<s>
In	O
computing	O
,	O
algorithmic	B-Language
skeletons	I-Language
,	O
or	O
parallelism	B-Language
patterns	I-Language
,	O
are	O
a	O
high-level	O
parallel	B-Application
programming	I-Application
model	I-Application
for	O
parallel	O
and	O
distributed	O
computing	O
.	O
</s>
<s>
Algorithmic	B-Language
skeletons	I-Language
take	O
advantage	O
of	O
common	O
programming	O
patterns	O
to	O
hide	O
the	O
complexity	O
of	O
parallel	O
and	O
distributed	O
applications	O
.	O
</s>
<s>
The	O
most	O
outstanding	O
feature	O
of	O
algorithmic	B-Language
skeletons	I-Language
,	O
which	O
differentiates	O
them	O
from	O
other	O
high-level	O
parallel	B-Application
programming	I-Application
models	I-Application
,	O
is	O
that	O
orchestration	O
and	O
synchronization	O
of	O
the	O
parallel	O
activities	O
is	O
implicitly	O
defined	O
by	O
the	O
skeleton	O
patterns	O
.	O
</s>
<s>
Second	O
,	O
that	O
algorithmic	B-Language
skeleton	I-Language
programming	O
reduces	O
the	O
number	O
of	O
errors	O
when	O
compared	O
to	O
traditional	O
lower-level	O
parallel	B-Application
programming	I-Application
models	I-Application
(	O
Threads	O
,	O
MPI	B-Application
)	O
.	O
</s>
<s>
The	O
generated	O
code	O
is	O
mostly	O
type	B-Language
safe	I-Language
,	O
using	O
the	O
types	O
provided	O
by	O
the	O
programmer	O
which	O
do	O
not	O
require	O
extension	O
of	O
superclass	O
,	O
but	O
fails	O
to	O
be	O
completely	O
type	B-Language
safe	I-Language
such	O
as	O
in	O
the	O
reduce( 	O
...	O
,	O
Object	O
reducer	O
)	O
method	O
in	O
the	O
mesh	O
pattern	O
.	O
</s>
<s>
Support	O
for	O
distributed	B-Operating_System
memory	I-Operating_System
architectures	O
in	O
CO2P3S	O
was	O
introduced	O
in	O
later	O
.	O
</s>
<s>
To	O
use	O
a	O
distributed	B-Operating_System
memory	I-Operating_System
pattern	O
,	O
programmers	O
must	O
change	O
the	O
pattern	O
's	O
memory	O
option	O
from	O
shared	O
to	O
distributed	O
,	O
and	O
generate	O
the	O
new	O
code	O
.	O
</s>
<s>
From	O
the	O
usage	O
perspective	O
,	O
the	O
distributed	B-Operating_System
memory	I-Operating_System
version	O
of	O
the	O
code	O
requires	O
the	O
management	O
of	O
remote	O
exceptions	O
.	O
</s>
<s>
As	O
such	O
,	O
it	O
provides	O
algorithmic	B-Language
skeleton	I-Language
programming	O
as	O
a	O
Java	O
library	O
.	O
</s>
<s>
Both	O
task	O
and	O
data	B-Operating_System
parallel	I-Operating_System
skeletons	O
are	O
fully	O
nestable	O
;	O
and	O
are	O
instantiated	O
via	O
parametric	O
skeleton	O
objects	O
,	O
not	O
inheritance	O
.	O
</s>
<s>
Additionally	O
,	O
Calcium	O
has	O
three	O
distinctive	O
features	O
for	O
algorithmic	B-Language
