	O
out	O
to	O
the	O
disks	O
which	O
slows	O
down	O
performance	O
,	O
but	O
nothing	O
else	O
will	O
happen	O
(	O
no	O
data	O
will	O
be	O
lost	O
)	O
.	O
</s>
<s>
Known	O
as	O
SLOG	O
or	O
ZIL	O
(	O
"	O
ZFS	B-Application
Intent	O
Log	O
"	O
)	O
–	O
the	O
terms	O
are	O
often	O
used	O
incorrectly	O
.	O
</s>
<s>
A	O
SLOG	O
(	O
secondary	O
log	O
device	O
)	O
is	O
an	O
optional	O
dedicated	O
cache	B-General_Concept
on	O
a	O
separate	O
device	O
,	O
for	O
recording	O
writes	O
,	O
in	O
the	O
event	O
of	O
a	O
system	O
issue	O
.	O
</s>
<s>
If	O
an	O
SLOG	O
device	O
exists	O
,	O
it	O
will	O
be	O
used	O
for	O
the	O
ZFS	B-Application
Intent	O
Log	O
as	O
a	O
second	O
level	O
log	O
,	O
and	O
if	O
no	O
separate	O
cache	B-General_Concept
device	O
is	O
provided	O
,	O
the	O
ZIL	O
will	O
be	O
created	O
on	O
the	O
main	O
storage	O
devices	O
instead	O
.	O
</s>
<s>
Strictly	O
speaking	O
,	O
ZFS	B-Application
does	O
not	O
use	O
the	O
SLOG	O
device	O
to	O
cache	B-General_Concept
its	O
disk	O
writes	O
.	O
</s>
<s>
The	O
SLOG	O
device	O
allows	O
ZFS	B-Application
to	O
speedily	O
store	O
writes	O
and	O
quickly	O
report	O
them	O
as	O
written	O
,	O
even	O
for	O
storage	O
devices	O
such	O
as	O
HDDs	B-Device
that	O
are	O
much	O
slower	O
.	O
</s>
<s>
In	O
the	O
normal	O
course	O
of	O
activity	O
,	O
the	O
SLOG	O
is	O
never	O
referred	O
to	O
or	O
read	O
,	O
and	O
it	O
does	O
not	O
act	O
as	O
a	O
cache	B-General_Concept
;	O
its	O
purpose	O
is	O
to	O
safeguard	O
data	B-General_Concept
in	I-General_Concept
flight	I-General_Concept
during	O
the	O
few	O
seconds	O
taken	O
for	O
collation	O
and	O
"	O
writing	O
out	O
"	O
,	O
in	O
case	O
the	O
eventual	O
write	O
were	O
to	O
fail	O
.	O
</s>
<s>
If	O
the	O
write	O
eventually	O
fails	O
,	O
or	O
the	O
system	O
suffers	O
a	O
crash	O
or	O
fault	O
preventing	O
its	O
writing	O
,	O
then	O
ZFS	B-Application
can	O
identify	O
all	O
the	O
writes	O
that	O
it	O
has	O
confirmed	O
were	O
written	O
,	O
by	O
reading	O
back	O
the	O
SLOG	O
(	O
the	O
only	O
time	O
it	O
is	O
read	O
from	O
)	O
,	O
and	O
use	O
this	O
to	O
completely	O
repair	O
the	O
data	O
loss	O
.	O
</s>
<s>
This	O
becomes	O
crucial	O
if	O
a	O
large	O
number	O
of	O
synchronous	O
writes	O
take	O
place	O
(	O
such	O
as	O
with	O
ESXi	B-Operating_System
,	O
NFS	B-Protocol
and	O
some	O
databases	O
)	O
,	O
where	O
the	O
client	O
requires	O
confirmation	O
of	O
successful	O
writing	O
before	O
continuing	O
its	O
activity	O
;	O
the	O
SLOG	O
allows	O
ZFS	B-Application
to	O
confirm	O
writing	O
is	O
successful	O
much	O
more	O
quickly	O
than	O
if	O
it	O
had	O
to	O
write	O
to	O
the	O
main	O
store	O
every	O
time	O
,	O
without	O
the	O
risk	O
involved	O
in	O
misleading	O
the	O
client	O
as	O
to	O
the	O
state	O
of	O
data	O
storage	O
.	O
</s>
<s>
In	O
earlier	O
versions	O
of	O
ZFS	B-Application
,	O
loss	O
of	O
the	O
log	O
device	O
could	O
result	O
in	O
loss	O
of	O
the	O
entire	O
zpool	O
,	O
although	O
this	O
is	O
no	O
longer	O
the	O
case	O
.	O
</s>
<s>
Therefore	O
,	O
one	O
should	O
upgrade	O
ZFS	B-Application
if	O
planning	O
to	O
use	O
a	O
separate	O
log	O
device	O
.	O
</s>
<s>
A	O
number	O
of	O
other	O
caches	B-General_Concept
,	O
cache	B-General_Concept
divisions	O
,	O
and	O
queues	O
also	O
exist	O
within	O
ZFS	B-Application
.	O
</s>
<s>
For	O
example	O
,	O
each	O
VDEV	O
has	O
its	O
own	O
data	B-General_Concept
cache	I-General_Concept
,	O
and	O
the	O
ARC	O
cache	B-General_Concept
is	O
divided	O
between	O
data	O
stored	O
by	O
the	O
user	O
and	O
metadata	O
used	O
by	O
ZFS	B-Application
,	O
with	O
control	O
over	O
the	O
balance	O
between	O
these	O
.	O
</s>
<s>
In	O
OpenZFS	B-Operating_System
0.8	O
and	O
later	O
,	O
it	O
is	O
possible	O
to	O
configure	O
a	O
Special	O
VDEV	O
class	O
to	O
preferentially	O
store	O
filesystem	B-Application
metadata	O
,	O
and	O
optionally	O
the	O
Data	B-General_Concept
Deduplication	I-General_Concept
Table	O
(	O
DDT	O
)	O
,	O
and	O
small	O
filesystem	B-Application
blocks	O
.	O
</s>
<s>
This	O
speeds	O
up	O
metadata-intensive	O
operations	O
such	O
as	O
filesystem	B-Application
traversal	O
,	O
scrub	B-Error_Name
,	O
and	O
resilver	O
,	O
without	O
the	O
expense	O
of	O
storing	O
the	O
entire	O
filesystem	B-Application
on	O
solid-state	O
storage	O
.	O
</s>
<s>
ZFS	B-Application
uses	O
a	O
copy-on-write	B-Language
transactional	B-General_Concept
object	B-Application
model	I-Application
.	O
</s>
<s>
All	O
block	B-General_Concept
pointers	O
within	O
the	O
filesystem	B-Application
contain	O
a	O
256-bit	O
checksum	B-Algorithm
or	O
256-bit	O
hash	B-Algorithm
(	O
currently	O
a	O
choice	O
between	O
Fletcher-2	B-Algorithm
,	O
Fletcher-4	B-Algorithm
,	O
or	O
SHA-256	B-Algorithm
)	O
of	O
the	O
target	O
