<s>
Data	B-Operating_System
parallelism	I-Operating_System
is	O
parallelization	B-Operating_System
across	O
multiple	O
processors	O
in	O
parallel	B-Operating_System
computing	I-Operating_System
environments	O
.	O
</s>
<s>
It	O
contrasts	O
to	O
task	B-Operating_System
parallelism	I-Operating_System
as	O
another	O
form	O
of	O
parallelism	B-Operating_System
.	O
</s>
<s>
A	O
data	B-Operating_System
parallel	I-Operating_System
job	O
on	O
an	O
array	O
of	O
n	O
elements	O
can	O
be	O
divided	O
equally	O
among	O
all	O
the	O
processors	O
.	O
</s>
<s>
On	O
the	O
other	O
hand	O
,	O
if	O
we	O
execute	O
this	O
job	O
as	O
a	O
data	B-Operating_System
parallel	I-Operating_System
job	O
on	O
4	O
processors	O
the	O
time	O
taken	O
would	O
reduce	O
to	O
(	O
n/4	O
)	O
×Ta	O
+	O
merging	O
overhead	O
time	O
units	O
.	O
</s>
<s>
One	O
important	O
thing	O
to	O
note	O
is	O
that	O
the	O
locality	B-General_Concept
of	I-General_Concept
data	I-General_Concept
references	I-General_Concept
plays	O
an	O
important	O
part	O
in	O
evaluating	O
the	O
performance	O
of	O
a	O
data	B-Operating_System
parallel	I-Operating_System
programming	O
model	O
.	O
</s>
<s>
Exploitation	O
of	O
the	O
concept	O
of	O
data	B-Operating_System
parallelism	I-Operating_System
started	O
in	O
1960s	O
with	O
the	O
development	O
of	O
Solomon	O
machine	O
.	O
</s>
<s>
The	O
Solomon	O
machine	O
,	O
also	O
called	O
a	O
vector	B-Operating_System
processor	I-Operating_System
,	O
was	O
developed	O
to	O
expedite	O
the	O
performance	O
of	O
mathematical	O
operations	O
by	O
working	O
on	O
a	O
large	O
data	O
array	O
(	O
operating	O
on	O
multiple	O
data	O
in	O
consecutive	O
time	O
steps	O
)	O
.	O
</s>
<s>
Concurrency	B-Application
of	O
data	O
operations	O
was	O
also	O
exploited	O
by	O
operating	O
on	O
multiple	O
data	O
at	O
the	O
same	O
time	O
using	O
a	O
single	O
instruction	O
.	O
</s>
<s>
These	O
processors	O
were	O
called	O
'	O
array	B-Operating_System
processors	I-Operating_System
 '	O
.	O
</s>
<s>
In	O
the	O
1980s	O
,	O
the	O
term	O
was	O
introduced	O
to	O
describe	O
this	O
programming	O
style	O
,	O
which	O
was	O
widely	O
used	O
to	O
program	O
Connection	B-Device
Machines	I-Device
in	O
data	B-Operating_System
parallel	I-Operating_System
languages	O
like	O
C*	B-Language
.	O
</s>
<s>
Today	O
,	O
data	B-Operating_System
parallelism	I-Operating_System
is	O
best	O
exemplified	O
in	O
graphics	B-Architecture
processing	I-Architecture
units	I-Architecture
(	O
GPUs	B-Architecture
)	O
,	O
which	O
use	O
both	O
the	O
techniques	O
of	O
operating	O
on	O
multiple	O
data	O
in	O
space	O
and	O
time	O
using	O
a	O
single	O
instruction	O
.	O
</s>
<s>
Most	O
data	B-Operating_System
parallel	I-Operating_System
hardware	O
supports	O
only	O
a	O
fixed	O
number	O
of	O
parallel	O
levels	O
,	O
often	O
only	O
one	O
.	O
</s>
<s>
This	O
means	O
that	O
within	O
a	O
parallel	O
operation	O
it	O
is	O
not	O
possible	O
to	O
launch	O
more	O
parallel	O
operations	O
recursively	O
,	O
and	O
means	O
that	O
programmers	O
cannot	O
make	O
use	O
of	O
nested	O
hardware	O
parallelism	B-Operating_System
.	O
</s>
<s>
The	O
programming	O
language	O
NESL	B-Language
was	O
an	O
early	O
effort	O
at	O
implementing	O
a	O
nested	O
data-parallel	O
programming	O
model	O
on	O
flat	O
parallel	B-Operating_System
machines	I-Operating_System
,	O
and	O
in	O
particular	O
introduced	O
the	O
flattening	O
transformation	O
that	O
transforms	O
nested	O
data	B-Operating_System
parallelism	I-Operating_System
to	O
flat	O
data	B-Operating_System
parallelism	I-Operating_System
.	O
</s>
<s>
This	O
work	O
was	O
continued	O
by	O
other	O
languages	O
such	O
as	O
Data	B-Operating_System
Parallel	I-Operating_System
Haskell	O
and	O
Futhark	B-Language
,	O
although	O