skeleton	I-Language
programming	O
.	O
</s>
<s>
Third	O
,	O
a	O
transparent	O
algorithmic	B-Language
skeleton	I-Language
file	O
access	O
model	O
,	O
which	O
enables	O
skeletons	O
for	O
data	O
intensive	O
applications	O
.	O
</s>
<s>
is	O
a	O
complete	O
re-implementation	O
of	O
Calcium	O
for	O
multi-core	B-Architecture
computing	O
.	O
</s>
<s>
Eden	O
is	O
a	O
parallel	O
programming	O
language	O
for	O
distributed	B-Operating_System
memory	I-Operating_System
environments	O
,	O
which	O
extends	O
Haskell	B-Language
.	O
</s>
<s>
Instead	O
,	O
skeletons	O
are	O
defined	O
on	O
top	O
of	O
Eden	O
's	O
lower-level	O
process	O
abstraction	O
,	O
supporting	O
both	O
task	O
and	O
data	B-Operating_System
parallelism	I-Operating_System
.	O
</s>
<s>
Because	O
Eden	O
is	O
an	O
extension	O
of	O
a	O
functional	O
language	O
,	O
Eden	O
skeletons	O
are	O
higher	B-Language
order	I-Language
functions	I-Language
.	O
</s>
<s>
Eden	O
introduces	O
the	O
concept	O
of	O
implementation	O
skeleton	O
,	O
which	O
is	O
an	O
architecture	O
independent	O
scheme	O
that	O
describes	O
a	O
parallel	O
implementation	O
of	O
an	O
algorithmic	B-Language
skeleton	I-Language
.	O
</s>
<s>
The	O
Edinburgh	O
Skeleton	O
Library	O
(	O
eSkel	O
)	O
is	O
provided	O
in	O
C	O
and	O
runs	O
on	O
top	O
of	O
MPI	B-Application
.	O
</s>
<s>
Being	O
initially	O
developed	O
to	O
target	O
multi-core	B-Architecture
platforms	O
,	O
it	O
has	O
been	O
successively	O
extended	O
to	O
target	O
heterogeneous	O
platforms	O
composed	O
of	O
clusters	O
of	O
shared-memory	O
platforms	O
,	O
possibly	O
equipped	O
with	O
computing	O
accelerators	O
such	O
as	O
NVidia	O
GPGPUs	O
,	O
Xeon	O
Phi	O
,	O
Tilera	O
TILE64	O
.	O
</s>
<s>
Like	O
other	O
high-level	O
programming	O
frameworks	O
,	O
such	O
as	O
Intel	B-Application
TBB	I-Application
and	O
OpenMP	B-Application
,	O
it	O
simplifies	O
the	O
design	O
and	O
engineering	O
of	O
portable	O
parallel	O
applications	O
.	O
</s>
<s>
However	O
,	O
it	O
has	O
a	O
clear	O
edge	O
in	O
terms	O
of	O
expressiveness	O
and	O
performance	O
with	O
respect	O
to	O
other	O
parallel	O
programming	O
frameworks	O
in	O
specific	O
application	O
scenarios	O
,	O
including	O
,	O
inter	O
alia	O
:	O
fine-grain	O
parallelism	O
on	O
cache-coherent	O
shared-memory	O
platforms	O
;	O
streaming	O
applications	O
;	O
coupled	O
usage	O
of	O
multi-core	B-Architecture
and	O
accelerators	O
.	O
</s>
<s>
In	O
other	O
cases	O
FastFlow	O
is	O
typically	O
comparable	O
to	O
(	O
and	O
is	O
some	O
cases	O
slightly	O
faster	O
than	O
)	O
state-of-the-art	O
parallel	O
programming	O
frameworks	O
such	O
as	O
Intel	B-Application