block	B-General_Concept
,	O
which	O
is	O
verified	O
when	O
the	O
block	B-General_Concept
is	O
read	O
.	O
</s>
<s>
Blocks	O
containing	O
active	O
data	O
are	O
never	O
overwritten	O
in	O
place	O
;	O
instead	O
,	O
a	O
new	O
block	B-General_Concept
is	O
allocated	O
,	O
modified	O
data	O
is	O
written	O
to	O
it	O
,	O
then	O
any	O
metadata	O
blocks	O
referencing	O
it	O
are	O
similarly	O
read	O
,	O
reallocated	O
,	O
and	O
written	O
.	O
</s>
<s>
To	O
reduce	O
the	O
overhead	O
of	O
this	O
process	O
,	O
multiple	O
updates	O
are	O
grouped	O
into	O
transaction	O
groups	O
,	O
and	O
ZIL	O
(	O
intent	O
log	O
)	O
write	O
cache	B-General_Concept
is	O
used	O
when	O
synchronous	O
write	O
semantics	O
are	O
required	O
.	O
</s>
<s>
The	O
blocks	O
are	O
arranged	O
in	O
a	O
tree	O
,	O
as	O
are	O
their	O
checksums	B-Algorithm
(	O
see	O
Merkle	O
signature	O
scheme	O
)	O
.	O
</s>
<s>
An	O
advantage	O
of	O
copy-on-write	B-Language
is	O
that	O
,	O
when	O
ZFS	B-Application
writes	O
new	O
data	O
,	O
the	O
blocks	O
containing	O
the	O
old	O
data	O
can	O
be	O
retained	O
,	O
allowing	O
a	O
snapshot	B-Application
version	O
of	O
the	O
file	B-Application
system	I-Application
to	O
be	O
maintained	O
.	O
</s>
<s>
ZFS	B-Application
snapshots	B-Application
are	O
consistent	O
(	O
they	O
reflect	O
the	O
entire	O
data	O
as	O
it	O
existed	O
at	O
a	O
single	O
point	O
in	O
time	O
)	O
,	O
and	O
can	O
be	O
created	O
extremely	O
quickly	O
,	O
since	O
all	O
the	O
data	O
composing	O
the	O
snapshot	B-Application
is	O
already	O
stored	O
,	O
with	O
the	O
entire	O
storage	O
pool	O
often	O
snapshotted	O
several	O
times	O
per	O
hour	O
.	O
</s>
<s>
They	O
are	O
also	O
space	O
efficient	O
,	O
since	O
any	O
unchanged	O
data	O
is	O
shared	O
among	O
the	O
file	B-Application
system	I-Application
and	O
its	O
snapshots	B-Application
.	O
</s>
<s>
Snapshots	B-Application
are	O
inherently	O
read-only	O
,	O
ensuring	O
they	O
will	O
not	O
be	O
modified	O
after	O
creation	O
,	O
although	O
they	O
should	O
not	O
be	O
relied	O
on	O
as	O
a	O
sole	O
means	O
of	O
backup	O
.	O
</s>
<s>
Entire	O
snapshots	B-Application
can	O
be	O
restored	O
and	O
also	O
files	O
and	O
directories	O
within	O
snapshots	B-Application
.	O
</s>
<s>
Writeable	O
snapshots	B-Application
(	O
"	O
clones	O
"	O
)	O
can	O
also	O
be	O
created	O
,	O
resulting	O
in	O
two	O
independent	O
file	B-Application
systems	I-Application
that	O
share	O
a	O
set	O
of	O
blocks	O
.	O
</s>
<s>
As	O
changes	O
are	O
made	O
to	O
any	O
of	O
the	O
clone	O
file	B-Application
systems	I-Application
,	O
new	O
data	O
blocks	O
are	O
created	O
to	O
reflect	O
those	O
changes	O
,	O
but	O
any	O
unchanged	O
blocks	O
continue	O
to	O
be	O
shared	O
,	O
no	O
matter	O
how	O
many	O
clones	O
exist	O
.	O
</s>
<s>
This	O
is	O
an	O
implementation	O
of	O
the	O
Copy-on-write	B-Language
principle	O
.	O
</s>
<s>
ZFS	B-Application
file	B-Application
systems	I-Application
can	O
be	O
moved	O
to	O
other	O
pools	O
,	O
also	O
on	O
remote	O
hosts	O
over	O
the	O
network	O
,	O
as	O
the	O
send	O
command	O
creates	O
a	O
stream	O
representation	O
of	O
the	O
file	B-Application
system	I-Application
's	O
state	O
.	O
</s>
<s>
This	O
stream	O
can	O
either	O
describe	O
complete	O
contents	O
of	O
the	O
file	B-Application
system	I-Application
at	O
a	O
given	O
snapshot	B-Application
,	O
or	O
it	O
can	O
be	O
a	O
delta	O
between	O
snapshots	B-Application
.	O
</s>
<s>
Computing	O
the	O
delta	O
stream	O
is	O
very	O
efficient	O
,	O
and	O
its	O
size	O
depends	O
on	O
the	O
number	O
of	O
blocks	O
changed	O
between	O
the	O
snapshots	B-Application
.	O
</s>
<s>
Dynamic	O
striping	B-General_Concept
across	O
all	O
devices	O
to	O
maximize	O
throughput	O
means	O
that	O
as	O
additional	O
devices	O
are	O
added	O
to	O
the	O
zpool	O
,	O
the	O
stripe	O
width	O
automatically	O
expands	O
to	O
include	O
them	O
;	O
thus	O
,	O
all	O
disks	O
in	O
a	O
pool	O
are	O
used	O
,	O
which	O
balances	O
the	O
write	O
load	O
across	O
them	O
.	O
</s>
<s>
ZFS	B-Application
uses	O
variable-sized	O
blocks	O
,	O
with	O
128KB	O
as	O
the	O
default	O
size	O
.	O
</s>
<s>
Available	O
features	O
allow	O
the	O
administrator	O
to	O
tune	O
the	O
maximum	O
block	B-General_Concept
size	I-General_Concept
which	O
is	O
used	O
,	O
as	O
certain	O
workloads	O
do	O
not	O
perform	O
well	O
with	O
large	O
blocks	O
.	O
</s>
<s>
If	O
data	B-General_Concept
compression	I-General_Concept
is	O
enabled	O
,	O
variable	O
block	B-General_Concept
sizes	I-General_Concept
are	O
used	O
.	O
</s>
<s>
If	O
a	O
block	B-General_Concept
can	O
be	O
compressed	O
to	O
fit	O
into	O
a	O
smaller	O
block	B-General_Concept
size	I-General_Concept
,	O
the	O
smaller	O
size	O
is	O
used	O
on	O
the	O
disk	O
to	O
