<s>
In	O
computer	B-General_Concept
science	I-General_Concept
,	O
a	O
parallel	B-Operating_System
algorithm	I-Operating_System
,	O
as	O
opposed	O
to	O
a	O
traditional	O
serial	B-Algorithm
algorithm	I-Algorithm
,	O
is	O
an	O
algorithm	O
which	O
can	O
do	O
multiple	O
operations	O
in	O
a	O
given	O
time	O
.	O
</s>
<s>
It	O
has	O
been	O
a	O
tradition	O
of	O
computer	B-General_Concept
science	I-General_Concept
to	O
describe	O
serial	B-Algorithm
algorithms	I-Algorithm
in	O
abstract	B-Application
machine	I-Application
models	O
,	O
often	O
the	O
one	O
known	O
as	O
random-access	B-Application
machine	I-Application
.	O
</s>
<s>
Similarly	O
,	O
many	O
computer	B-General_Concept
science	I-General_Concept
researchers	O
have	O
used	O
a	O
so-called	O
parallel	B-Operating_System
random-access	I-Operating_System
machine	I-Operating_System
(	O
PRAM	O
)	O
as	O
a	O
parallel	O
abstract	B-Application
machine	I-Application
(	O
shared-memory	B-Operating_System
)	O
.	O
</s>
<s>
Many	O
parallel	B-Operating_System
algorithms	I-Operating_System
are	O
executed	O
concurrently	B-Architecture
–	O
though	O
in	O
general	O
concurrent	B-Architecture
algorithms	I-Architecture
are	O
a	O
distinct	O
concept	O
–	O
and	O
thus	O
these	O
concepts	O
are	O
often	O
conflated	O
,	O
with	O
which	O
aspect	O
of	O
an	O
algorithm	O
is	O
parallel	O
and	O
which	O
is	O
concurrent	B-Operating_System
not	O
being	O
clearly	O
distinguished	O
.	O
</s>
<s>
Further	O
,	O
non-parallel	O
,	O
non-concurrent	O
algorithms	O
are	O
often	O
referred	O
to	O
as	O
"	O
sequential	B-Algorithm
algorithms	I-Algorithm
"	O
,	O
by	O
contrast	O
with	O
concurrent	B-Architecture
algorithms	I-Architecture
.	O
</s>
<s>
Some	O
problems	O
are	O
easy	O
to	O
divide	O
up	O
into	O
pieces	O
in	O
this	O
way	O
–	O
these	O
are	O
called	O
embarrassingly	B-Operating_System
parallel	I-Operating_System
problems	I-Operating_System
.	O
</s>
<s>
Examples	O
include	O
many	O
algorithms	O
to	O
solve	O
Rubik	O
's	O
Cubes	O
and	O
find	O
values	O
which	O
result	O
in	O
a	O
given	O
hash	B-Application
.	O
</s>
<s>
Some	O
problems	O
cannot	O
be	O
split	O
up	O
into	O
parallel	O
portions	O
,	O
as	O
they	O
require	O
the	O
results	O
from	O
a	O
preceding	O
step	O
to	O
effectively	O
carry	O
on	O
with	O
the	O
next	O
step	O
–	O
these	O
are	O
called	O
s	O
.	O
Examples	O
include	O
iterative	O
numerical	B-General_Concept
methods	I-General_Concept
,	O
such	O
as	O
Newton	O
's	O
method	O
,	O
iterative	O
solutions	O
to	O
the	O
three-body	O
problem	O
,	O
and	O
most	O
of	O
the	O
available	O
algorithms	O
to	O
compute	O
pi	O
( π	O
)	O
.	O
</s>
<s>
Some	O
sequential	B-Algorithm
algorithms	I-Algorithm
can	O
be	O
converted	O
into	O
parallel	B-Operating_System
algorithms	I-Operating_System
using	O
automatic	O
parallelization	B-Operating_System
.	O
</s>
<s>
Parallel	B-Operating_System
algorithms	I-Operating_System
on	O
individual	O
devices	O
have	O
become	O
more	O
common	O
since	O
the	O
early	O
2000s	O
because	O
of	O
substantial	O
improvements	O
in	O
multiprocessing	B-Operating_System
systems	O
and	O
the	O
rise	O
of	O
multi-core	B-Architecture
processors	I-Architecture
.	O
</s>
<s>
Up	O
until	O
the	O
end	O
of	O
2004	O
,	O
single-core	O
processor	O
performance	O
rapidly	O
increased	O
via	O