TBB	I-Application
,	O
OpenMP	B-Application
,	O
Cilk	O
,	O
etc	O
.	O
</s>
<s>
Higher-order	O
Divide	O
and	O
Conquer	O
(	O
HDC	O
)	O
is	O
a	O
subset	O
of	O
the	O
functional	O
language	O
Haskell	B-Language
.	O
</s>
<s>
Functional	O
programs	O
are	O
presented	O
as	O
polymorphic	O
higher-order	B-Language
functions	I-Language
,	O
which	O
can	O
be	O
compiled	O
into	O
C/MPI	O
,	O
and	O
linked	O
with	O
skeleton	O
implementations	O
.	O
</s>
<s>
The	O
evaluation	O
of	O
a	O
skeleton	O
application	O
follows	O
a	O
formal	O
definition	O
of	O
operational	O
semantics	O
introduced	O
by	O
Aldinucci	O
and	O
Danelutto	O
,	O
which	O
can	O
handle	O
both	O
task	O
and	O
data	B-Operating_System
parallelism	I-Operating_System
.	O
</s>
<s>
For	O
local	O
search	O
heuristics	O
Mallba	O
supports	O
:	O
hill	B-Algorithm
climbing	I-Algorithm
,	O
metropolis	O
,	O
simulated	B-Algorithm
annealing	I-Algorithm
,	O
and	O
tabu	B-Algorithm
search	I-Algorithm
;	O
and	O
also	O
population	O
based	O
heuristics	O
derived	O
from	O
evolutionary	B-Algorithm
algorithms	I-Algorithm
such	O
as	O
genetic	B-Algorithm
algorithms	I-Algorithm
,	O
evolution	O
strategy	O
,	O
and	O
others	O
(	O
CHC	O
)	O
.	O
</s>
<s>
simulated	B-Algorithm
annealing	I-Algorithm
,	O
and	O
CHCCES	O
which	O
combines	O
CHC	O
and	O
ES	O
.	O
</s>
<s>
The	O
skeletons	O
are	O
provided	O
as	O
a	O
C++	O
library	O
and	O
are	O
not	O
nestable	O
but	O
type	B-Language
safe	I-Language
.	O
</s>
<s>
A	O
custom	O
MPI	B-Application
abstraction	O
layer	O
is	O
used	O
,	O
NetStream	O
,	O
which	O
takes	O
care	O
of	O
primitive	O
data	O
type	O
marshalling	O
,	O
synchronization	O
,	O
etc	O
.	O
</s>
<s>
Marrow	O
is	O
a	O
C++	O
algorithmic	B-Language
skeleton	I-Language
framework	O
for	O
the	O
orchestration	O
of	O
OpenCL	B-Application
computations	O
in	O
,	O
possibly	O
heterogeneous	O
,	O
multi-GPU	O
environments	O
.	O
</s>
<s>
The	O
leaf	O
nodes	O
of	O
the	O
resulting	O
composition	O
trees	O
represent	O
the	O
GPU	B-Architecture
computational	O
kernels	O
,	O
while	O
the	O
remainder	O
nodes	O
denote	O
the	O
skeleton	O
applied	O
to	O
the	O
nested	O
sub-tree	O
.	O
</s>
<s>
The	O
framework	O
takes	O
upon	O
itself	O
the	O
entire	O
host-side	O
orchestration	O
required	O
to	O
correctly	O
execute	O
these	O
trees	O
in	O
heterogeneous	O
multi-GPU	O
environments	O
,	O
including	O
the	O
proper	O
ordering	O
of	O
the	O
data-transfer	O
and	O
of	O
the	O
execution	O
requests	O
,	O
and	O
the	O
communication	O
required	O
between	O
the	O
tree	O
's	O
nodes	O
.	O
</s>
<s>
Among	O
Marrow	O
's	O
most	O
distinguishable	O
features	O
are	O
a	O
set	O
of	O