use	O
less	O
storage	O
and	O
improve	O
IO	O
throughput	O
(	O
though	O
at	O
the	O
cost	O
of	O
increased	O
CPU	O
use	O
for	O
the	O
compression	O
and	O
decompression	O
operations	O
)	O
.	O
</s>
<s>
In	O
ZFS	B-Application
,	O
filesystem	B-Application
manipulation	O
within	O
a	O
storage	O
pool	O
is	O
easier	O
than	O
volume	O
manipulation	O
within	O
a	O
traditional	O
filesystem	B-Application
;	O
the	O
time	O
and	O
effort	O
required	O
to	O
create	O
or	O
expand	O
a	O
ZFS	B-Application
filesystem	B-Application
is	O
closer	O
to	O
that	O
of	O
making	O
a	O
new	O
directory	O
than	O
it	O
is	O
to	O
volume	O
manipulation	O
in	O
some	O
other	O
systems	O
.	O
</s>
<s>
Pools	O
and	O
their	O
associated	O
ZFS	B-Application
file	B-Application
systems	I-Application
can	O
be	O
moved	O
between	O
different	O
platform	O
architectures	O
,	O
including	O
systems	O
implementing	O
different	O
byte	O
orders	O
.	O
</s>
<s>
The	O
ZFS	B-Application
block	B-General_Concept
pointer	O
format	O
stores	O
filesystem	B-Application
metadata	O
in	O
an	O
endian-adaptive	O
way	O
;	O
individual	O
metadata	O
blocks	O
are	O
written	O
with	O
the	O
native	O
byte	O
order	O
of	O
the	O
system	O
writing	O
the	O
block	B-General_Concept
.	O
</s>
<s>
Data	B-General_Concept
deduplication	I-General_Concept
capabilities	O
were	O
added	O
to	O
the	O
ZFS	B-Application
source	O
repository	O
at	O
the	O
end	O
of	O
October	O
2009	O
,	O
and	O
relevant	O
OpenSolaris	B-Operating_System
ZFS	B-Application
development	O
packages	O
have	O
been	O
available	O
since	O
December	O
3	O
,	O
2009	O
(	O
build	O
128	O
)	O
.	O
</s>
<s>
Effective	O
use	O
of	O
deduplication	B-General_Concept
may	O
require	O
large	O
RAM	B-Architecture
capacity	O
;	O
recommendations	O
range	O
between	O
1	O
and	O
5GB	O
of	O
RAM	B-Architecture
for	O
every	O
TB	O
of	O
storage	O
.	O
</s>
<s>
An	O
accurate	O
assessment	O
of	O
the	O
memory	O
required	O
for	O
deduplication	B-General_Concept
is	O
made	O
by	O
referring	O
to	O
the	O
number	O
of	O
unique	O
blocks	O
in	O
the	O
pool	O
,	O
and	O
the	O
number	O
of	O
bytes	O
on	O
disk	O
and	O
in	O
RAM	B-Architecture
(	O
"	O
core	O
"	O
)	O
required	O
to	O
store	O
each	O
record	O
—	O
these	O
figures	O
are	O
reported	O
by	O
inbuilt	O
commands	O
such	O
as	O
zpool	O
and	O
zdb	O
.	O
</s>
<s>
Insufficient	O
physical	O
memory	O
or	O
lack	O
of	O
ZFS	B-Application
cache	B-General_Concept
can	O
result	O
in	O
virtual	O
memory	B-General_Concept
thrashing	I-General_Concept
when	O
using	O
deduplication	B-General_Concept
,	O
which	O
can	O
cause	O
performance	O
to	O
plummet	O
,	O
or	O
result	O
in	O
complete	O
memory	O
starvation	O
.	O
</s>
<s>
Because	O
deduplication	B-General_Concept
occurs	O
at	O
write-time	O
,	O
it	O
is	O
also	O
very	O
CPU-intensive	O
and	O
this	O
can	O
also	O
significantly	O
slow	O
down	O
a	O
system	O
.	O
</s>
<s>
Other	O
storage	O
vendors	O
use	O
modified	O
versions	O
of	O
ZFS	B-Application
to	O
achieve	O
very	O
high	O
data	B-General_Concept
compression	I-General_Concept
ratios	I-General_Concept
.	O
</s>
<s>
In	O
May	O
2014	O
,	O
Oracle	B-Application
bought	O
GreenBytes	O
for	O
its	O
ZFS	B-Application
deduplication	B-General_Concept
and	O
replication	B-Architecture
technology	O
.	O
</s>
<s>
As	O
described	O
above	O
,	O
deduplication	B-General_Concept
is	O
usually	O
not	O
recommended	O
due	O
to	O
its	O
heavy	O
resource	O
requirements	O
(	O
especially	O
RAM	B-Architecture
)	O
and	O
impact	O
on	O
performance	O
(	O
especially	O
when	O
writing	O
)	O
,	O
other	O
than	O
in	O
specific	O
circumstances	O
where	O
the	O
system	O
and	O
data	O
are	O
well-suited	O
to	O
this	O
space-saving	O
technique	O
.	O
</s>
<s>
Explicit	O
I/O	B-General_Concept
priority	O
with	O
deadline	O
scheduling	O
.	O
</s>
<s>
Claimed	O
globally	O
optimal	O
I/O	B-General_Concept
sorting	O
and	O
aggregation	O
.	O
</s>
<s>
End-to-end	O
checksumming	O
,	O
using	O
a	O
kind	O
of	O
"	O
Data	B-Error_Name
Integrity	I-Error_Name
Field	I-Error_Name
"	O
,	O
allowing	O
data	O
corruption	O
detection	O
(	O
and	O
recovery	O
if	O
you	O
have	O
redundancy	B-Application
in	O
the	O
pool	O
)	O
.	O
</s>
<s>
A	O
choice	O
of	O
3	O
hashes	B-Algorithm
can	O
be	O
used	O
,	O
optimized	O
for	O
speed	O
(	O
fletcher	O
)	O
,	O
standardization	O
and	O
security	O
(	O
SHA256	B-Algorithm
)	O
and	O
salted	O
hashes	B-Algorithm
(	O
Skein	B-Algorithm
)	O
.	O
</s>
<s>
Transparent	O
filesystem	B-Application
compression	O
.	O
</s>
<s>
Supports	O
LZJB	B-Application
,	O
gzip	B-Application
,	O
LZ4	B-Language
and	O
Zstd	B-Language
.	O
</s>
<s>
Intelligent	O
scrubbing	B-Error_Name
and	O
resilvering	B-General_Concept
(	O
resyncing	O
)	O
.	O
</s>
<s>
Ditto	O
blocks	O
:	O
Configurable	O