frequency	B-General_Concept
scaling	I-General_Concept
,	O
and	O
thus	O
it	O
was	O
easier	O
to	O
construct	O
a	O
computer	O
with	O
a	O
single	O
fast	O
core	O
than	O
one	O
with	O
many	O
slower	O
cores	O
with	O
the	O
same	O
throughput	O
,	O
so	O
multicore	B-Architecture
systems	O
were	O
of	O
more	O
limited	O
use	O
.	O
</s>
<s>
Since	O
2004	O
however	O
,	O
frequency	B-General_Concept
scaling	I-General_Concept
hit	O
a	O
wall	O
,	O
and	O
thus	O
multicore	B-Architecture
systems	O
have	O
become	O
more	O
widespread	O
,	O
making	O
parallel	B-Operating_System
algorithms	I-Operating_System
of	O
more	O
general	O
use	O
.	O
</s>
<s>
The	O
cost	O
or	O
complexity	O
of	O
serial	B-Algorithm
algorithms	I-Algorithm
is	O
estimated	O
in	O
terms	O
of	O
the	O
space	O
(	O
memory	O
)	O
and	O
time	O
(	O
processor	O
cycles	O
)	O
that	O
they	O
take	O
.	O
</s>
<s>
Parallel	B-Operating_System
algorithms	I-Operating_System
need	O
to	O
optimize	O
one	O
more	O
resource	O
,	O
the	O
communication	O
between	O
different	O
processors	O
.	O
</s>
<s>
There	O
are	O
two	O
ways	O
parallel	B-Operating_System
processors	I-Operating_System
communicate	O
,	O
shared	B-Operating_System
memory	I-Operating_System
or	O
message	B-Architecture
passing	I-Architecture
.	O
</s>
<s>
Shared	B-Operating_System
memory	I-Operating_System
processing	O
needs	O
additional	O
locking	B-Operating_System
for	O
the	O
data	O
,	O
imposes	O
the	O
overhead	O
of	O
additional	O
processor	O
and	O
bus	O
cycles	O
,	O
and	O
also	O
serializes	O
some	O
portion	O
of	O
the	O
algorithm	O
.	O
</s>
<s>
Message	B-Architecture
passing	I-Architecture
processing	O
uses	O
channels	O
and	O
message	O
boxes	O
but	O
this	O
communication	O
adds	O
transfer	O
overhead	O
on	O
the	O
bus	O
,	O
additional	O
memory	O
need	O
for	O
queues	O
and	O
message	O
boxes	O
and	O
latency	O
in	O
the	O
messages	O
.	O
</s>
<s>
Designs	O
of	O
parallel	B-Operating_System
processors	I-Operating_System
use	O
special	O
buses	B-General_Concept
like	O
crossbar	O
so	O
that	O
the	O
communication	O
overhead	O
will	O
be	O
small	O
but	O
it	O
is	O
the	O
parallel	B-Operating_System
algorithm	I-Operating_System
that	O
decides	O
the	O
volume	O
of	O
the	O
traffic	O
.	O
</s>
<s>
If	O
the	O
communication	O
overhead	O
of	O
additional	O
processors	O
outweighs	O
the	O
benefit	O
of	O
adding	O
another	O
processor	O
,	O
one	O
encounters	O
parallel	B-Operating_System
slowdown	I-Operating_System
.	O
</s>
<s>
Another	O
problem	O
with	O
parallel	B-Operating_System
algorithms	I-Operating_System
is	O
ensuring	O
that	O
they	O
are	O
suitably	O
load	B-Application
balanced	I-Application
,	O
by	O
ensuring	O
that	O
load	O
(	O
overall	O
work	O
)	O
is	O
balanced	O
,	O
rather	O
than	O
input	O
size	O
being	O
balanced	O
.	O
</s>
<s>
A	O
subtype	O
of	O
parallel	B-Operating_System
algorithms	I-Operating_System
,	O
distributed	B-Operating_System
algorithms	I-Operating_System
,	O
are	O
algorithms	O
designed	O
to	O
work	O
in	O
cluster	B-Architecture
computing	I-Architecture
and	O
distributed	B-Architecture
computing	I-Architecture
environments	O
,	O
where	O
additional	O
concerns	O
beyond	O
the	O
scope	O
of	O
"	O
classical	O
"	O
parallel	B-Operating_System
algorithms	I-Operating_System
need	O
to	O
be	O
addressed	O
.	O
</s>