skeletons	O
previously	O
unavailable	O
in	O
the	O
GPU	B-Architecture
context	O
,	O
such	O
as	O
Pipeline	O
and	O
Loop	O
,	O
and	O
the	O
skeleton	O
nesting	O
ability	O
–	O
a	O
feature	O
also	O
new	O
in	O
this	O
context	O
.	O
</s>
<s>
The	O
parallel	O
execution	O
of	O
a	O
Marrow	O
composition	O
tree	O
by	O
multiple	O
GPUs	B-Architecture
follows	O
a	O
data-parallel	O
decomposition	O
strategy	O
,	O
that	O
concurrently	O
applies	O
the	O
entire	O
computational	O
tree	O
to	O
different	O
partitions	O
of	O
the	O
input	O
dataset	O
.	O
</s>
<s>
Other	O
than	O
expressing	O
which	O
kernel	O
parameters	O
may	O
be	O
decomposed	O
and	O
,	O
when	O
required	O
,	O
defining	O
how	O
the	O
partial	O
results	O
should	O
be	O
merged	O
,	O
the	O
programmer	O
is	O
completely	O
abstracted	O
from	O
the	O
underlying	O
multi-GPU	O
architecture	O
.	O
</s>
<s>
higher	B-Language
order	I-Language
functions	I-Language
,	O
currying	O
,	O
and	O
polymorphic	O
types	O
.	O
</s>
<s>
It	O
is	O
built	O
on	O
top	O
of	O
MPI	B-Application
1.2	O
and	O
OpenMP	B-Application
2.5	O
and	O
supports	O
,	O
unlike	O
many	O
other	O
skeleton	O
libraries	O
,	O
both	O
task	O
and	O
data	B-Operating_System
parallel	I-Operating_System
skeletons	O
.	O
</s>
<s>
task	O
parallel	O
skeletons	O
can	O
be	O
nested	O
arbitrarily	O
while	O
data	B-Operating_System
parallel	I-Operating_System
skeletons	O
cannot	O
,	O
but	O
may	O
be	O
used	O
at	O
the	O
leaves	O
of	O
a	O
task	O
parallel	O
nesting	O
tree	O
.	O
</s>
<s>
However	O
,	O
the	O
library	O
implements	O
an	O
automated	O
serialization	O
mechanism	O
inspired	O
by	O
such	O
that	O
,	O
in	O
addition	O
to	O
the	O
standard	O
MPI	B-Application
data	O
types	O
,	O
arbitrary	O
user-defined	O
data	O
types	O
can	O
be	O
used	O
within	O
the	O
skeletons	O
.	O
</s>
<s>
Data	B-Operating_System
parallel	I-Operating_System
skeletons	O
,	O
such	O
as	O
fold	O
(	O
reduce	O
)	O
,	O
map	O
,	O
permute	O
,	O
zip	O
,	O
and	O
their	O
variants	O
are	O
implemented	O
as	O
higher	O
order	O
member	O
functions	O
of	O
a	O
distributed	O
data	O
structure	O
.	O
</s>
<s>
As	O
a	O
unique	O
feature	O
,	O
Muesli	O
's	O
data	B-Operating_System
parallel	I-Operating_System
skeletons	O
automatically	O
scale	O
both	O
on	O
single	O
-	O
as	O
well	O
as	O
on	O
multi-core	B-Architecture
,	O
multi-node	O
cluster	O
architectures	O
.	O
</s>
<s>
Here	O
,	O
scalability	O
across	O
nodes	O
and	O
cores	O
is	O
ensured	O
by	O
simultaneously	O
using	O
MPI	B-Application
and	O
OpenMP	B-Application
,	O
respectively	O
.	O
</s>
<s>
This	O
is	O
ensured	O
by	O
providing	O
a	O
very	O
thin	O
OpenMP	B-Application
abstraction	O