arbitrary	O
nested	O
data	B-Operating_System
parallelism	I-Operating_System
is	O
not	O
widely	O
available	O
in	O
current	O
data-parallel	O
programming	O
languages	O
.	O
</s>
<s>
In	O
a	O
multiprocessor	O
system	O
executing	O
a	O
single	O
set	O
of	O
instructions	O
(	O
SIMD	B-Device
)	O
,	O
data	B-Operating_System
parallelism	I-Operating_System
is	O
achieved	O
when	O
each	O
processor	O
performs	O
the	O
same	O
task	O
on	O
different	O
distributed	O
data	O
.	O
</s>
<s>
Below	O
is	O
the	O
sequential	O
pseudo-code	B-Language
for	O
multiplication	O
and	O
addition	O
of	O
two	O
matrices	O
where	O
the	O
result	O
is	O
stored	O
in	O
the	O
matrix	O
.	O
</s>
<s>
The	O
pseudo-code	B-Language
for	O
multiplication	O
calculates	O
the	O
dot	O
product	O
of	O
two	O
matrices	O
,	O
and	O
stores	O
the	O
result	O
into	O
the	O
output	O
matrix	O
.	O
</s>
<s>
We	O
can	O
exploit	O
data	B-Operating_System
parallelism	I-Operating_System
in	O
the	O
preceding	O
code	O
to	O
execute	O
it	O
faster	O
as	O
the	O
arithmetic	O
is	O
loop	O
independent	O
.	O
</s>
<s>
Parallelization	B-Operating_System
of	O
the	O
matrix	O
multiplication	O
code	O
is	O
achieved	O
by	O
using	O
OpenMP	B-Application
.	O
</s>
<s>
An	O
OpenMP	B-Application
directive	O
,	O
"	O
omp	O
parallel	O
for	O
"	O
instructs	O
the	O
compiler	O
to	O
execute	O
the	O
code	O
in	O
the	O
for	O
loop	O
in	O
parallel	O
.	O
</s>
<s>
For	O
addition	O
of	O
arrays	O
in	O
a	O
data	B-Operating_System
parallel	I-Operating_System
implementation	O
,	O
let	O
's	O
assume	O
a	O
more	O
modest	O
system	O
with	O
two	O
central	B-General_Concept
processing	I-General_Concept
units	I-General_Concept
(	O
CPU	O
)	O
A	O
and	O
B	O
,	O
CPU	O
A	O
could	O
add	O
all	O
elements	O
from	O
the	O
top	O
half	O
of	O
the	O
arrays	O
,	O
while	O
CPU	O
B	O
could	O
add	O
all	O
elements	O
from	O
the	O
bottom	O
half	O
of	O
the	O
arrays	O
.	O
</s>
<s>
The	O
program	O
expressed	O
in	O
pseudocode	B-Language
below	O
—	O
which	O
applies	O
some	O
arbitrary	O
operation	O
,	O
foo	O
,	O
on	O
every	O
element	O
in	O
the	O
array	O
d	O
—	O
illustrates	O
data	B-Operating_System
parallelism	I-Operating_System
:	O
</s>
<s>
In	O
an	O
SPMD	B-Operating_System
system	O
executed	O
on	O
2	O
processor	O
system	O
,	O
both	O
CPUs	O
will	O
execute	O
the	O
code	O
.	O
</s>
<s>
Data	B-Operating_System
parallelism	I-Operating_System
emphasizes	O
the	O
distributed	O
(	O
parallel	O
)	O
nature	O
of	O
the	O
data	O
,	O
as	O
opposed	O
to	O
the	O
processing	O
(	O
task	B-Operating_System
parallelism	I-Operating_System
)	O
.	O
</s>
<s>
Most	O
real	O
programs	O
fall	O
somewhere	O
on	O
a	O
continuum	O
between	O
task	B-Operating_System
parallelism	I-Operating_System
and	O
data	B-Operating_System
parallelism	I-Operating_System
.	O
</s>
<s>
Data	O
parallelismTask	O
parallelismSame	O
operations	O
are	O
performed	O
on	O
different	O
subsets	O
of	O
same	O
data.Different	O
operations	O
are	O
performed	O
on	O
the	O
same	O
or	O
different	O
data.Synchronous	O
computationAsynchronous	O
computationSpeedup	O
is	O
more	O
as	O
there	O
is	O
only	O
one	O
execution	O
thread	O
operating	O
on	O
all	O
sets	O
of	O
data.Speedup	O
is	O
less	O
as	O
each	O
processor	O
will	O
execute	O
a	O
different	O
thread	O
or	O
process	O
on	O
the	O
same	O
or	O
different	O
set	O
of	O
data.Amount	O
of	O
parallelization	B-Operating_System
is	O
proportional	O
to	O
the	O
input	O
data	O
size.Amount	O
of	O
parallelization	B-Operating_System
is	O
proportional	O
to	O
the	O
number	O
of	O
independent	O
tasks	O
to	O
be	O
performed.Designed	O
for	O
optimum	O
load	B-Application
balance	I-Application
on	O
multi	O
processor	O
system.Load	O
balancing	O
depends	O
on	O
the	O
availability	O
of	O
the	O
hardware	O
and	O
scheduling	O