data	B-Architecture
replication	I-Architecture
per	O
filesystem	B-Application
,	O
with	O
zero	O
,	O
one	O
or	O
two	O
extra	O
copies	O
requested	O
per	O
write	O
for	O
user	O
data	O
,	O
and	O
with	O
that	O
same	O
base	O
number	O
of	O
copies	O
plus	O
one	O
or	O
two	O
for	O
metadata	O
(	O
according	O
to	O
metadata	O
importance	O
)	O
.	O
</s>
<s>
If	O
the	O
pool	O
has	O
several	O
devices	O
,	O
ZFS	B-Application
tries	O
to	O
replicate	O
over	O
different	O
devices	O
.	O
</s>
<s>
ZFS	B-Application
design	O
(	O
copy-on-write	B-Language
+	O
superblocks	O
)	O
is	O
safe	O
when	O
using	O
disks	O
with	O
write	O
cache	B-General_Concept
enabled	O
,	O
if	O
they	O
honor	O
the	O
write	O
barriers	O
.	O
</s>
<s>
This	O
feature	O
provides	O
safety	O
and	O
a	O
performance	O
boost	O
compared	O
with	O
some	O
other	O
filesystems	B-Application
.	O
</s>
<s>
On	O
Solaris	B-Application
,	O
when	O
entire	O
disks	O
are	O
added	O
to	O
a	O
ZFS	B-Application
pool	O
,	O
ZFS	B-Application
automatically	O
enables	O
their	O
write	O
cache	B-General_Concept
.	O
</s>
<s>
This	O
is	O
not	O
done	O
when	O
ZFS	B-Application
only	O
manages	O
discrete	O
slices	O
of	O
the	O
disk	O
,	O
since	O
it	O
does	O
not	O
know	O
if	O
other	O
slices	O
are	O
managed	O
by	O
non-write-cache	O
safe	O
filesystems	B-Application
,	O
like	O
UFS	B-Operating_System
.	O
</s>
<s>
The	O
FreeBSD	B-Operating_System
implementation	O
can	O
handle	O
disk	O
flushes	O
for	O
partitions	O
thanks	O
to	O
its	O
GEOM	B-Application
framework	O
,	O
and	O
therefore	O
does	O
not	O
suffer	O
from	O
this	O
limitation	O
.	O
</s>
<s>
Per-user	O
,	O
per-group	O
,	O
per-project	O
,	O
and	O
per-dataset	O
quota	B-Application
limits	O
.	O
</s>
<s>
Filesystem	B-Application
encryption	O
since	O
Solaris	B-Application
11	I-Application
Express	I-Application
,	O
and	O
OpenZFS	B-Operating_System
(	O
ZoL	O
)	O
0.8	O
.	O
</s>
<s>
(	O
on	O
some	O
other	O
systems	O
ZFS	B-Application
can	O
utilize	O
encrypted	O
disks	O
for	O
a	O
similar	O
effect	O
;	O
GELI	B-Application
on	O
FreeBSD	B-Operating_System
can	O
be	O
used	O
this	O
way	O
to	O
create	O
fully	O
encrypted	O
ZFS	B-Application
storage	O
)	O
.	O
</s>
<s>
ZFS	B-Application
is	O
not	O
a	O
clustered	B-Application
filesystem	I-Application
;	O
however	O
,	O
clustered	O
ZFS	B-Application
is	O
available	O
from	O
third	O
parties	O
.	O
</s>
<s>
Snapshots	B-Application
can	O
be	O
taken	O
manually	O
or	O
automatically	O
.	O
</s>
<s>
The	O
older	O
versions	O
of	O
the	O
stored	O
data	O
that	O
they	O
contain	O
can	O
be	O
exposed	O
as	O
full	O
read-only	O
file	B-Application
systems	I-Application
.	O
</s>
<s>
They	O
can	O
also	O
be	O
exposed	O
as	O
historic	O
versions	O
of	O
files	O
and	O
folders	O
when	O
used	O
with	O
CIFS	B-Operating_System
(	O
also	O
known	O
as	O
SMB	B-Operating_System
,	I-Operating_System
Samba	I-Operating_System
or	O
file	O
shares	O
)	O
;	O
this	O
is	O
known	O
as	O
"	O
Previous	O
versions	O
"	O
,	O
"	O
VSS	O
shadow	O
copies	O
"	O
,	O
or	O
"	O
File	O
history	O
"	O
on	O
Windows	O
,	O
or	O
AFP	B-Protocol
and	O
"	O
Apple	O
Time	O
Machine	O
"	O
on	O
Apple	O
devices	O
.	O
</s>
<s>
There	O
are	O
several	O
limitations	O
of	O
the	O
ZFS	B-Application
filesystem	B-Application
.	O
</s>
<s>
The	O
authors	O
of	O
a	O
2010	O
study	O
that	O
examined	O
the	O
ability	O
of	O
file	B-Application
systems	I-Application
to	O
detect	O
and	O
prevent	O
data	O
corruption	O
,	O
with	O
particular	O
focus	O
on	O
ZFS	B-Application
,	O
observed	O
that	O
ZFS	B-Application
itself	O
is	O
effective	O
in	O
detecting	O
and	O
correcting	O
data	O
errors	O
on	O
storage	O
devices	O
,	O
but	O
that	O
it	O
assumes	O
data	O
in	O
RAM	B-Architecture
is	O
"	O
safe	O
"	O
,	O
and	O
not	O
prone	O
to	O
error	O
.	O
</s>
<s>
The	O
study	O
comments	O
that	O
"	O
a	O
single	O
bit	O
flip	O
in	O
memory	O
causes	O
a	O
small	O
but	O
non-negligible	O
percentage	O
of	O
runs	O
to	O
experience	O
a	O
failure	O
,	O
with	O
the	O
probability	O
of	O
committing	O
bad	O
data	O
to	O
disk	O
varying	O
from	O
0%	O
to	O
3.6	O
%	O
(	O
according	O
to	O
the	O
workload	O
)	O
"	O
,	O
and	O
that	O
when	O
ZFS	B-Application
caches	B-General_Concept
pages	O
or	O
stores	O
copies	O
of	O
metadata	O
in	O
RAM	B-Architecture
,	O
or	O
holds	O
data	O
in	O
its	O
"	O
dirty	O
"	O
cache	B-General_Concept
for	O
writing	O
to	O
disk	O
,	O
no	O
test	O
is	O
made	O
whether	O
the	O
checksums	B-Algorithm
still	O
match	O
the	O
data	O
at	O
the	O
point	O
of	O
use	O
.	O
</s>
<s>
According	O
to	O
the	O
authors	O
,	O
by	O
using	O
ECC	B-General_Concept
RAM	I-General_Concept
;	O
however	O
,	O
the	O
authors	O
considered	O
that	O
adding	O
error	B-Error_Name
detection	I-Error_Name
related	O
to	O
the	O