layer	O
such	O
that	O
the	O
support	O
of	O
multi-core	B-Architecture
architectures	O
can	O
be	O
switched	O
on/off	O
by	O
simply	O
providing/omitting	O
the	O
OpenMP	B-Application
compiler	O
flag	O
when	O
compiling	O
the	O
program	O
.	O
</s>
<s>
Modules	O
can	O
be	O
nested	O
using	O
the	O
two	O
tier	O
model	O
,	O
where	O
the	O
outer	O
level	O
is	O
composed	O
of	O
task	O
parallel	O
skeletons	O
,	O
while	O
data	B-Operating_System
parallel	I-Operating_System
skeletons	O
may	O
be	O
used	O
in	O
the	O
inner	O
level	O
 [ 64 ] 	O
.	O
</s>
<s>
Contrary	O
to	O
Skil	O
,	O
another	O
C	O
like	O
skeleton	O
framework	O
,	O
type	B-Language
safety	I-Language
is	O
not	O
addressed	O
in	O
SKELib	O
.	O
</s>
<s>
PAS	O
(	O
Parallel	O
Architectural	O
Skeletons	O
)	O
is	O
a	O
framework	O
for	O
skeleton	O
programming	O
developed	O
in	O
C++	O
and	O
MPI	B-Application
.	O
</s>
<s>
Elementary	O
skeletons	O
correspond	O
to	O
data	B-Operating_System
parallel	I-Operating_System
skeletons	O
such	O
as	O
map	O
,	O
scan	O
,	O
and	O
fold	O
.	O
</s>
<s>
SkePU	O
SkePU	O
is	O
a	O
skeleton	O
programming	O
framework	O
for	O
multicore	B-Architecture
CPUs	I-Architecture
and	O
multi-GPU	O
systems	O
.	O
</s>
<s>
It	O
is	O
a	O
C++	O
template	O
library	O
with	O
six	O
data-parallel	O
and	O
one	O
task-parallel	O
skeletons	O
,	O
two	O
container	O
types	O
,	O
and	O
support	O
for	O
execution	O
on	O
multi-GPU	O
systems	O
both	O
with	O
CUDA	O
and	O
OpenCL	B-Application
.	O
</s>
<s>
SkePU	O
is	O
being	O
extended	O
for	O
GPU	B-Architecture
clusters	O
.	O
</s>
<s>
SKiPPER	O
is	O
a	O
domain	O
specific	O
skeleton	O
library	O
for	O
vision	O
applications	O
which	O
provides	O
skeletons	O
in	O
CAML	O
,	O
and	O
thus	O
relies	O
on	O
CAML	O
for	O
type	B-Language
safety	I-Language
.	O
</s>
<s>
QUAFF	O
is	O
a	O
more	O
recent	O
skeleton	O
library	O
written	O
in	O
C++	O
and	O
MPI	B-Application
.	O
</s>
<s>
compilation	O
time	O
,	O
new	O
C/MPI	O
code	O
.	O
</s>
<s>
The	O
SkeTo	O
project	O
is	O
a	O
C++	O
library	O
which	O
achieves	O
parallelization	O
using	O
MPI	B-Application
.	O
</s>
<s>
SkeTo	O
is	O
different	O
from	O
other	O
skeleton	O
libraries	O
because	O
instead	O
of	O
providing	O
nestable	O
parallelism	B-Language
patterns	I-Language
,	O
SkeTo	O
provides	O
parallel	O
skeletons	O
for	O
parallel	O
data	O
structures	O
such	O
as	O
:	O
lists	O
,	O
trees	O
,	O
and	O
matrices	O
.	O
</s>
<s>
Skil	O
uses	O
a	O
subset	O
of	O
C	O
language	O
which	O
provides	O
functional	O
language	O
like	O
features	O
such	O
as	O
higher	B-Language
order	I-Language
functions	I-Language
,	O
curring	O
and	O
polymorphic	O