algorithms	O
like	O
static	O
and	O
dynamic	O
scheduling	O
.	O
</s>
<s>
Data	O
and	O
task	B-Operating_System
parallelism	I-Operating_System
,	O
can	O
be	O
simultaneously	O
implemented	O
by	O
combining	O
them	O
together	O
for	O
the	O
same	O
application	O
.	O
</s>
<s>
This	O
is	O
called	O
Mixed	O
data	O
and	O
task	B-Operating_System
parallelism	I-Operating_System
.	O
</s>
<s>
Mixed	O
parallelism	B-Operating_System
requires	O
sophisticated	O
scheduling	O
algorithms	O
and	O
software	O
support	O
.	O
</s>
<s>
It	O
is	O
the	O
best	O
kind	O
of	O
parallelism	B-Operating_System
when	O
communication	O
is	O
slow	O
and	O
number	O
of	O
processors	O
is	O
large	O
.	O
</s>
<s>
Mixed	O
data	O
and	O
task	B-Operating_System
parallelism	I-Operating_System
has	O
many	O
applications	O
.	O
</s>
<s>
Mixed	O
data	O
and	O
task	B-Operating_System
parallelism	I-Operating_System
finds	O
applications	O
in	O
the	O
global	O
climate	O
modeling	O
.	O
</s>
<s>
Large	O
data	B-Operating_System
parallel	I-Operating_System
computations	I-Operating_System
are	O
performed	O
by	O
creating	O
grids	O
of	O
data	O
representing	O
earth	O
's	O
atmosphere	O
and	O
oceans	O
and	O
task	B-Operating_System
parallelism	I-Operating_System
is	O
employed	O
for	O
simulating	O
the	O
function	O
and	O
model	O
of	O
the	O
physical	O
processes	O
.	O
</s>
<s>
In	O
timing	O
based	O
circuit	B-Language
simulation	I-Language
.	O
</s>
<s>
The	O
data	O
is	O
divided	O
among	O
different	O
sub-circuits	O
and	O
parallelism	B-Operating_System
is	O
achieved	O
with	O
orchestration	O
from	O
the	O
tasks	O
.	O
</s>
<s>
A	O
variety	O
of	O
data	B-Operating_System
parallel	I-Operating_System
programming	O
environments	O
are	O
available	O
today	O
,	O
most	O
widely	O
used	O
of	O
which	O
are	O
:	O
</s>
<s>
Message	B-Application
Passing	I-Application
Interface	I-Application
:	O
It	O
is	O
a	O
cross-platform	O
message	O
passing	O
programming	O
interface	O
for	O
parallel	B-Operating_System
computers	I-Operating_System
.	O
</s>
<s>
Open	O
Multi	O
Processing	O
(	O
Open	B-Application
MP	I-Application
)	O
:	O
It	O
's	O
an	O
Application	O
Programming	O
Interface	O
(	O
API	O
)	O
which	O
supports	O
shared	B-Operating_System
memory	I-Operating_System
programming	I-Operating_System
models	O
on	O
multiple	O
platforms	O
of	O
multiprocessor	O
systems	O
.	O
</s>
<s>
CUDA	B-Architecture
and	O
OpenACC	B-Operating_System
:	O
CUDA	B-Architecture
and	O
OpenACC	B-Operating_System
(	O
respectively	O
)	O
are	O
parallel	B-Operating_System
computing	I-Operating_System
API	O
platforms	O
designed	O
to	O
allow	O
a	O
software	O
engineer	O
to	O
utilize	O
GPU	B-Architecture
's	O
computational	O
units	O
for	O
general	O
purpose	O
processing	O
.	O
</s>
<s>
Threading	B-Application
Building	I-Application
Blocks	I-Application
and	O
RaftLib	B-Language
:	O
Both	O
open	O
source	O
programming	O
environments	O
that	O
enable	O
mixed	O
data/task	O
parallelism	O
in	O
C/C	O
++	O
environments	O
across	O
heterogeneous	O
resources	O
.	O
</s>
<s>
Data	B-Operating_System
parallelism	I-Operating_System
finds	O
its	O
applications	O
in	O
a	O
variety	O
of	O
fields	O
ranging	O
from	O
physics	O
,	O
chemistry	O
,	O
biology	O
,	O
material	O
sciences	O
to	O
signal	O
processing	O
.	O
</s>
<s>
Sciences	O
imply	O
data	B-Operating_System
parallelism	I-Operating_System
for	O
simulating	O
models	O
like	O
molecular	O
dynamics	O
,	O
sequence	O
analysis	O
of	O
genome	O
data	O
and	O
other	O
physical	O
phenomenon	O
.	O
</s>
<s>
Driving	O
forces	O
in	O
signal	O
processing	O
for	O
data	B-Operating_System
parallelism	I-Operating_System
are	O
video	O
encoding	O
,	O
image	O
and	O
graphics	O
processing	O
,	O
wireless	O
communications	O
to	O
name	O
a	O
few	O
.	O
</s>