page	B-General_Concept
cache	I-General_Concept
and	O
heap	O
would	O
allow	O
ZFS	B-Application
to	O
handle	O
certain	O
classes	O
of	O
error	O
more	O
robustly	O
.	O
</s>
<s>
One	O
of	O
the	O
main	O
architects	O
of	O
ZFS	B-Application
,	O
Matt	O
Ahrens	O
,	O
explains	O
there	O
is	O
an	O
option	O
to	O
enable	O
checksumming	O
of	O
data	O
in	O
memory	O
by	O
using	O
the	O
ZFS_DEBUG_MODIFY	O
flag	O
(	O
zfs_flags	O
=	O
0x10	O
)	O
which	O
addresses	O
these	O
concerns	O
.	O
</s>
<s>
Capacity	O
expansion	O
is	O
normally	O
achieved	O
by	O
adding	O
groups	O
of	O
disks	O
as	O
a	O
top-level	O
vdev	O
:	O
simple	O
device	O
,	O
RAID-Z	O
,	O
RAIDZ2	O
,	O
RAIDZ3	O
,	O
or	O
mirrored	O
.	O
</s>
<s>
It	O
is	O
also	O
possible	O
to	O
expand	O
the	O
array	O
by	O
iteratively	O
swapping	O
each	O
drive	O
in	O
the	O
array	O
with	O
a	O
bigger	O
drive	O
and	O
waiting	O
for	O
ZFS	B-Application
to	O
self-heal	O
;	O
the	O
heal	O
time	O
will	O
depend	O
on	O
the	O
amount	O
of	O
stored	O
information	O
,	O
not	O
the	O
disk	O
size	O
.	O
</s>
<s>
As	O
of	O
Solaris	B-Application
10	I-Application
Update	O
11	O
and	O
Solaris	O
11.2	O
,	O
it	O
was	O
neither	O
possible	O
to	O
reduce	O
the	O
number	O
of	O
top-level	O
vdevs	O
in	O
a	O
pool	O
except	O
hot	O
spares	O
,	O
cache	B-General_Concept
,	O
and	O
log	O
devices	O
,	O
nor	O
to	O
otherwise	O
reduce	O
pool	O
capacity	O
.	O
</s>
<s>
Enhancements	O
to	O
allow	O
reduction	O
of	O
vdevs	O
is	O
under	O
development	O
in	O
OpenZFS	B-Operating_System
.	O
</s>
<s>
Online	O
shrinking	O
by	O
removing	O
non-redundant	O
top-level	O
vdevs	O
is	O
supported	O
since	O
Solaris	O
11.4	O
released	O
in	O
August	O
2018	O
and	O
OpenZFS	B-Operating_System
(	O
ZoL	O
)	O
0.8	O
released	O
May	O
2019	O
.	O
</s>
<s>
Some	O
traditional	O
nested	O
RAID	B-Architecture
configurations	O
,	O
such	O
as	O
RAID51	O
(	O
a	O
mirror	O
of	O
RAID5	O
groups	O
)	O
,	O
are	O
not	O
configurable	O
in	O
ZFS	B-Application
,	O
without	O
some	O
3rd-party	O
tools	O
.	O
</s>
<s>
Vdevs	O
can	O
only	O
be	O
composed	O
of	O
raw	O
disks	O
or	O
files	O
,	O
not	O
other	O
vdevs	O
,	O
using	O
the	O
default	O
ZFS	B-Application
management	O
commands	O
.	O
</s>
<s>
However	O
,	O
a	O
ZFS	B-Application
pool	O
effectively	O
creates	O
a	O
stripe	O
(	O
RAID0	O
)	O
across	O
its	O
vdevs	O
,	O
so	O
the	O
equivalent	O
of	O
a	O
RAID50	O
or	O
RAID60	O
is	O
common	O
.	O
</s>
<s>
Reconfiguring	O
the	O
number	O
of	O
devices	O
in	O
a	O
top-level	O
vdev	O
requires	O
copying	O
data	O
offline	O
,	O
destroying	O
the	O
pool	O
,	O
and	O
recreating	O
the	O
pool	O
with	O
the	O
new	O
top-level	O
vdev	O
configuration	O
,	O
except	O
for	O
adding	O
extra	O
redundancy	B-Application
to	O
an	O
existing	O
mirror	O
,	O
which	O
can	O
be	O
done	O
at	O
any	O
time	O
or	O
if	O
all	O
top	O
level	O
vdevs	O
are	O
mirrors	O
with	O
sufficient	O
redundancy	B-Application
the	O
zpool	O
split	O
command	O
can	O
be	O
used	O
to	O
remove	O
a	O
vdev	O
from	O
each	O
top	O
level	O
vdev	O
in	O
the	O
pool	O
,	O
creating	O
a	O
2nd	O
pool	O
with	O
identical	O
data	O
.	O
</s>
<s>
IOPS	O
performance	O
of	O
a	O
ZFS	B-Application
storage	O
pool	O
can	O
suffer	O
if	O
the	O
ZFS	B-Application
raid	B-Architecture
is	O
not	O
appropriately	O
configured	O
.	O
</s>
<s>
This	O
applies	O
to	O
all	O
types	O
of	O
RAID	B-Architecture
,	O
in	O
one	O
way	O
or	O
another	O
.	O
</s>
<s>
However	O
,	O
there	O
are	O
ways	O
to	O
mitigate	O
this	O
IOPS	O
performance	O
problem	O
,	O
for	O
instance	O
add	O
SSDs	B-Device
as	O
L2ARC	B-Application
cache	B-General_Concept
—	O
which	O
can	O
boost	O
IOPS	O
into	O
100,000	O
s	O
.	O
</s>
<s>
Resilver	O
(	O
repair	O
)	O
of	O
a	O
failed	O
disk	O
in	O
a	O
ZFS	B-Application
RAID	B-Architecture
can	O
take	O
a	O
long	O
time	O
which	O
is	O
not	O
unique	O
to	O
ZFS	B-Application
,	O
it	O
applies	O
to	O
all	O
types	O
of	O
RAID	B-Architecture
,	O
in	O
one	O
way	O
or	O
another	O
.	O
</s>
<s>
This	O
means	O
that	O
very	O
large	O
volumes	O
can	O
take	O
several	O
days	O
to	O
repair	O
or	O
to	O
being	O
back	O
to	O
full	O
redundancy	B-Application
after	O
severe	O
data	O
corruption	O
or	O
failure	O
,	O
and	O
during	O
this	O
time	O
a	O
second	O
disk	O
failure	O
may	O
occur	O
,	O
especially	O
as	O
the	O
repair	O
puts	O
additional	O
stress	O
on	O
the	O
system	O
as	O
a	O
whole	O
.	O
</s>
<s>
ZFS	B-Application
RAID	B-Architecture
differs	O
from	O
conventional	O
RAID	B-Architecture
by	O
only	O
reconstructing	O
live	O
data	O
and	O
metadata	O
when	O
replacing	O
a	O
disk	O
,	O
not	O
the	O
entirety	O
of	O
the	O
disk	O
including	O
blank	O
and	O
garbage	O
blocks	O
,	O
which	O
means	O
that	O
replacing	O
a	O
member	O
disk	O
on	O
a	O
ZFS	B-Application
pool	O
that	O
is	O
only	O