types	O
.	O
</s>
<s>
Data	B-Operating_System
parallelism	I-Operating_System
is	O
achieved	O
using	O
specific	O
data	B-Operating_System
parallel	I-Operating_System
structures	O
,	O
for	O
example	O
to	O
spread	O
arrays	O
among	O
available	O
processors	O
.	O
</s>
<s>
The	O
system	O
relied	O
heavily	O
on	O
functional	O
programming	O
properties	O
,	O
and	O
five	O
skeletons	O
were	O
defined	O
as	O
higher	B-Language
order	I-Language
functions	I-Language
:	O
Divide-and-Conquer	O
,	O
Farm	O
,	O
Map	O
,	O
Pipe	O
and	O
RaMP	O
.	O
</s>
<s>
The	O
big	O
favorite	O
in	O
this	O
sense	O
has	O
been	O
MPI	B-Application
,	O
which	O
is	O
not	O
surprising	O
since	O
it	O
integrates	O
well	O
with	O
the	O
C	O
language	O
,	O
and	O
is	O
probably	O
the	O
most	O
used	O
tool	O
for	O
parallelism	O
in	O
cluster	O
computing	O
.	O
</s>
<s>
For	O
example	O
,	O
CO2	O
P3	O
S	O
can	O
use	O
Threads	O
,	O
RMI	O
or	O
Sockets	O
;	O
Mallba	O
can	O
use	O
Netstream	O
or	O
MPI	B-Application
;	O
or	O
JaSkel	O
which	O
uses	O
AspectJ	O
to	O
execute	O
the	O
skeleton	O
applications	O
on	O
different	O
skeleton	O
frameworks	O
.	O
</s>
<s>
Type	B-Language
safety	I-Language
refers	O
to	O
the	O
capability	O
of	O
detecting	O
type	O
incompatibility	O
errors	O
in	O
skeleton	O
program	O
.	O
</s>
<s>
Since	O
the	O
first	O
skeleton	O
frameworks	O
were	O
built	O
on	O
functional	O
languages	O
such	O
as	O
Haskell	B-Language
,	O
type	B-Language
safety	I-Language
was	O
simply	O
inherited	O
from	O
the	O
host	O
language	O
.	O
</s>
<s>
Recently	O
however	O
,	O
as	O
we	O
began	O
to	O
host	O
skeleton	O
frameworks	O
on	O
object-oriented	O
languages	O
with	O
full	O
nesting	O
,	O
the	O
type	B-Language
safety	I-Language
issue	O
has	O
resurfaced	O
.	O
</s>
<s>
+	O
Non-object	O
oriented	O
algorithmic	B-Language
skeleton	I-Language
frameworks	O
Activity	O
years	O
Programming	O
language	O
Execution	O
language	O
Distribution	O
library	O
Type	B-Language
safe	I-Language
Skeleton	O
nesting	O
File	O
access	O
Skeleton	O
set	O
ASSIST	O
2004	O
–	O
2007	O
Custom	O
control	O
language	O
C++	O
TCP/IP	O
+	O
ssh/scp	O
explicit	O
seq	O
,	O
parmod	O
SBSACO	O
2004	O
–	O
2006	O
Custom	O
composition	O
language	O
C++	O
MPI	B-Application
farm	O
,	O
pipe	O
,	O
multi-block	O
eSkel	O
2004	O
–	O
2005	O
C	O
C	O
MPI	B-Application
?	O
</s>
<s>
pipeline	O
,	O
farm	O
,	O
deal	O
,	O
butterfly	O
,	O
hallowSwap	O
HDC	O
2004	O
–	O
2005	O
Haskell	B-Language
subset	O
C	O
MPI	B-Application
?	O
</s>
<s>
dcA	O
,	O
dcB	O
,	O
dcD	O
,	O
dcE	O
,	O
dcF	O
,	O
map	O
,	O
red	O
,	O
scan	O
,	O
filter	O
SKELib	O
2000-2000	O
C	O
C	O
MPI	B-Application