partially	O
full	O
will	O
take	O
proportionally	O
less	O
time	O
compared	O
to	O
conventional	O
RAID	B-Architecture
.	O
</s>
<s>
Historically	O
,	O
ZFS	B-Application
has	O
not	O
shipped	O
with	O
tools	O
such	O
as	O
fsck	B-Device
to	O
repair	O
damaged	O
file	B-Application
systems	I-Application
,	O
because	O
the	O
file	B-Application
system	I-Application
itself	O
was	O
designed	O
to	O
self-repair	O
,	O
so	O
long	O
as	O
it	O
had	O
been	O
built	O
with	O
sufficient	O
attention	O
to	O
the	O
design	O
of	O
storage	O
and	O
redundancy	B-Application
of	O
data	O
.	O
</s>
<s>
If	O
the	O
pool	O
was	O
compromised	O
because	O
of	O
poor	O
hardware	B-Architecture
,	O
inadequate	O
design	O
or	O
redundancy	B-Application
,	O
or	O
unfortunate	O
mishap	O
,	O
to	O
the	O
point	O
that	O
ZFS	B-Application
was	O
unable	O
to	O
mount	B-Application
the	O
pool	O
,	O
traditionally	O
there	O
were	O
no	O
tools	O
which	O
allowed	O
an	O
end-user	O
to	O
attempt	O
partial	O
salvage	O
of	O
the	O
stored	O
data	O
.	O
</s>
<s>
This	O
led	O
to	O
threads	O
in	O
online	O
forums	O
where	O
ZFS	B-Application
developers	O
sometimes	O
tried	O
to	O
provide	O
ad-hoc	O
help	O
to	O
home	O
and	O
other	O
small	O
scale	O
users	O
,	O
facing	O
loss	O
of	O
data	O
due	O
to	O
their	O
inadequate	O
design	O
or	O
poor	O
system	O
management	O
.	O
</s>
<s>
Modern	O
ZFS	B-Application
has	O
improved	O
considerably	O
on	O
this	O
situation	O
over	O
time	O
,	O
and	O
continues	O
to	O
do	O
so	O
:	O
</s>
<s>
Removal	O
or	O
abrupt	O
failure	O
of	O
caching	B-General_Concept
devices	O
no	O
longer	O
causes	O
pool	O
loss	O
.	O
</s>
<s>
Loss	O
of	O
the	O
L2ARC	B-Application
cache	B-General_Concept
does	O
not	O
affect	O
data	O
.	O
)	O
</s>
<s>
If	O
the	O
pool	O
is	O
unmountable	O
,	O
modern	O
versions	O
of	O
ZFS	B-Application
will	O
attempt	O
to	O
identify	O
the	O
most	O
recent	O
consistent	O
point	O
at	O
which	O
the	O
pool	O
can	O
be	O
recovered	O
,	O
at	O
the	O
cost	O
of	O
losing	O
some	O
of	O
the	O
most	O
recent	O
changes	O
to	O
the	O
contents	O
.	O
</s>
<s>
Copy	B-Language
on	I-Language
write	I-Language
means	O
that	O
older	O
versions	O
of	O
data	O
,	O
including	O
top-level	O
records	O
and	O
metadata	O
,	O
may	O
still	O
exist	O
even	O
though	O
they	O
are	O
superseded	O
,	O
and	O
if	O
so	O
,	O
the	O
pool	O
can	O
be	O
wound	O
back	O
to	O
a	O
consistent	O
state	O
based	O
on	O
them	O
.	O
</s>
<s>
Informally	O
,	O
tools	O
exist	O
to	O
probe	O
the	O
reason	O
why	O
ZFS	B-Application
is	O
unable	O
to	O
mount	B-Application
a	O
pool	O
,	O
and	O
guide	O
the	O
user	O
or	O
a	O
developer	O
as	O
to	O
manual	O
changes	O
required	O
to	O
force	O
the	O
pool	O
to	O
mount	B-Application
.	O
</s>
<s>
These	O
include	O
using	O
zdb	O
(	O
ZFS	B-Application
debug	O
)	O
to	O
find	O
a	O
valid	O
importable	O
point	O
in	O
the	O
pool	O
,	O
using	O
dtrace	B-Application
or	O
similar	O
to	O
identify	O
the	O
issue	O
causing	O
mount	B-Application
failure	O
,	O
or	O
manually	O
bypassing	O
health	O
checks	O
that	O
cause	O
the	O
mount	B-Application
process	O
to	O
abort	O
,	O
and	O
allow	O
mounting	B-Application
of	O
the	O
damaged	O
pool	O
.	O
</s>
<s>
,	O
a	O
range	O
of	O
significantly	O
enhanced	O
methods	O
are	O
gradually	O
being	O
rolled	O
out	O
within	O
OpenZFS	B-Operating_System
.	O
</s>
<s>
Code	O
refactoring	O
,	O
and	O
more	O
detailed	O
diagnostic	O
and	O
debug	O
information	O
on	O
mount	B-Application
failures	O
,	O
to	O
simplify	O
diagnosis	O
and	O
fixing	O
of	O
corrupt	O
pool	O
issues	O
;	O
</s>
<s>
Oracle	B-Application
Corporation	I-Application
ceased	O
the	O
public	O
development	O
of	O
both	O
ZFS	B-Application
and	O
OpenSolaris	B-Operating_System
after	O
the	O
acquisition	O
of	O
Sun	O
in	O
2010	O
.	O
</s>
<s>
Some	O
developers	O
forked	B-Application
the	O
last	O
public	O
release	O
of	O
OpenSolaris	B-Operating_System
as	O
the	O
Illumos	B-Application
project	O
.	O
</s>
<s>
Because	O
of	O
the	O
significant	O
advantages	O
present	O
in	O
ZFS	B-Application
,	O
it	O
has	O
been	O
ported	O
to	O
several	O
different	O
platforms	O
with	O
different	O
features	O
and	O
commands	O
.	O
</s>
<s>
For	O
coordinating	O
the	O
development	O
efforts	O
and	O
to	O
avoid	O
fragmentation	O
,	O
OpenZFS	B-Operating_System
was	O
founded	O
in	O
2013	O
.	O
</s>
<s>
According	O
to	O
Matt	O
Ahrens	O
,	O
one	O
of	O
the	O
main	O
architects	O
of	O
ZFS	B-Application
,	O
over	O
50%	O
of	O
the	O
original	O
OpenSolaris	B-Operating_System
ZFS	B-Application
code	O
has	O
been	O
replaced	O
in	O
OpenZFS	B-Operating_System
with	O
community	O
contributions	O
as	O
of	O
2019	O
,	O
making	O
“	O
Oracle	B-Application
ZFS	I-Application
”	O
and	O
“	O
OpenZFS	B-Operating_System
”	O
politically	O
and	O
technologically	O
incompatible	O
.	O
</s>
<s>
2008	O
:	O
Sun	O
shipped	O
a	O
line	O
of	O
ZFS-based	O
7000-series	O
storage	O
appliances	O
.	O
</s>
<s>
2013	O
:	O
Oracle	B-Application
shipped	O
ZS3	O
series	O
of	O
ZFS-based	O
filers	O
and	O
seized	O
first	O
place	O
in	O
the	O
SPC-2	O
benchmark	O
with	O
one	O
of	O
them	O
.	O
</s>
<s>
2013	O
:	O
iXsystems	O
ships	O
ZFS-based	O
NAS	O
devices	O
called	O
FreeNAS	B-Application
,	O
(	O
now	O
named	O
TrueNAS	B-Application
CORE	O
)	O
,	O
for	O
SOHO	O
and	O
TrueNAS	B-Application
for	O
the	O
enterprise	O
.	O
</s>
<s>
2014	O
:	O
Netgear	O
ships	O
a	O
line	O
of	O
ZFS-based	O
NAS	O
devices	O
called	O
ReadyDATA	O
,	O
designed	O
to	O
be	O
used	O
in	O
the	O
enterprise	O
.	O
</s>
<s>
2015	O
:	O
rsync.net	O
announces	O
a	O
cloud	O
storage	O
platform	O
that	O
allows	O
customers	O
to	O
provision	O
their	O
own	O
zpool	O
and	O
import	O
and	O
export	O
data	O
using	O
zfs	B-Application
send	O
and	O
zfs	B-Application
receive	O
.	O
</s>
<s>
2020	O
:	O
iXsystems	O
Begins	O
development	O
of	O
a	O
ZFS-based	O
hyperconverged	O
software	O
called	O
TrueNAS	B-Application
SCALE	O
,	O
for	O
SOHO	O
and	O
TrueNAS	B-Application
for	O
the	O
enterprise	O
.	O
</s>
<s>
In	O
January	O
2010	O
,	O
Oracle	B-Application
Corporation	I-Application
acquired	O
Sun	O
Microsystems	O
,	O
and	O
quickly	O
discontinued	O
the	O
OpenSolaris	B-Operating_System
distribution	O
and	O
the	O
open	B-Application
source	I-Application
development	O
model	O
.	O
</s>
<s>
In	O
August	O
2010	O
,	O
Oracle	B-Application
discontinued	O
providing	O
public	O
updates	O
to	O
the	O
source	O
code	O
of	O
the	O
Solaris	O
OS/Networking	O
repository	O
,	O
effectively	O
turning	O
Solaris	B-Application
11	I-Application
back	O
into	O
a	O
closed	B-Application
source	I-Application
proprietary	B-Application
operating	B-General_Concept
system	I-General_Concept
.	O
</s>
<s>
In	O
response	O
to	O
the	O
changing	O
landscape	O
of	O
Solaris	B-Application
and	O
OpenSolaris	B-Operating_System
,	O
the	O
illumos	B-Application
project	O
was	O
launched	O
via	O
webinar	B-Protocol
on	O
Thursday	O
,	O
August	O
3	O
,	O
2010	O
,	O
as	O
a	O
community	O
effort	O
of	O
some	O
core	O
Solaris	B-Application
engineers	O
to	O
continue	O
developing	O
the	O
open	B-Application
source	I-Application
version	O
of	O
Solaris	B-Application
,	O
and	O
complete	O
the	O
open	O
sourcing	O
of	O
those	O
parts	O
not	O
already	O
open	O
sourced	O
by	O
Sun	O
.	O
</s>
<s>
illumos	B-Application
was	O
founded	O
as	O
a	O
Foundation	O
,	O
the	O
illumos	B-Application
Foundation	O
,	O
incorporated	O
in	O
the	O
State	O
of	O
California	O
as	O
a	O
501(c )	O
6	O
trade	O
association	O
.	O
</s>
<s>
The	O
original	O
plan	O
explicitly	O
stated	O
that	O
illumos	B-Application
would	O
not	O
be	O
a	O
distribution	O
or	O
a	O
fork	B-Application
.	O
</s>
<s>
However	O
,	O
after	O
Oracle	B-Application
announced	O
discontinuing	O
OpenSolaris	B-Operating_System
,	O
plans	O
were	O
made	O
to	O
fork	B-Application
the	O
final	O
version	O
of	O
the	O
Solaris	B-Application
ON	O
,	O
allowing	O
illumos	B-Application
to	O
evolve	O
into	O
an	O
operating	B-General_Concept
system	I-General_Concept
of	O
its	O
own	O
.	O
</s>
<s>
As	O
part	O
of	O
OpenSolaris	B-Operating_System
,	O
an	O
open	B-Application
source	I-Application
version	O
of	O
ZFS	B-Application
was	O
therefore	O
integral	O
within	O
illumos	B-Application
.	O
</s>
<s>
ZFS	B-Application
was	O
widely	O
used	O
within	O
numerous	O
platforms	O
,	O
as	O
well	O
as	O
Solaris	B-Application
.	O
</s>
<s>
Therefore	O
,	O
in	O
2013	O
,	O
the	O
co-ordination	O
of	O
development	O
work	O
on	O
the	O
open	B-Application
source	I-Application
version	O
of	O
ZFS	B-Application
was	O
passed	O
to	O
an	O
umbrella	O
project	O
,	O
OpenZFS	B-Operating_System
.	O
</s>
<s>
The	O
OpenZFS	B-Operating_System
framework	O
allows	O
any	O
interested	O
parties	O
to	O
collaboratively	O
develop	O
the	O
core	O
ZFS	B-Application
codebase	B-General_Concept
in	O
common	O
,	O
while	O
individually	O
maintaining	O
any	O
specific	O
extra	O
code	O
which	O
ZFS	B-Application
requires	O
to	O
function	O
and	O
integrate	O
within	O
their	O
own	O
systems	O
.	O
</s>
<s>
ZFS	B-Application
Filesystem	B-Application
Version	O
Number	O
Release	O
date	O
Significant	O
changes	O
OpenSolaris	B-Operating_System
Nevada	O
"	O
While	O
under	O
Sun	O
Microsystems	O
 '	O
control	O
,	O
there	O
were	O
bi-weekly	O
snapshots	B-Application
of	O
Solaris	B-Application
Nevada	O
(	O
the	O
codename	O
for	O
the	O
next-generation	O
Solaris	B-Application
OS	I-Application
to	O
eventually	O
succeed	O
Solaris	B-Application
10	I-Application
)	O
and	O
this	O
new	O
code	O
was	O
then	O
pulled	O
into	O