farm	O
,	O
pipe	O
SkiPPER	O
1999	O
–	O
2002	O
CAML	O
C	O
SynDex	O
limited	O
scm	O
,	O
df	O
,	O
tf	O
,	O
intermem	O
SkIE	O
1999-1999	O
GUI/Custom	O
control	O
language	O
C++	O
MPI	B-Application
limited	O
pipe	O
,	O
farm	O
,	O
map	O
,	O
reduce	O
,	O
loop	O
Eden	O
1997	O
–	O
2011	O
Haskell	B-Language
extension	O
Haskell	B-Language
PVM/MPI	O
map	O
,	O
farm	O
,	O
workpool	O
,	O
nr	O
,	O
dc	O
,	O
pipe	O
,	O
iterUntil	O
,	O
torus	O
,	O
ring	O
P3L	O
1995	O
–	O
1998	O
Custom	O
control	O
language	O
C	O
MPI	B-Application
limited	O
map	O
,	O
reduce	O
,	O
scan	O
,	O
comp	O
,	O
pipe	O
,	O
farm	O
,	O
seq	O
,	O
loop	O
Skil	O
1995	O
–	O
1998	O
C	O
subset	O
C	O
?	O
</s>
<s>
+	O
Object-oriented	O
algorithmic	B-Language
skeleton	I-Language
frameworks	O
Activity	O
years	O
Programming	O
language	O
Execution	O
language	O
Distribution	O
library	O
Type	B-Language
safe	I-Language
Skeleton	O
nesting	O
File	O
access	O
Skeleton	O
set	O
Skandium	O
2009	O
–	O
2012	O
Java	O
Java	O
Threads	O
seq	O
,	O
pipe	O
,	O
farm	O
,	O
for	O
,	O
while	O
,	O
map	O
,	O
d&c	O
,	O
fork	O
FastFlow	O
2009	O
–	O
C++	O
C++11	O
/	O
CUDA	O
/	O
OpenCL	B-Application
C++11	O
threads	O
/	O
Posix	O
threads	O
/	O
TCP-IP	O
/	O
OFED-IB	O
/	O
CUDA	O
/	O
OpenCL	B-Application
Pipeline	O
,	O
Farm	O
,	O
ParallelFor	O
,	O
ParallelForReduce	O
,	O
MapReduce	O
,	O
StencilReduce	O
,	O
PoolEvolution	O
,	O
MacroDataFlow	O
Calcium	O
2006	O
–	O
2008	O
Java	O
Java	O
ProActive	B-Operating_System
seq	O
,	O
pipe	O
,	O
farm	O
,	O
for	O
,	O
while	O
,	O
map	O
,	O
d&c	O
,	O
fork	O
QUAFF	O
2006	O
–	O
2007	O
C++	O
C	O
MPI	B-Application
seq	O
,	O
pipe	O
,	O
farm	O
,	O
scm	O
,	O
pardo	O
JaSkel	O
2006	O
–	O
2007	O
Java	O
Java/AspectJ	O
MPP	O
/	O
RMI	O
farm	O
,	O
pipeline	O
,	O
heartbeat	O
Muskel	O
2005	O
–	O
2008	O
Java	O
Java	O
RMI	O
farm	O
,	O
pipe	O
,	O
seq	O
,	O
+	O
custom	O
MDF	O
Graphs	O
HOC-SA	O
2004	O
–	O
2008	O
Java	O
Java	O
Globus	O
,	O
KOALA	O
farm	O
,	O
pipeline	O
,	O
wavefront	O
SkeTo	O
2003	O
–	O
2013	O
C++	O
C++	O
MPI	B-Application
list	O
,	O
matrix	O
,	O
tree	O
Mallba	O
2002	O
–	O
2007	O
C++	O
C++	O
NetStream	O
/	O
MPI	B-Application
exact	O
,	O
heuristic	O
,	O
hybrid	O
Marrow	O
2013	O
–	O
C++	O
C++	O
plus	O
OpenCL	B-Application
(	O
none	O
)	O
data	B-Operating_System
parallel	I-Operating_System
:	O
map	O
,	O
map-reduce	O
.	O
</s>
<s>
task	O
parallel	O
:	O
pipeline	O
,	O
loop	O
,	O
for	O
Muesli	O
2002	O
–	O
2013	O
C++	O
C++	O
MPI	B-Application
/	O
OpenMP	B-Application
data	B-Operating_System
parallel	I-Operating_System
:	O
fold	O
,	O
map	O
,	O
permute	O
,	O
scan	O
,	O
zip	O
,	O
and	O
variants	O
.	O
</s>