new	O
OpenSolaris	B-Operating_System
preview	O
snapshots	B-Application
available	O
at	O
Genunix.org	O
.	O
</s>
<s>
The	O
stable	O
releases	O
of	O
OpenSolaris	B-Operating_System
are	O
based	O
off	O
of	O
these	O
Nevada	O
builds.	O
"	O
</s>
<s>
build	O
36	O
First	O
release	O
OpenSolaris	B-Operating_System
Nevada	O
b69	O
Enhanced	O
directory	O
entries	O
.	O
</s>
<s>
OpenSolaris	B-Operating_System
Nevada	O
b77	O
Support	O
for	O
sharing	O
ZFS	B-Application
file	B-Application
systems	I-Application
over	O
SMB	B-Operating_System
.	O
</s>
<s>
ZFS	B-Application
Pool	O
Version	O
Number	O
Release	O
date	O
Significant	O
changes	O
OpenSolaris	B-Operating_System
Nevada	O
b36	O
First	O
release	O
OpenSolaris	B-Operating_System
Nevada	O
b38	O
Ditto	O
Blocks	O
OpenSolaris	B-Operating_System
Nevada	O
b42	O
Hot	O
spares	O
,	O
double-parity	O
RAID-Z	O
(	O
raidz2	O
)	O
,	O
improved	O
RAID-Z	O
accounting	O
OpenSolaris	B-Operating_System
Nevada	O
b62	O
zpool	O
history	O
OpenSolaris	B-Operating_System
Nevada	O
b62	O
gzip	B-Application
compression	I-Application
for	O
ZFS	B-Application
datasets	O
OpenSolaris	B-Operating_System
Nevada	O
b62	O
"	O
bootfs	O
"	O
pool	O
property	O
OpenSolaris	B-Operating_System
Nevada	O
b68	O
ZIL	O
:	O
adds	O
the	O
capability	O
to	O
specify	O
a	O
separate	O
Intent	O
Log	O
device	O
or	O
devices	O
OpenSolaris	B-Operating_System
Nevada	O
b69	O
ability	O
to	O
delegate	O
zfs(1M )	O
administrative	O
tasks	O
to	O
ordinary	O
users	O
OpenSolaris	B-Operating_System
Nevada	O
b77	O
CIFS	B-Operating_System
server	B-Application
support	O
,	O
dataset	O
quotas	B-Application
OpenSolaris	B-Operating_System
Nevada	O
b77	O
Devices	O
can	O
be	O
added	O
to	O
a	O
storage	O
pool	O
as	O
"	O
cache	B-General_Concept
devices	O
"	O
OpenSolaris	B-Operating_System
Nevada	O
b94	O
Improved	O
zpool	O
scrub	B-Error_Name
/	O
resilver	O
performance	O
OpenSolaris	B-Operating_System
Nevada	O
b96Snapshot	O
properties	O
OpenSolaris	B-Operating_System
Nevada	O
b98	O
Properties	O
:	O
usedbysnapshots	O
,	O
usedbychildren	O
,	O
usedbyrefreservation	O
,	O
and	O
usedbydataset	O
OpenSolaris	B-Operating_System
Nevada	O
b103	O
passthrough-x	O
aclinherit	O
property	O
support	O
OpenSolaris	B-Operating_System
Nevada	O
b114	O
Properties	O
:	O
userquota	O
,	O
groupquota	O
,	O
usuerused	O
and	O
groupused	O
;	O
also	O
required	O
FS	O
v4	O
OpenSolaris	B-Operating_System
Nevada	O
b116	O
STMF	O
property	O
support	O
OpenSolaris	B-Operating_System
Nevada	O
b120	O
triple-parity	O
RAID-Z	O
OpenSolaris	B-Operating_System
Nevada	O
b121	O
ZFS	B-Application
snapshot	B-Application
holds	O
OpenSolaris	B-Operating_System
Nevada	O
b125	O
ZFS	B-Application
log	O
device	O
removal	O
OpenSolaris	B-Operating_System
Nevada	O
b128	O
zle	O
compression	B-General_Concept
algorithm	I-General_Concept
that	O
is	O
needed	O
to	O
support	O
the	O
ZFS	B-Application
deduplication	B-General_Concept
properties	O
in	O
ZFS	B-Application
pool	O
version	O
21	O
,	O
which	O
were	O
released	O
concurrently	O
OpenSolaris	B-Operating_System
Nevada	O
b128	O
Deduplication	B-General_Concept
OpenSolaris	B-Operating_System
Nevada	O
b128	O
zfs	B-Application
receive	O
properties	O
OpenSolaris	B-Operating_System
Nevada	O
b135	O
slim	O
ZIL	O
OpenSolaris	B-Operating_System
Nevada	O
b137	O
System	O
attributes	O
.	O
</s>
<s>
Note	O
:	O
The	O
Solaris	B-Application
version	O
under	O
development	O
by	O
Sun	O
since	O
the	O
release	O
of	O
Solaris	B-Application
10	I-Application
in	O
2005	O
was	O
codenamed	O
'	O
Nevada	O
 '	O
,	O
and	O
was	O
derived	O
from	O
what	O
was	O
the	O
OpenSolaris	B-Operating_System
codebase	B-General_Concept
.	O
</s>
<s>
'	O
Solaris	B-Application
Nevada	O
 '	O
is	O
the	O
codename	O
for	O
the	O
next-generation	O
Solaris	B-Application
OS	I-Application
to	O
eventually	O
succeed	O
Solaris	B-Application
10	I-Application
and	O
this	O
new	O
code	O
was	O
then	O
pulled	O
successively	O
into	O
new	O
OpenSolaris	B-Operating_System
'	O
Nevada	O
 '	O
snapshot	B-Application
builds	O
.	O
</s>
<s>
OpenSolaris	B-Operating_System
is	O
now	O
discontinued	O
and	O
OpenIndiana	B-Application
forked	B-Application
from	O
it	O
.	O
</s>
<s>
A	O
final	O
build	O
(	O
b134	O
)	O
of	O
OpenSolaris	B-Operating_System
was	O
published	O
by	O
Oracle	B-Application
(	O
2010-Nov-12	O
)	O
as	O
an	O
upgrade	O
path	O
to	O
Solaris	B-Application
11	I-Application
Express	I-Application
.	O
</s>
<s>
List	O
of	O
Operating	B-General_Concept
Systems	I-General_Concept
,	O
distributions	O
and	O
add-ons	O
that	O
support	O
ZFS	B-Application
,	O
the	O
zpool	O
version	O
it	O
supports	O
,	O
and	O
the	O
Solaris	B-Application
build	O
they	O
are	O
based	O
on	O
(	O
if	O
any	O
)	O
:	O
</s>
